feat(hlapi): bind string [r]split_once

Bumps [actions/download-artifact](https://github.com/actions/download-artifact) from 4.1.9 to 4.2.1.
- [Release notes](https://github.com/actions/download-artifact/releases)
- [Commits](cc20338598...95815c38cf)

---
updated-dependencies:
- dependency-name: actions/download-artifact
  dependency-type: direct:production
  update-type: version-update:semver-minor
...

Signed-off-by: dependabot[bot] <support@github.com>

.editorconfig

@@ -8,8 +8,14 @@ root = true
 end_of_line = lf
 insert_final_newline = true
 
-# 4 space indentation
-[*.rs]
+# 4 space indentation for rust and toml
+[*.{rs,toml}]
 charset = utf-8
 indent_style = space
 indent_size = 4
+
+# 2 for c and js
+[*.{js,json,c,h}]
+charset = utf-8
+indent_style = space
+indent_size = 2

.github/actionlint.yaml

@@ -5,6 +5,7 @@ self-hosted-runner:
     - 4090-desktop
     - large_windows_16_latest
     - large_ubuntu_16
+    - large_ubuntu_16-22.04
 # Configuration variables in array of strings defined in your repository or
 # organization. `null` means disabling configuration variables check.
 # Empty array means no configuration variable is allowed.

.github/actions/gpu_setup/action.yml

@@ -0,0 +1,63 @@
+name: Setup Cuda
+description: Setup Cuda on Hyperstack or GitHub instance
+
+inputs:
+  cuda-version:
+    description: Version of Cuda to use
+    required: true
+  gcc-version:
+    description: Version of GCC to use
+    required: true
+  cmake-version:
+    description: Version of cmake to use
+    default: 3.29.6
+  github-instance:
+    description: Instance is hosted on GitHub
+    default: 'false'
+
+runs:
+  using: "composite"
+  steps:
+    # Mandatory on hyperstack since a bootable volume is not re-usable yet.
+    - name: Install dependencies
+      shell: bash
+      run: |
+        sudo apt update
+        curl -fsSL https://apt.kitware.com/keys/kitware-archive-latest.asc | sudo gpg --dearmour -o /etc/apt/trusted.gpg.d/kitware.gpg
+        sudo chmod 644 /etc/apt/trusted.gpg.d/kitware.gpg
+        echo 'deb [signed-by=/etc/apt/trusted.gpg.d/kitware.gpg] https://apt.kitware.com/ubuntu/ jammy main' | sudo tee /etc/apt/sources.list.d/kitware.list >/dev/null
+        sudo apt update
+        sudo apt install -y cmake cmake-format libclang-dev
+
+    - name: Install CUDA
+      if: inputs.github-instance == 'true'
+      shell: bash
+      run: |
+        TOOLKIT_VERSION="$(echo ${{ inputs.cuda-version }} | sed 's/\(.*\)\.\(.*\)/\1-\2/')"
+        wget https://developer.download.nvidia.com/compute/cuda/repos/ubuntu2204/x86_64/cuda-keyring_1.1-1_all.deb
+        sudo dpkg -i cuda-keyring_1.1-1_all.deb
+        sudo apt update
+        sudo apt -y install cuda-toolkit-${TOOLKIT_VERSION}
+
+    - name: Export CUDA variables
+      shell: bash
+      run: |
+        CUDA_PATH=/usr/local/cuda-${{ inputs.cuda-version }}
+        echo "CUDA_PATH=$CUDA_PATH" >> "${GITHUB_ENV}"
+        echo "PATH=$PATH:$CUDA_PATH/bin" >> "${GITHUB_PATH}"
+        echo "LD_LIBRARY_PATH=$CUDA_PATH/lib64:$LD_LIBRARY_PATH" >> "${GITHUB_ENV}"
+        echo "CUDA_MODULE_LOADER=EAGER" >> "${GITHUB_ENV}"
+
+    # Specify the correct host compilers
+    - name: Export gcc and g++ variables
+      shell: bash
+      run: |
+        {
+          echo "CC=/usr/bin/gcc-${{ inputs.gcc-version }}";
+          echo "CXX=/usr/bin/g++-${{ inputs.gcc-version }}";
+          echo "CUDAHOSTCXX=/usr/bin/g++-${{ inputs.gcc-version }}";
+        } >> "${GITHUB_ENV}"
+
+    - name: Check device is detected
+      shell: bash
+      run: nvidia-smi

.github/workflows/aws_tfhe_backward_compat_tests.yml

@@ -11,6 +11,12 @@ env:
   SLACK_ICON: https://pbs.twimg.com/profile_images/1274014582265298945/OjBKP9kn_400x400.png
   SLACK_USERNAME: ${{ secrets.BOT_USERNAME }}
   SLACK_WEBHOOK: ${{ secrets.SLACK_WEBHOOK }}
+  SLACKIFY_MARKDOWN: true
+  PULL_REQUEST_MD_LINK: ""
+  CHECKOUT_TOKEN: ${{ secrets.REPO_CHECKOUT_TOKEN || secrets.GITHUB_TOKEN }}
+  # Secrets will be available only to zama-ai organization members
+  SECRETS_AVAILABLE: ${{ secrets.JOB_SECRET != '' }}
+  EXTERNAL_CONTRIBUTION_RUNNER: "large_ubuntu_16"
 
 on:
   # Allows you to run this workflow manually from the Actions tab as an alternative.
@@ -22,11 +28,12 @@ jobs:
     name: Setup instance (backward-compat-tests)
     runs-on: ubuntu-latest
     outputs:
-      runner-name: ${{ steps.start-instance.outputs.label }}
+      runner-name: ${{ steps.start-remote-instance.outputs.label || steps.start-github-instance.outputs.runner_group }}
     steps:
-      - name: Start instance
-        id: start-instance
-        uses: zama-ai/slab-github-runner@801df0b8db5ea2b06128b7476c652f5ed5f193a8
+      - name: Start remote instance
+        id: start-remote-instance
+        if: env.SECRETS_AVAILABLE == 'true'
+        uses: zama-ai/slab-github-runner@79939325c3c429837c10d6041e4fd8589d328bac
         with:
           mode: start
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}
@@ -35,11 +42,18 @@ jobs:
           backend: aws
           profile: cpu-small
 
+      # This instance will be spawned especially for pull-request from forked repository
+      - name: Start GitHub instance
+        id: start-github-instance
+        if: env.SECRETS_AVAILABLE == 'false'
+        run: |
+          echo "runner_group=${{ env.EXTERNAL_CONTRIBUTION_RUNNER }}" >> "$GITHUB_OUTPUT"
+
   backward-compat-tests:
     name: Backward compatibility tests
     needs: [ setup-instance ]
     concurrency:
-      group: ${{ github.workflow }}_${{ github.ref }}
+      group: ${{ github.workflow_ref }}
       cancel-in-progress: true
     runs-on: ${{ needs.setup-instance.outputs.runner-name }}
     steps:
@@ -47,21 +61,17 @@ jobs:
         uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
         with:
           persist-credentials: 'false'
-          token: ${{ secrets.FHE_ACTIONS_TOKEN }}
+          token: ${{ env.CHECKOUT_TOKEN }}
 
       - name: Install latest stable
-        uses: dtolnay/rust-toolchain@7b1c307e0dcbda6122208f10795a713336a9b35a
+        uses: dtolnay/rust-toolchain@888c2e1ea69ab0d4330cbf0af1ecc7b68f368cc1
         with:
           toolchain: stable
 
-      - name: Install git-lfs
-        run: |
-          sudo apt update && sudo apt -y install git-lfs
-
       - name: Use specific data branch
         if: ${{ contains(github.event.pull_request.labels.*.name, 'data_PR') }}
         env:
-          PR_BRANCH: ${{ github.head_ref || github.ref_name }}
+          PR_BRANCH: ${{ github.ref_name }}
         run: |
           echo "BACKWARD_COMPAT_DATA_BRANCH=${PR_BRANCH}" >> "${GITHUB_ENV}"
 
@@ -71,12 +81,28 @@ jobs:
           BRANCH="$(make backward_compat_branch)"
           echo "branch=${BRANCH}" >> "${GITHUB_OUTPUT}"
 
+      - name: Get backward compat branch head SHA
+        id: backward_compat_sha
+        env:
+          REPO_URL: "https://github.com/zama-ai/tfhe-backward-compat-data"
+        run: |
+          SHA=$(git ls-remote ${{ env.REPO_URL }} refs/heads/${{ steps.backward_compat_branch.outputs.branch }} | awk '{print $1}')
+          echo "sha=${SHA}" >> "${GITHUB_OUTPUT}"
+
+      - name: Retrieve data from cache
+        id: retrieve-data-cache
+        uses: actions/cache/restore@5a3ec84eff668545956fd18022155c47e93e2684 #v4.2.3
+        with:
+          path: tests/tfhe-backward-compat-data
+          key: ${{ steps.backward_compat_branch.outputs.branch }}_${{ steps.backward_compat_sha.outputs.sha }}
+
       - name: Clone test data
+        if: steps.retrieve-data-cache.outputs.cache-hit != 'true'
         uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
         with:
           persist-credentials: 'false'
           repository: zama-ai/tfhe-backward-compat-data
-          path: tfhe/tfhe-backward-compat-data
+          path: tests/tfhe-backward-compat-data
           lfs: 'true'
           ref: ${{ steps.backward_compat_branch.outputs.branch }}
 
@@ -84,23 +110,37 @@ jobs:
         run: |
           make test_backward_compatibility_ci
 
+      - name: Store data in cache
+        if: steps.retrieve-data-cache.outputs.cache-hit != 'true'
+        continue-on-error: true
+        uses: actions/cache/save@5a3ec84eff668545956fd18022155c47e93e2684 #v4.2.3
+        with:
+          path: tests/tfhe-backward-compat-data
+          key: ${{ steps.backward_compat_branch.outputs.branch }}_${{ steps.backward_compat_sha.outputs.sha }}
+
+      - name: Set pull-request URL
+        if: ${{ failure() && github.event_name == 'pull_request' }}
+        run: |
+          echo "PULL_REQUEST_MD_LINK=[pull-request](${{ vars.PR_BASE_URL }}${{ github.event.pull_request.number }}), "  >> "${GITHUB_ENV}"
+
       - name: Slack Notification
         if: ${{ failure() }}
         continue-on-error: true
         uses: rtCamp/action-slack-notify@c33737706dea87cd7784c687dadc9adf1be59990
         env:
           SLACK_COLOR: ${{ job.status }}
-          SLACK_MESSAGE: "Backward compatibility tests finished with status: ${{ job.status }}. (${{ env.ACTION_RUN_URL }})"
+          SLACK_MESSAGE: "Backward compatibility tests finished with status: ${{ job.status }}. (${{ env.PULL_REQUEST_MD_LINK }}[action run](${{ env.ACTION_RUN_URL }}))"
 
   teardown-instance:
     name: Teardown instance (backward-compat-tests)
-    if: ${{ always() && needs.setup-instance.result != 'skipped' }}
+    if: ${{ always() && needs.setup-instance.result == 'success' }}
     needs: [ setup-instance, backward-compat-tests ]
     runs-on: ubuntu-latest
     steps:
-      - name: Stop instance
+      - name: Stop remote instance
         id: stop-instance
-        uses: zama-ai/slab-github-runner@801df0b8db5ea2b06128b7476c652f5ed5f193a8
+        if: env.SECRETS_AVAILABLE == 'true'
+        uses: zama-ai/slab-github-runner@79939325c3c429837c10d6041e4fd8589d328bac
         with:
           mode: stop
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}

.github/workflows/aws_tfhe_fast_tests.yml

@@ -11,7 +11,13 @@ env:
   SLACK_ICON: https://pbs.twimg.com/profile_images/1274014582265298945/OjBKP9kn_400x400.png
   SLACK_USERNAME: ${{ secrets.BOT_USERNAME }}
   SLACK_WEBHOOK: ${{ secrets.SLACK_WEBHOOK }}
+  SLACKIFY_MARKDOWN: true
   IS_PULL_REQUEST: ${{ github.event_name == 'pull_request' }}
+  PULL_REQUEST_MD_LINK: ""
+  CHECKOUT_TOKEN: ${{ secrets.REPO_CHECKOUT_TOKEN || secrets.GITHUB_TOKEN }}
+  # Secrets will be available only to zama-ai organization members
+  SECRETS_AVAILABLE: ${{ secrets.JOB_SECRET != '' }}
+  EXTERNAL_CONTRIBUTION_RUNNER: "large_ubuntu_64-22.04"
 
 on:
   # Allows you to run this workflow manually from the Actions tab as an alternative.
@@ -22,7 +28,7 @@ jobs:
   should-run:
     runs-on: ubuntu-latest
     permissions:
-      pull-requests: write
+      pull-requests: read
     outputs:
       csprng_test: ${{ env.IS_PULL_REQUEST == 'false' || steps.changed-files.outputs.csprng_any_changed }}
       zk_pok_test: ${{ env.IS_PULL_REQUEST == 'false' || steps.changed-files.outputs.zk_pok_any_changed }}
@@ -54,12 +60,13 @@ jobs:
         uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
         with:
           fetch-depth: 0
+          persist-credentials: 'false'
+          token: ${{ env.CHECKOUT_TOKEN }}
 
       - name: Check for file changes
         id: changed-files
-        uses: tj-actions/changed-files@4edd678ac3f81e2dc578756871e4d00c19191daf
+        uses: tj-actions/changed-files@26a38635fc1173cc5820336ce97be6188d0de9f5 # v46.0.2
         with:
-          since_last_remote_commit: true
           files_yaml: |
             dependencies:
               - tfhe/Cargo.toml
@@ -102,7 +109,7 @@ jobs:
             user_docs:
               - tfhe/src/**
               - '!tfhe/src/c_api/**'
-              - 'tfhe/docs/**.md'
+              - 'tfhe/docs/**/**.md'
               - README.md
 
       - name: Aggregate file changes
@@ -123,16 +130,17 @@ jobs:
 
   setup-instance:
     name: Setup instance (fast-tests)
-    if: github.event_name != 'pull_request' ||
-      needs.should-run.outputs.any_file_changed == 'true'
+    if: github.event_name == 'workflow_dispatch' ||
+      (github.event_name != 'workflow_dispatch' && needs.should-run.outputs.any_file_changed == 'true')
     needs: should-run
     runs-on: ubuntu-latest
     outputs:
-      runner-name: ${{ steps.start-instance.outputs.label }}
+      runner-name: ${{ steps.start-remote-instance.outputs.label || steps.start-github-instance.outputs.runner_group }}
     steps:
-      - name: Start instance
-        id: start-instance
-        uses: zama-ai/slab-github-runner@801df0b8db5ea2b06128b7476c652f5ed5f193a8
+      - name: Start remote instance
+        id: start-remote-instance
+        if: env.SECRETS_AVAILABLE == 'true'
+        uses: zama-ai/slab-github-runner@79939325c3c429837c10d6041e4fd8589d328bac
         with:
           mode: start
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}
@@ -141,13 +149,18 @@ jobs:
           backend: aws
           profile: cpu-big
 
+      # This instance will be spawned especially for pull-request from forked repository
+      - name: Start GitHub instance
+        id: start-github-instance
+        if: env.SECRETS_AVAILABLE == 'false'
+        run: |
+          echo "runner_group=${{ env.EXTERNAL_CONTRIBUTION_RUNNER }}" >> "$GITHUB_OUTPUT"
+
   fast-tests:
     name: Fast CPU tests
-    if: github.event_name != 'pull_request' ||
-      (github.event_name == 'pull_request' && needs.setup-instance.result != 'skipped')
     needs: [ should-run, setup-instance ]
     concurrency:
-      group: ${{ github.workflow }}_${{ github.ref }}
+      group: ${{ github.workflow_ref }}
       cancel-in-progress: true
     runs-on: ${{ needs.setup-instance.outputs.runner-name }}
     steps:
@@ -155,10 +168,10 @@ jobs:
         uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
         with:
           persist-credentials: 'false'
-          token: ${{ secrets.FHE_ACTIONS_TOKEN }}
+          token: ${{ env.CHECKOUT_TOKEN }}
 
       - name: Install latest stable
-        uses: dtolnay/rust-toolchain@7b1c307e0dcbda6122208f10795a713336a9b35a
+        uses: dtolnay/rust-toolchain@888c2e1ea69ab0d4330cbf0af1ecc7b68f368cc1
         with:
           toolchain: stable
 
@@ -198,7 +211,7 @@ jobs:
 
       - name: Node cache restoration
         id: node-cache
-        uses: actions/cache/restore@6849a6489940f00c2f30c0fb92c6274307ccb58a #v4.1.2
+        uses: actions/cache/restore@5a3ec84eff668545956fd18022155c47e93e2684 #v4.2.3
         with:
           path: |
             ~/.nvm
@@ -211,7 +224,7 @@ jobs:
           make install_node
 
       - name: Node cache save
-        uses: actions/cache/save@6849a6489940f00c2f30c0fb92c6274307ccb58a #v4.1.2
+        uses: actions/cache/save@5a3ec84eff668545956fd18022155c47e93e2684 #v4.2.3
         if: steps.node-cache.outputs.cache-hit != 'true'
         with:
           path: |
@@ -253,23 +266,29 @@ jobs:
         run: |
           make test_zk
 
+      - name: Set pull-request URL
+        if: ${{ failure() && github.event_name == 'pull_request' }}
+        run: |
+          echo "PULL_REQUEST_MD_LINK=[pull-request](${{ vars.PR_BASE_URL }}${{ github.event.pull_request.number }}), "  >> "${GITHUB_ENV}"
+
       - name: Slack Notification
-        if: ${{ failure() }}
+        if: ${{ failure() && env.SECRETS_AVAILABLE == 'true' }}
         continue-on-error: true
         uses: rtCamp/action-slack-notify@c33737706dea87cd7784c687dadc9adf1be59990
         env:
           SLACK_COLOR: ${{ job.status }}
-          SLACK_MESSAGE: "Fast AWS tests finished with status: ${{ job.status }}. (${{ env.ACTION_RUN_URL }})"
+          SLACK_MESSAGE: "Fast AWS tests finished with status: ${{ job.status }}. (${{ env.PULL_REQUEST_MD_LINK }}[action run](${{ env.ACTION_RUN_URL }}))"
 
   teardown-instance:
     name: Teardown instance (fast-tests)
-    if: ${{ always() && needs.setup-instance.result != 'skipped' }}
+    if: ${{ always() && needs.setup-instance.result == 'success' }}
     needs: [ setup-instance, fast-tests ]
     runs-on: ubuntu-latest
     steps:
-      - name: Stop instance
+      - name: Stop remote instance
         id: stop-instance
-        uses: zama-ai/slab-github-runner@801df0b8db5ea2b06128b7476c652f5ed5f193a8
+        if: env.SECRETS_AVAILABLE == 'true'
+        uses: zama-ai/slab-github-runner@79939325c3c429837c10d6041e4fd8589d328bac
         with:
           mode: stop
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}

.github/workflows/aws_tfhe_integer_tests.yml

@@ -10,16 +10,22 @@ env:
   SLACK_ICON: https://pbs.twimg.com/profile_images/1274014582265298945/OjBKP9kn_400x400.png
   SLACK_USERNAME: ${{ secrets.BOT_USERNAME }}
   SLACK_WEBHOOK: ${{ secrets.SLACK_WEBHOOK }}
+  SLACKIFY_MARKDOWN: true
+  PULL_REQUEST_MD_LINK: ""
   # We clear the cache to reduce memory pressure because of the numerous processes of cargo
   # nextest
   TFHE_RS_CLEAR_IN_MEMORY_KEY_CACHE: "1"
   NO_BIG_PARAMS: FALSE
+  CHECKOUT_TOKEN: ${{ secrets.REPO_CHECKOUT_TOKEN || secrets.GITHUB_TOKEN }}
+  # Secrets will be available only to zama-ai organization members
+  SECRETS_AVAILABLE: ${{ secrets.JOB_SECRET != '' }}
+  EXTERNAL_CONTRIBUTION_RUNNER: "large_ubuntu_64-22.04"
 
 on:
   # Allows you to run this workflow manually from the Actions tab as an alternative.
   workflow_dispatch:
   pull_request:
-    types: [labeled]
+    types: [ labeled ]
   push:
     branches:
       - main
@@ -28,12 +34,11 @@ jobs:
   should-run:
     if:
       (github.event_name == 'push' && github.repository == 'zama-ai/tfhe-rs') ||
-      (github.event_name == 'schedule' && github.repository == 'zama-ai/tfhe-rs') ||
       (github.event_name == 'pull_request' && contains(github.event.label.name, 'approved')) ||
       github.event_name == 'workflow_dispatch'
     runs-on: ubuntu-latest
     permissions:
-      pull-requests: write
+      pull-requests: read
     outputs:
       integer_test: ${{ github.event_name == 'workflow_dispatch' ||
         steps.changed-files.outputs.integer_any_changed }}
@@ -42,13 +47,13 @@ jobs:
         uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
         with:
           fetch-depth: 0
-          persist-credentials: "false"
+          persist-credentials: 'false'
+          token: ${{ env.CHECKOUT_TOKEN }}
 
       - name: Check for file changes
         id: changed-files
-        uses: tj-actions/changed-files@4edd678ac3f81e2dc578756871e4d00c19191daf
+        uses: tj-actions/changed-files@26a38635fc1173cc5820336ce97be6188d0de9f5 # v46.0.2
         with:
-          since_last_remote_commit: true
           files_yaml: |
             integer:
               - tfhe/Cargo.toml
@@ -58,6 +63,7 @@ jobs:
               - tfhe/src/core_crypto/**
               - tfhe/src/shortint/**
               - tfhe/src/integer/**
+              - .github/workflows/aws_tfhe_integer_tests.yml
 
   setup-instance:
     name: Setup instance (unsigned-integer-tests)
@@ -65,15 +71,16 @@ jobs:
     if:
       (github.event_name == 'push' && github.repository == 'zama-ai/tfhe-rs' && needs.should-run.outputs.integer_test == 'true') ||
       (github.event_name == 'schedule' && github.repository == 'zama-ai/tfhe-rs') ||
-      (github.event_name == 'pull_request' && contains(github.event.label.name, 'approved')) ||
+      (github.event.action == 'labeled' && github.event.label.name == 'approved' && needs.should-run.outputs.integer_test == 'true') ||
       github.event_name == 'workflow_dispatch'
     runs-on: ubuntu-latest
     outputs:
-      runner-name: ${{ steps.start-instance.outputs.label }}
+      runner-name: ${{ steps.start-remote-instance.outputs.label || steps.start-github-instance.outputs.runner_group }}
     steps:
-      - name: Start instance
-        id: start-instance
-        uses: zama-ai/slab-github-runner@801df0b8db5ea2b06128b7476c652f5ed5f193a8
+      - name: Start remote instance
+        id: start-remote-instance
+        if: env.SECRETS_AVAILABLE == 'true'
+        uses: zama-ai/slab-github-runner@79939325c3c429837c10d6041e4fd8589d328bac
         with:
           mode: start
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}
@@ -82,11 +89,18 @@ jobs:
           backend: aws
           profile: cpu-big
 
+      # This instance will be spawned especially for pull-request from forked repository
+      - name: Start GitHub instance
+        id: start-github-instance
+        if: env.SECRETS_AVAILABLE == 'false'
+        run: |
+          echo "runner_group=${{ env.EXTERNAL_CONTRIBUTION_RUNNER }}" >> "$GITHUB_OUTPUT"
+
   unsigned-integer-tests:
     name: Unsigned integer tests
     needs: setup-instance
     concurrency:
-      group: ${{ github.workflow }}_${{ github.ref }}${{ github.ref == 'refs/heads/main' && github.sha || '' }}
+      group: ${{ github.workflow_ref }}
       cancel-in-progress: ${{ github.ref != 'refs/heads/main' }}
     runs-on: ${{ needs.setup-instance.outputs.runner-name }}
     steps:
@@ -94,10 +108,10 @@ jobs:
         uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
         with:
           persist-credentials: "false"
-          token: ${{ secrets.FHE_ACTIONS_TOKEN }}
+          token: ${{ env.CHECKOUT_TOKEN }}
 
       - name: Install latest stable
-        uses: dtolnay/rust-toolchain@7b1c307e0dcbda6122208f10795a713336a9b35a
+        uses: dtolnay/rust-toolchain@888c2e1ea69ab0d4330cbf0af1ecc7b68f368cc1
         with:
           toolchain: stable
 
@@ -122,23 +136,29 @@ jobs:
         run: |
           AVX512_SUPPORT=ON NO_BIG_PARAMS=${{ env.NO_BIG_PARAMS }} BIG_TESTS_INSTANCE=TRUE make test_unsigned_integer_ci
 
+      - name: Set pull-request URL
+        if: ${{ failure() && github.event_name == 'pull_request' }}
+        run: |
+          echo "PULL_REQUEST_MD_LINK=[pull-request](${{ vars.PR_BASE_URL }}${{ github.event.pull_request.number }}), "  >> "${GITHUB_ENV}"
+
       - name: Slack Notification
         if: ${{ failure() }}
         continue-on-error: true
         uses: rtCamp/action-slack-notify@c33737706dea87cd7784c687dadc9adf1be59990
         env:
           SLACK_COLOR: ${{ job.status }}
-          SLACK_MESSAGE: "Unsigned Integer tests finished with status: ${{ job.status }}. (${{ env.ACTION_RUN_URL }})"
+          SLACK_MESSAGE: "Unsigned Integer tests finished with status: ${{ job.status }}. (${{ env.PULL_REQUEST_MD_LINK }}[action run](${{ env.ACTION_RUN_URL }}))"
 
   teardown-instance:
     name: Teardown instance (unsigned-integer-tests)
-    if: ${{ always() && needs.setup-instance.result != 'skipped' }}
+    if: ${{ always() && needs.setup-instance.result == 'success' }}
     needs: [setup-instance, unsigned-integer-tests]
     runs-on: ubuntu-latest
     steps:
-      - name: Stop instance
+      - name: Stop remote instance
         id: stop-instance
-        uses: zama-ai/slab-github-runner@801df0b8db5ea2b06128b7476c652f5ed5f193a8
+        if: env.SECRETS_AVAILABLE == 'true'
+        uses: zama-ai/slab-github-runner@79939325c3c429837c10d6041e4fd8589d328bac
         with:
           mode: stop
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}

.github/workflows/aws_tfhe_signed_integer_tests.yml

@@ -10,16 +10,22 @@ env:
   SLACK_ICON: https://pbs.twimg.com/profile_images/1274014582265298945/OjBKP9kn_400x400.png
   SLACK_USERNAME: ${{ secrets.BOT_USERNAME }}
   SLACK_WEBHOOK: ${{ secrets.SLACK_WEBHOOK }}
+  SLACKIFY_MARKDOWN: true
+  PULL_REQUEST_MD_LINK: ""
   # We clear the cache to reduce memory pressure because of the numerous processes of cargo
   # nextest
   TFHE_RS_CLEAR_IN_MEMORY_KEY_CACHE: "1"
   NO_BIG_PARAMS: FALSE
+  CHECKOUT_TOKEN: ${{ secrets.REPO_CHECKOUT_TOKEN || secrets.GITHUB_TOKEN }}
+  # Secrets will be available only to zama-ai organization members
+  SECRETS_AVAILABLE: ${{ secrets.JOB_SECRET != '' }}
+  EXTERNAL_CONTRIBUTION_RUNNER: "large_ubuntu_64-22.04"
 
 on:
   # Allows you to run this workflow manually from the Actions tab as an alternative.
   workflow_dispatch:
   pull_request:
-    types: [labeled]
+    types: [ labeled ]
   push:
     branches:
       - main
@@ -33,7 +39,7 @@ jobs:
       github.event_name == 'workflow_dispatch'
     runs-on: ubuntu-latest
     permissions:
-      pull-requests: write
+      pull-requests: read
     outputs:
       integer_test: ${{ github.event_name == 'workflow_dispatch' ||
         steps.changed-files.outputs.integer_any_changed }}
@@ -42,13 +48,13 @@ jobs:
         uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
         with:
           fetch-depth: 0
-          persist-credentials: "false"
+          persist-credentials: 'false'
+          token: ${{ env.CHECKOUT_TOKEN }}
 
       - name: Check for file changes
         id: changed-files
-        uses: tj-actions/changed-files@4edd678ac3f81e2dc578756871e4d00c19191daf
+        uses: tj-actions/changed-files@26a38635fc1173cc5820336ce97be6188d0de9f5 # v46.0.2
         with:
-          since_last_remote_commit: true
           files_yaml: |
             integer:
               - tfhe/Cargo.toml
@@ -58,6 +64,7 @@ jobs:
               - tfhe/src/core_crypto/**
               - tfhe/src/shortint/**
               - tfhe/src/integer/**
+              - .github/workflows/aws_tfhe_signed_integer_tests.yml
 
   setup-instance:
     name: Setup instance (unsigned-integer-tests)
@@ -65,15 +72,16 @@ jobs:
     if:
       (github.event_name == 'push' && github.repository == 'zama-ai/tfhe-rs' && needs.should-run.outputs.integer_test == 'true') ||
       (github.event_name == 'schedule' && github.repository == 'zama-ai/tfhe-rs') ||
-      (github.event_name == 'pull_request' && contains(github.event.label.name, 'approved')) ||
+      (github.event.action == 'labeled' && github.event.label.name == 'approved' && needs.should-run.outputs.integer_test == 'true') ||
       github.event_name == 'workflow_dispatch'
     runs-on: ubuntu-latest
     outputs:
-      runner-name: ${{ steps.start-instance.outputs.label }}
+      runner-name: ${{ steps.start-remote-instance.outputs.label || steps.start-github-instance.outputs.runner_group }}
     steps:
-      - name: Start instance
-        id: start-instance
-        uses: zama-ai/slab-github-runner@801df0b8db5ea2b06128b7476c652f5ed5f193a8
+      - name: Start remote instance
+        id: start-remote-instance
+        if: env.SECRETS_AVAILABLE == 'true'
+        uses: zama-ai/slab-github-runner@79939325c3c429837c10d6041e4fd8589d328bac
         with:
           mode: start
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}
@@ -82,11 +90,18 @@ jobs:
           backend: aws
           profile: cpu-big
 
+      # This instance will be spawned especially for pull-request from forked repository
+      - name: Start GitHub instance
+        id: start-github-instance
+        if: env.SECRETS_AVAILABLE == 'false'
+        run: |
+          echo "runner_group=${{ env.EXTERNAL_CONTRIBUTION_RUNNER }}" >> "$GITHUB_OUTPUT"
+
   signed-integer-tests:
     name: Signed integer tests
     needs: setup-instance
     concurrency:
-      group: ${{ github.workflow }}_${{ github.ref }}${{ github.ref == 'refs/heads/main' && github.sha || '' }}
+      group: ${{ github.workflow_ref }}
       cancel-in-progress: ${{ github.ref != 'refs/heads/main' }}
     runs-on: ${{ needs.setup-instance.outputs.runner-name }}
     steps:
@@ -94,10 +109,10 @@ jobs:
         uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
         with:
           persist-credentials: "false"
-          token: ${{ secrets.FHE_ACTIONS_TOKEN }}
+          token: ${{ env.CHECKOUT_TOKEN }}
 
       - name: Install latest stable
-        uses: dtolnay/rust-toolchain@7b1c307e0dcbda6122208f10795a713336a9b35a
+        uses: dtolnay/rust-toolchain@888c2e1ea69ab0d4330cbf0af1ecc7b68f368cc1
         with:
           toolchain: stable
 
@@ -126,23 +141,29 @@ jobs:
         run: |
           AVX512_SUPPORT=ON NO_BIG_PARAMS=${{ env.NO_BIG_PARAMS }} BIG_TESTS_INSTANCE=TRUE make test_signed_integer_ci
 
+      - name: Set pull-request URL
+        if: ${{ failure() && github.event_name == 'pull_request' }}
+        run: |
+          echo "PULL_REQUEST_MD_LINK=[pull-request](${{ vars.PR_BASE_URL }}${{ github.event.pull_request.number }}), "  >> "${GITHUB_ENV}"
+
       - name: Slack Notification
         if: ${{ failure() }}
         continue-on-error: true
         uses: rtCamp/action-slack-notify@c33737706dea87cd7784c687dadc9adf1be59990
         env:
           SLACK_COLOR: ${{ job.status }}
-          SLACK_MESSAGE: "Signed Integer tests finished with status: ${{ job.status }}. (${{ env.ACTION_RUN_URL }})"
+          SLACK_MESSAGE: "Signed Integer tests finished with status: ${{ job.status }}. (${{ env.PULL_REQUEST_MD_LINK }}[action run](${{ env.ACTION_RUN_URL }}))"
 
   teardown-instance:
     name: Teardown instance (signed-integer-tests)
-    if: ${{ always() && needs.setup-instance.result != 'skipped' }}
+    if: ${{ always() && needs.setup-instance.result == 'success' }}
     needs: [setup-instance, signed-integer-tests]
     runs-on: ubuntu-latest
     steps:
-      - name: Stop instance
+      - name: Stop remote instance
         id: stop-instance
-        uses: zama-ai/slab-github-runner@801df0b8db5ea2b06128b7476c652f5ed5f193a8
+        if: env.SECRETS_AVAILABLE == 'true'
+        uses: zama-ai/slab-github-runner@79939325c3c429837c10d6041e4fd8589d328bac
         with:
           mode: stop
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}

.github/workflows/aws_tfhe_tests.yml

@@ -10,7 +10,13 @@ env:
   SLACK_ICON: https://pbs.twimg.com/profile_images/1274014582265298945/OjBKP9kn_400x400.png
   SLACK_USERNAME: ${{ secrets.BOT_USERNAME }}
   SLACK_WEBHOOK: ${{ secrets.SLACK_WEBHOOK }}
+  SLACKIFY_MARKDOWN: true
   IS_PULL_REQUEST: ${{ github.event_name == 'pull_request' }}
+  PULL_REQUEST_MD_LINK: ""
+  CHECKOUT_TOKEN: ${{ secrets.REPO_CHECKOUT_TOKEN || secrets.GITHUB_TOKEN }}
+  # Secrets will be available only to zama-ai organization members
+  SECRETS_AVAILABLE: ${{ secrets.JOB_SECRET != '' }}
+  EXTERNAL_CONTRIBUTION_RUNNER: "large_ubuntu_64-22.04"
 
 on:
   # Allows you to run this workflow manually from the Actions tab as an alternative.
@@ -27,7 +33,7 @@ jobs:
     if: github.event_name != 'schedule' ||
       (github.event_name == 'schedule' && github.repository == 'zama-ai/tfhe-rs')
     permissions:
-      pull-requests: write
+      pull-requests: read
     outputs:
       csprng_test: ${{ env.IS_PULL_REQUEST == 'false' || steps.changed-files.outputs.csprng_any_changed }}
       zk_pok_test: ${{ env.IS_PULL_REQUEST == 'false' || steps.changed-files.outputs.zk_pok_any_changed }}
@@ -63,12 +69,13 @@ jobs:
         uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
         with:
           fetch-depth: 0
+          persist-credentials: 'false'
+          token: ${{ env.CHECKOUT_TOKEN }}
 
       - name: Check for file changes
         id: changed-files
-        uses: tj-actions/changed-files@4edd678ac3f81e2dc578756871e4d00c19191daf
+        uses: tj-actions/changed-files@26a38635fc1173cc5820336ce97be6188d0de9f5 # v46.0.2
         with:
-          since_last_remote_commit: true
           files_yaml: |
             dependencies:
               - tfhe/Cargo.toml
@@ -110,7 +117,7 @@ jobs:
             user_docs:
               - tfhe/src/**
               - '!tfhe/src/c_api/**'
-              - 'tfhe/docs/**.md'
+              - 'tfhe/docs/**/**.md'
               - README.md
 
       - name: Aggregate file changes
@@ -137,11 +144,12 @@ jobs:
     needs: should-run
     runs-on: ubuntu-latest
     outputs:
-      runner-name: ${{ steps.start-instance.outputs.label }}
+      runner-name: ${{ steps.start-remote-instance.outputs.label || steps.start-github-instance.outputs.runner_group }}
     steps:
-      - name: Start instance
-        id: start-instance
-        uses: zama-ai/slab-github-runner@801df0b8db5ea2b06128b7476c652f5ed5f193a8
+      - name: Start remote instance
+        id: start-remote-instance
+        if: env.SECRETS_AVAILABLE == 'true'
+        uses: zama-ai/slab-github-runner@79939325c3c429837c10d6041e4fd8589d328bac
         with:
           mode: start
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}
@@ -150,13 +158,20 @@ jobs:
           backend: aws
           profile: cpu-big
 
+      # This instance will be spawned especially for pull-request from forked repository
+      - name: Start GitHub instance
+        id: start-github-instance
+        if: env.SECRETS_AVAILABLE == 'false'
+        run: |
+          echo "runner_group=${{ env.EXTERNAL_CONTRIBUTION_RUNNER }}" >> "$GITHUB_OUTPUT"
+
   cpu-tests:
     name: CPU tests
     if: github.event_name != 'pull_request' ||
       (github.event_name == 'pull_request' && needs.setup-instance.result != 'skipped')
     needs: [ should-run, setup-instance ]
     concurrency:
-      group: ${{ github.workflow }}_${{github.event_name}}_${{ github.ref }}
+      group: ${{ github.workflow_ref }}_${{github.event_name}}
       cancel-in-progress: true
     runs-on: ${{ needs.setup-instance.outputs.runner-name }}
     steps:
@@ -164,10 +179,10 @@ jobs:
         uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
         with:
           persist-credentials: 'false'
-          token: ${{ secrets.FHE_ACTIONS_TOKEN }}
+          token: ${{ env.CHECKOUT_TOKEN }}
 
       - name: Install latest stable
-        uses: dtolnay/rust-toolchain@7b1c307e0dcbda6122208f10795a713336a9b35a
+        uses: dtolnay/rust-toolchain@888c2e1ea69ab0d4330cbf0af1ecc7b68f368cc1
         with:
           toolchain: stable
 
@@ -233,23 +248,29 @@ jobs:
           make test_trivium
           make test_kreyvium
 
+      - name: Set pull-request URL
+        if: ${{ failure() && github.event_name == 'pull_request' }}
+        run: |
+          echo "PULL_REQUEST_MD_LINK=[pull-request](${{ vars.PR_BASE_URL }}${{ github.event.pull_request.number }}), "  >> "${GITHUB_ENV}"
+
       - name: Slack Notification
         if: ${{ failure() }}
         continue-on-error: true
         uses: rtCamp/action-slack-notify@c33737706dea87cd7784c687dadc9adf1be59990
         env:
           SLACK_COLOR: ${{ job.status }}
-          SLACK_MESSAGE: "CPU tests finished with status: ${{ job.status }}. (${{ env.ACTION_RUN_URL }})"
+          SLACK_MESSAGE: "CPU tests finished with status: ${{ job.status }}. (${{ env.PULL_REQUEST_MD_LINK }}[action run](${{ env.ACTION_RUN_URL }}))"
 
   teardown-instance:
     name: Teardown instance (cpu-tests)
-    if: ${{ always() && needs.setup-instance.result != 'skipped' }}
+    if: ${{ always() && needs.setup-instance.result == 'success' }}
     needs: [ setup-instance, cpu-tests ]
     runs-on: ubuntu-latest
     steps:
-      - name: Stop instance
+      - name: Stop remote instance
         id: stop-instance
-        uses: zama-ai/slab-github-runner@801df0b8db5ea2b06128b7476c652f5ed5f193a8
+        if: env.SECRETS_AVAILABLE == 'true'
+        uses: zama-ai/slab-github-runner@79939325c3c429837c10d6041e4fd8589d328bac
         with:
           mode: stop
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}

.github/workflows/aws_tfhe_wasm_tests.yml

@@ -10,6 +10,12 @@ env:
   SLACK_ICON: https://pbs.twimg.com/profile_images/1274014582265298945/OjBKP9kn_400x400.png
   SLACK_USERNAME: ${{ secrets.BOT_USERNAME }}
   SLACK_WEBHOOK: ${{ secrets.SLACK_WEBHOOK }}
+  SLACKIFY_MARKDOWN: true
+  PULL_REQUEST_MD_LINK: ""
+  CHECKOUT_TOKEN: ${{ secrets.REPO_CHECKOUT_TOKEN || secrets.GITHUB_TOKEN }}
+  # Secrets will be available only to zama-ai organization members
+  SECRETS_AVAILABLE: ${{ secrets.JOB_SECRET != '' }}
+  EXTERNAL_CONTRIBUTION_RUNNER: "large_ubuntu_16"
 
 on:
   # Allows you to run this workflow manually from the Actions tab as an alternative.
@@ -23,11 +29,12 @@ jobs:
     if: ${{ github.event_name == 'workflow_dispatch' || contains(github.event.label.name, 'approved') }}
     runs-on: ubuntu-latest
     outputs:
-      runner-name: ${{ steps.start-instance.outputs.label }}
+      runner-name: ${{ steps.start-remote-instance.outputs.label || steps.start-github-instance.outputs.runner_group }}
     steps:
-      - name: Start instance
-        id: start-instance
-        uses: zama-ai/slab-github-runner@801df0b8db5ea2b06128b7476c652f5ed5f193a8
+      - name: Start remote instance
+        id: start-remote-instance
+        if: env.SECRETS_AVAILABLE == 'true'
+        uses: zama-ai/slab-github-runner@79939325c3c429837c10d6041e4fd8589d328bac
         with:
           mode: start
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}
@@ -36,11 +43,18 @@ jobs:
           backend: aws
           profile: cpu-small
 
+      # This instance will be spawned especially for pull-request from forked repository
+      - name: Start GitHub instance
+        id: start-github-instance
+        if: env.SECRETS_AVAILABLE == 'false'
+        run: |
+          echo "runner_group=${{ env.EXTERNAL_CONTRIBUTION_RUNNER }}" >> "$GITHUB_OUTPUT"
+
   wasm-tests:
     name: WASM tests
     needs: setup-instance
     concurrency:
-      group: ${{ github.workflow }}_${{ github.ref }}
+      group: ${{ github.workflow_ref }}
       cancel-in-progress: true
     runs-on: ${{ needs.setup-instance.outputs.runner-name }}
     steps:
@@ -48,10 +62,10 @@ jobs:
         uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
         with:
           persist-credentials: 'false'
-          token: ${{ secrets.FHE_ACTIONS_TOKEN }}
+          token: ${{ env.CHECKOUT_TOKEN }}
 
       - name: Install latest stable
-        uses: dtolnay/rust-toolchain@7b1c307e0dcbda6122208f10795a713336a9b35a
+        uses: dtolnay/rust-toolchain@888c2e1ea69ab0d4330cbf0af1ecc7b68f368cc1
         with:
           toolchain: stable
 
@@ -61,7 +75,7 @@ jobs:
 
       - name: Node cache restoration
         id: node-cache
-        uses: actions/cache/restore@6849a6489940f00c2f30c0fb92c6274307ccb58a #v4.1.2
+        uses: actions/cache/restore@5a3ec84eff668545956fd18022155c47e93e2684 #v4.2.3
         with:
           path: |
             ~/.nvm
@@ -74,7 +88,7 @@ jobs:
           make install_node
 
       - name: Node cache save
-        uses: actions/cache/save@6849a6489940f00c2f30c0fb92c6274307ccb58a #v4.1.2
+        uses: actions/cache/save@5a3ec84eff668545956fd18022155c47e93e2684 #v4.2.3
         if: steps.node-cache.outputs.cache-hit != 'true'
         with:
           path: |
@@ -103,23 +117,29 @@ jobs:
         run: |
           make test_zk_wasm_x86_compat_ci
 
+      - name: Set pull-request URL
+        if: ${{ failure() && github.event_name == 'pull_request' }}
+        run: |
+          echo "PULL_REQUEST_MD_LINK=[pull-request](${{ vars.PR_BASE_URL }}${{ github.event.pull_request.number }}), "  >> "${GITHUB_ENV}"
+
       - name: Slack Notification
         if: ${{ failure() }}
         continue-on-error: true
         uses: rtCamp/action-slack-notify@c33737706dea87cd7784c687dadc9adf1be59990
         env:
           SLACK_COLOR: ${{ job.status }}
-          SLACK_MESSAGE: "WASM tests finished with status: ${{ job.status }}. (${{ env.ACTION_RUN_URL }})"
+          SLACK_MESSAGE: "WASM tests finished with status: ${{ job.status }}. (${{ env.PULL_REQUEST_MD_LINK }}[action run](${{ env.ACTION_RUN_URL }}))"
 
   teardown-instance:
     name: Teardown instance (wasm-tests)
-    if: ${{ always() && needs.setup-instance.result != 'skipped' }}
+    if: ${{ always() && needs.setup-instance.result == 'success' }}
     needs: [ setup-instance, wasm-tests ]
     runs-on: ubuntu-latest
     steps:
-      - name: Stop instance
+      - name: Stop remote instance
         id: stop-instance
-        uses: zama-ai/slab-github-runner@801df0b8db5ea2b06128b7476c652f5ed5f193a8
+        if: env.SECRETS_AVAILABLE == 'true'
+        uses: zama-ai/slab-github-runner@79939325c3c429837c10d6041e4fd8589d328bac
         with:
           mode: stop
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}

.github/workflows/benchmark_boolean.yml

@@ -29,7 +29,7 @@ jobs:
     steps:
       - name: Start instance
         id: start-instance
-        uses: zama-ai/slab-github-runner@801df0b8db5ea2b06128b7476c652f5ed5f193a8
+        uses: zama-ai/slab-github-runner@79939325c3c429837c10d6041e4fd8589d328bac
         with:
           mode: start
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}
@@ -43,7 +43,7 @@ jobs:
     needs: setup-instance
     runs-on: ${{ needs.setup-instance.outputs.runner-name }}
     concurrency:
-      group: ${{ github.workflow }}_${{ github.ref }}
+      group: ${{ github.workflow_ref }}
       cancel-in-progress: ${{ github.ref != 'refs/heads/main' }}
     continue-on-error: true
     steps:
@@ -51,7 +51,8 @@ jobs:
         uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
         with:
           fetch-depth: 0
-          token: ${{ secrets.FHE_ACTIONS_TOKEN }}
+          persist-credentials: 'false'
+          token: ${{ secrets.REPO_CHECKOUT_TOKEN }}
 
       - name: Get benchmark details
         run: |
@@ -62,7 +63,7 @@ jobs:
           } >> "${GITHUB_ENV}"
 
       - name: Install rust
-        uses: dtolnay/rust-toolchain@7b1c307e0dcbda6122208f10795a713336a9b35a
+        uses: dtolnay/rust-toolchain@888c2e1ea69ab0d4330cbf0af1ecc7b68f368cc1
         with:
           toolchain: nightly
 
@@ -93,7 +94,7 @@ jobs:
           --append-results
 
       - name: Upload parsed results artifact
-        uses: actions/upload-artifact@b4b15b8c7c6ac21ea08fcf65892d2ee8f75cf882
+        uses: actions/upload-artifact@ea165f8d65b6e75b540449e92b4886f43607fa02
         with:
           name: ${{ github.sha }}_boolean
           path: ${{ env.RESULTS_FILENAME }}
@@ -103,7 +104,8 @@ jobs:
         with:
           repository: zama-ai/slab
           path: slab
-          token: ${{ secrets.FHE_ACTIONS_TOKEN }}
+          persist-credentials: 'false'
+          token: ${{ secrets.REPO_CHECKOUT_TOKEN }}
 
       - name: Send data to Slab
         shell: bash
@@ -121,13 +123,13 @@ jobs:
 
   teardown-instance:
     name: Teardown instance (boolean-benchmarks)
-    if: ${{ always() && needs.setup-instance.result != 'skipped' }}
+    if: ${{ always() && needs.setup-instance.result == 'success' }}
     needs: [ setup-instance, boolean-benchmarks ]
     runs-on: ubuntu-latest
     steps:
       - name: Stop instance
         id: stop-instance
-        uses: zama-ai/slab-github-runner@801df0b8db5ea2b06128b7476c652f5ed5f193a8
+        uses: zama-ai/slab-github-runner@79939325c3c429837c10d6041e4fd8589d328bac
         with:
           mode: stop
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}

.github/workflows/benchmark_core_crypto.yml

@@ -3,6 +3,9 @@ name: Core crypto benchmarks
 
 on:
   workflow_dispatch:
+  schedule:
+    # Weekly benchmarks will be triggered each Saturday at 5a.m.
+    - cron: '0 5 * * 6'
 
 env:
   CARGO_TERM_COLOR: always
@@ -26,7 +29,7 @@ jobs:
     steps:
       - name: Start instance
         id: start-instance
-        uses: zama-ai/slab-github-runner@801df0b8db5ea2b06128b7476c652f5ed5f193a8
+        uses: zama-ai/slab-github-runner@79939325c3c429837c10d6041e4fd8589d328bac
         with:
           mode: start
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}
@@ -40,14 +43,15 @@ jobs:
     needs: setup-instance
     runs-on: ${{ needs.setup-instance.outputs.runner-name }}
     concurrency:
-      group: ${{ github.workflow }}_${{ github.ref }}
+      group: ${{ github.workflow_ref }}
       cancel-in-progress: ${{ github.ref != 'refs/heads/main' }}
     steps:
       - name: Checkout tfhe-rs repo with tags
         uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
         with:
           fetch-depth: 0
-          token: ${{ secrets.FHE_ACTIONS_TOKEN }}
+          persist-credentials: 'false'
+          token: ${{ secrets.REPO_CHECKOUT_TOKEN }}
 
       - name: Get benchmark details
         run: |
@@ -58,12 +62,13 @@ jobs:
           } >> "${GITHUB_ENV}"
 
       - name: Install rust
-        uses: dtolnay/rust-toolchain@7b1c307e0dcbda6122208f10795a713336a9b35a
+        uses: dtolnay/rust-toolchain@888c2e1ea69ab0d4330cbf0af1ecc7b68f368cc1
         with:
           toolchain: nightly
 
       - name: Run benchmarks with AVX512
         run: |
+          make bench_ks_pbs
           make bench_pbs
           make bench_pbs128
           make bench_ks
@@ -81,7 +86,7 @@ jobs:
           --walk-subdirs
 
       - name: Upload parsed results artifact
-        uses: actions/upload-artifact@b4b15b8c7c6ac21ea08fcf65892d2ee8f75cf882
+        uses: actions/upload-artifact@ea165f8d65b6e75b540449e92b4886f43607fa02
         with:
           name: ${{ github.sha }}_core_crypto
           path: ${{ env.RESULTS_FILENAME }}
@@ -91,7 +96,8 @@ jobs:
         with:
           repository: zama-ai/slab
           path: slab
-          token: ${{ secrets.FHE_ACTIONS_TOKEN }}
+          persist-credentials: 'false'
+          token: ${{ secrets.REPO_CHECKOUT_TOKEN }}
 
       - name: Send data to Slab
         shell: bash
@@ -109,13 +115,13 @@ jobs:
 
   teardown-instance:
     name: Teardown instance (core-crypto-benchmarks)
-    if: ${{ always() && needs.setup-instance.result != 'skipped' }}
+    if: ${{ always() && needs.setup-instance.result == 'success' }}
     needs: [ setup-instance, core-crypto-benchmarks ]
     runs-on: ubuntu-latest
     steps:
       - name: Stop instance
         id: stop-instance
-        uses: zama-ai/slab-github-runner@801df0b8db5ea2b06128b7476c652f5ed5f193a8
+        uses: zama-ai/slab-github-runner@79939325c3c429837c10d6041e4fd8589d328bac
         with:
           mode: stop
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}

.github/workflows/benchmark_erc20.yml

@@ -29,7 +29,7 @@ jobs:
     steps:
       - name: Start instance
         id: start-instance
-        uses: zama-ai/slab-github-runner@801df0b8db5ea2b06128b7476c652f5ed5f193a8
+        uses: zama-ai/slab-github-runner@79939325c3c429837c10d6041e4fd8589d328bac
         with:
           mode: start
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}
@@ -43,7 +43,7 @@ jobs:
     needs: setup-instance
     runs-on: ${{ needs.setup-instance.outputs.runner-name }}
     concurrency:
-      group: ${{ github.workflow }}_${{ github.ref }}
+      group: ${{ github.workflow_ref }}
       cancel-in-progress: ${{ github.ref != 'refs/heads/main' }}
     continue-on-error: true
     timeout-minutes: 720  # 12 hours
@@ -52,7 +52,8 @@ jobs:
         uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
         with:
           fetch-depth: 0
-          token: ${{ secrets.FHE_ACTIONS_TOKEN }}
+          persist-credentials: 'false'
+          token: ${{ secrets.REPO_CHECKOUT_TOKEN }}
 
       - name: Get benchmark details
         run: |
@@ -63,7 +64,7 @@ jobs:
           } >> "${GITHUB_ENV}"
 
       - name: Install rust
-        uses: dtolnay/rust-toolchain@7b1c307e0dcbda6122208f10795a713336a9b35a
+        uses: dtolnay/rust-toolchain@888c2e1ea69ab0d4330cbf0af1ecc7b68f368cc1
         with:
           toolchain: nightly
 
@@ -72,7 +73,8 @@ jobs:
         with:
           repository: zama-ai/slab
           path: slab
-          token: ${{ secrets.FHE_ACTIONS_TOKEN }}
+          persist-credentials: 'false'
+          token: ${{ secrets.REPO_CHECKOUT_TOKEN }}
 
       - name: Run benchmarks
         run: |
@@ -97,7 +99,7 @@ jobs:
           --append-results
 
       - name: Upload parsed results artifact
-        uses: actions/upload-artifact@b4b15b8c7c6ac21ea08fcf65892d2ee8f75cf882
+        uses: actions/upload-artifact@ea165f8d65b6e75b540449e92b4886f43607fa02
         with:
           name: ${{ github.sha }}_erc20
           path: ${{ env.RESULTS_FILENAME }}
@@ -118,13 +120,13 @@ jobs:
 
   teardown-instance:
     name: Teardown instance (erc20-benchmarks)
-    if: ${{ always() && needs.setup-instance.result != 'skipped' }}
+    if: ${{ always() && needs.setup-instance.result == 'success' }}
     needs: [ setup-instance, erc20-benchmarks ]
     runs-on: ubuntu-latest
     steps:
       - name: Stop instance
         id: stop-instance
-        uses: zama-ai/slab-github-runner@801df0b8db5ea2b06128b7476c652f5ed5f193a8
+        uses: zama-ai/slab-github-runner@79939325c3c429837c10d6041e4fd8589d328bac
         with:
           mode: stop
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}

.github/workflows/benchmark_gpu.yml

@@ -15,6 +15,7 @@ on:
           - "4-h100 (n3-H100x4)"
           - "multi-h100 (n3-H100x8)"
           - "multi-h100-nvlink (n3-H100x8-NVLink)"
+          - "multi-h100-sxm5 (n3-H100x8-SXM5)"
           - "multi-a100-nvlink (n3-A100x8-NVLink)"
       command:
         description: "Benchmark command to run"
@@ -25,6 +26,7 @@ on:
           - integer_multi_bit
           - integer_compression
           - pbs
+          - pbs128
           - ks
       op_flavor:
         description: "Operations set to run"
@@ -67,7 +69,7 @@ jobs:
   run-benchmarks:
     name: Run benchmarks
     needs: parse-inputs
-    uses: ./.github/workflows/benchmark_gpu_integer_common.yml
+    uses: ./.github/workflows/benchmark_gpu_common.yml
     with:
       profile: ${{ needs.parse-inputs.outputs.profile }}
       hardware_name: ${{ needs.parse-inputs.outputs.hardware_name }}

.github/workflows/benchmark_gpu_4090.yml

@@ -17,7 +17,7 @@ on:
   # Allows you to run this workflow manually from the Actions tab as an alternative.
   workflow_dispatch:
   pull_request:
-    types: [labeled]
+    types: [ labeled ]
   schedule:
     # Weekly benchmarks will be triggered each Friday at 9p.m.
     - cron: "0 21 * * 5"
@@ -29,20 +29,17 @@ jobs:
       github.event_name == 'schedule' && github.repository == 'zama-ai/tfhe-rs' ||
       contains(github.event.label.name, '4090_bench') }}
     concurrency:
-      group: ${{ github.workflow }}_${{ github.ref }}_cuda_integer_bench
+      group: ${{ github.workflow_ref }}_cuda_integer_bench
       cancel-in-progress: ${{ github.ref != 'refs/heads/main' }}
     runs-on: ["self-hosted", "4090-desktop"]
     timeout-minutes: 1440 # 24 hours
-    strategy:
-      fail-fast: false
-      max-parallel: 1
-
     steps:
       - name: Checkout tfhe-rs
         uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
         with:
           fetch-depth: 0
-          token: ${{ secrets.FHE_ACTIONS_TOKEN }}
+          persist-credentials: 'false'
+          token: ${{ secrets.REPO_CHECKOUT_TOKEN }}
 
       - name: Get benchmark details
         run: |
@@ -54,7 +51,7 @@ jobs:
           echo "FAST_BENCH=TRUE" >> "${GITHUB_ENV}"
 
       - name: Install rust
-        uses: dtolnay/rust-toolchain@7b1c307e0dcbda6122208f10795a713336a9b35a
+        uses: dtolnay/rust-toolchain@888c2e1ea69ab0d4330cbf0af1ecc7b68f368cc1
         with:
           toolchain: nightly
 
@@ -63,7 +60,8 @@ jobs:
         with:
           repository: zama-ai/slab
           path: slab
-          token: ${{ secrets.FHE_ACTIONS_TOKEN }}
+          persist-credentials: 'false'
+          token: ${{ secrets.REPO_CHECKOUT_TOKEN }}
 
       - name: Run integer benchmarks
         run: |
@@ -82,7 +80,7 @@ jobs:
           --walk-subdirs
 
       - name: Upload parsed results artifact
-        uses: actions/upload-artifact@b4b15b8c7c6ac21ea08fcf65892d2ee8f75cf882
+        uses: actions/upload-artifact@ea165f8d65b6e75b540449e92b4886f43607fa02
         with:
           name: ${{ github.sha }}_integer_multi_bit_gpu_default
           path: ${{ env.RESULTS_FILENAME }}
@@ -106,7 +104,7 @@ jobs:
     if: ${{ github.event_name == 'workflow_dispatch' || github.event_name == 'schedule' || contains(github.event.label.name, '4090_bench') }}
     needs: cuda-integer-benchmarks
     concurrency:
-      group: ${{ github.workflow }}_${{ github.ref }}_cuda_core_crypto_bench
+      group: ${{ github.workflow_ref }}_cuda_core_crypto_bench
       cancel-in-progress: ${{ github.ref != 'refs/heads/main' }}
     runs-on: ["self-hosted", "4090-desktop"]
     timeout-minutes: 1440 # 24 hours
@@ -116,6 +114,8 @@ jobs:
         uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
         with:
           fetch-depth: 0
+          persist-credentials: 'false'
+          token: ${{ secrets.REPO_CHECKOUT_TOKEN }}
 
       - name: Get benchmark details
         run: |
@@ -126,7 +126,7 @@ jobs:
           } >> "${GITHUB_ENV}"
 
       - name: Install rust
-        uses: dtolnay/rust-toolchain@7b1c307e0dcbda6122208f10795a713336a9b35a
+        uses: dtolnay/rust-toolchain@888c2e1ea69ab0d4330cbf0af1ecc7b68f368cc1
         with:
           toolchain: nightly
 
@@ -135,7 +135,8 @@ jobs:
         with:
           repository: zama-ai/slab
           path: slab
-          token: ${{ secrets.FHE_ACTIONS_TOKEN }}
+          persist-credentials: 'false'
+          token: ${{ secrets.REPO_CHECKOUT_TOKEN }}
 
       - name: Run core crypto benchmarks
         run: |
@@ -156,7 +157,7 @@ jobs:
       
 
       - name: Upload parsed results artifact
-        uses: actions/upload-artifact@b4b15b8c7c6ac21ea08fcf65892d2ee8f75cf882
+        uses: actions/upload-artifact@ea165f8d65b6e75b540449e92b4886f43607fa02
         with:
           name: ${{ github.sha }}_core_crypto
           path: ${{ env.RESULTS_FILENAME }}

.github/workflows/benchmark_gpu_common.yml

@@ -1,4 +1,4 @@
-# Run integer benchmarks on CUDA instance and return parsed results to Slab CI bot.
+# Run benchmarks on CUDA instance and return parsed results to Slab CI bot.
 name: Cuda benchmarks - common
 
 on:
@@ -26,7 +26,7 @@ on:
         type: boolean
         default: false
     secrets:
-      FHE_ACTIONS_TOKEN:
+      REPO_CHECKOUT_TOKEN:
         required: true
       SLAB_ACTION_TOKEN:
         required: true
@@ -83,7 +83,7 @@ jobs:
       - name: Set multiple operations flavors
         if: ${{ contains(inputs.op_flavor, ',')}}
         run: |
-          PARSED_OP_FLAVOR=$(echo "${{ inputs.op_flavor }}" | sed 's/[[:space:]]*,[[:space:]]*/\\", \\"/g')
+          PARSED_OP_FLAVOR=$(echo "${{ inputs.op_flavor }}" | sed 's/[[:space:]]*,[[:space:]]*/", "/g')
           echo "OP_FLAVOR=[\"${PARSED_OP_FLAVOR}\"]" >> "${GITHUB_ENV}"
 
       - name: Set benchmark types
@@ -114,11 +114,17 @@ jobs:
     needs: prepare-matrix
     runs-on: ubuntu-latest
     outputs:
-      runner-name: ${{ steps.start-instance.outputs.label }}
+      # Use permanent remote instance label first as on-demand remote instance label output is set before the end of start-remote-instance step.
+      # If the latter fails due to a failed GitHub action runner set up, we have to fallback on the permanent instance.
+      # Since the on-demand remote label is set before failure, we have to do the logical OR in this order,
+      # otherwise we'll try to run the next job on a non-existing on-demand instance.
+      runner-name: ${{ steps.use-permanent-instance.outputs.runner_group || steps.start-remote-instance.outputs.label }}
+      remote-instance-outcome: ${{ steps.start-remote-instance.outcome }}
     steps:
-      - name: Start instance
-        id: start-instance
-        uses: zama-ai/slab-github-runner@801df0b8db5ea2b06128b7476c652f5ed5f193a8
+      - name: Start remote instance
+        id: start-remote-instance
+        continue-on-error: true
+        uses: zama-ai/slab-github-runner@79939325c3c429837c10d6041e4fd8589d328bac
         with:
           mode: start
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}
@@ -127,9 +133,43 @@ jobs:
           backend: ${{ inputs.backend }}
           profile: ${{ inputs.profile }}
 
+      # This will allow to fallback on permanent instances running on Hyperstack.
+      - name: Use permanent remote instance
+        id: use-permanent-instance
+        if: steps.start-remote-instance.outcome == 'failure' &&
+          inputs.profile == 'single-h100'
+        run: |
+          echo "runner_group=h100x1" >> "$GITHUB_OUTPUT"
+
+  # Install dependencies only once since cuda-benchmarks uses a matrix strategy, thus running multiple times.
+  install-dependencies:
+    name: Install dependencies
+    needs: [ setup-instance ]
+    runs-on: ${{ needs.setup-instance.outputs.runner-name }}
+    strategy:
+      matrix:
+        # explicit include-based build matrix, of known valid options
+        include:
+          - cuda: "12.2"
+            gcc: 11
+    steps:
+      - name: Checkout tfhe-rs repo
+        uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
+        with:
+          fetch-depth: 0
+          persist-credentials: 'false'
+          token: ${{ secrets.REPO_CHECKOUT_TOKEN }}
+
+      - name: Setup Hyperstack dependencies
+        if: needs.setup-instance.outputs.remote-instance-outcome == 'success'
+        uses: ./.github/actions/gpu_setup
+        with:
+          cuda-version: ${{ matrix.cuda }}
+          gcc-version: ${{ matrix.gcc }}
+
   cuda-benchmarks:
     name: Cuda benchmarks (${{ inputs.profile }})
-    needs: [ prepare-matrix, setup-instance ]
+    needs: [ prepare-matrix, setup-instance, install-dependencies ]
     runs-on: ${{ needs.setup-instance.outputs.runner-name }}
     timeout-minutes: 1440 # 24 hours
     continue-on-error: true
@@ -140,32 +180,16 @@ jobs:
         command: ${{ fromJSON(needs.prepare-matrix.outputs.command) }}
         op_flavor: ${{ fromJSON(needs.prepare-matrix.outputs.op_flavor) }}
         bench_type: ${{ fromJSON(needs.prepare-matrix.outputs.bench_type) }}
-        # explicit include-based build matrix, of known valid options
         include:
-          - os: ubuntu-22.04
-            cuda: "12.2"
+          - cuda: "12.2"
             gcc: 11
-    env:
-      CUDA_PATH: /usr/local/cuda-${{ matrix.cuda }}
-      CMAKE_VERSION: 3.29.6
     steps:
-      # Mandatory on hyperstack since a bootable volume is not re-usable yet.
-      - name: Install dependencies
-        run: |
-          sudo apt update
-          sudo apt install -y checkinstall zlib1g-dev libssl-dev libclang-dev
-          wget https://github.com/Kitware/CMake/releases/download/v${{ env.CMAKE_VERSION }}/cmake-${{ env.CMAKE_VERSION }}.tar.gz
-          tar -zxvf cmake-${{ env.CMAKE_VERSION }}.tar.gz
-          cd cmake-${{ env.CMAKE_VERSION }}
-          ./bootstrap
-          make -j"$(nproc)"
-          sudo make install
-
       - name: Checkout tfhe-rs repo with tags
         uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
         with:
           fetch-depth: 0
-          token: ${{ secrets.FHE_ACTIONS_TOKEN }}
+          persist-credentials: 'false'
+          token: ${{ secrets.REPO_CHECKOUT_TOKEN }}
 
       - name: Get benchmark details
         run: |
@@ -175,47 +199,31 @@ jobs:
             echo "COMMIT_HASH=$(git describe --tags --dirty)";
           } >> "${GITHUB_ENV}"
 
-      - name: Set up home
-        # "Install rust" step require root user to have a HOME directory which is not set.
+      # Re-export environment variables as dependencies setup perform this task in the previous job.
+      # Local env variables are cleaned at the end of each job.
+      - name: Export CUDA variables
+        shell: bash
         run: |
-          echo "HOME=/home/ubuntu" >> "${GITHUB_ENV}"
+          CUDA_PATH=/usr/local/cuda-${{ matrix.cuda }}
+          echo "CUDA_PATH=$CUDA_PATH" >> "${GITHUB_ENV}"
+          echo "PATH=$PATH:$CUDA_PATH/bin" >> "${GITHUB_PATH}"
+          echo "LD_LIBRARY_PATH=$CUDA_PATH/lib64:$LD_LIBRARY_PATH" >> "${GITHUB_ENV}"
+          echo "CUDA_MODULE_LOADER=EAGER" >> "${GITHUB_ENV}"
+
+      - name: Export gcc and g++ variables
+        shell: bash
+        run: |
+          {
+          echo "CC=/usr/bin/gcc-${{ matrix.gcc }}";
+          echo "CXX=/usr/bin/g++-${{ matrix.gcc }}";
+          echo "CUDAHOSTCXX=/usr/bin/g++-${{ matrix.gcc }}";
+          } >> "${GITHUB_ENV}"
 
       - name: Install rust
-        uses: dtolnay/rust-toolchain@7b1c307e0dcbda6122208f10795a713336a9b35a
+        uses: dtolnay/rust-toolchain@888c2e1ea69ab0d4330cbf0af1ecc7b68f368cc1
         with:
           toolchain: nightly
 
-      - name: Export CUDA variables
-        if: ${{ !cancelled() }}
-        run: |
-          {
-            echo "CUDA_PATH=$CUDA_PATH";
-            echo "LD_LIBRARY_PATH=$CUDA_PATH/lib:$LD_LIBRARY_PATH";
-            echo "CUDACXX=/usr/local/cuda-${{ matrix.cuda }}/bin/nvcc";
-          } >> "${GITHUB_ENV}"
-          echo "$CUDA_PATH/bin" >> "${GITHUB_PATH}"
-
-      # Specify the correct host compilers
-      - name: Export gcc and g++ variables
-        if: ${{ !cancelled() }}
-        run: |
-          {
-            echo "CC=/usr/bin/gcc-${{ matrix.gcc }}";
-            echo "CXX=/usr/bin/g++-${{ matrix.gcc }}";
-            echo "CUDAHOSTCXX=/usr/bin/g++-${{ matrix.gcc }}";
-          } >> "${GITHUB_ENV}"
-
-      - name: Checkout Slab repo
-        uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
-        with:
-          repository: zama-ai/slab
-          path: slab
-          token: ${{ secrets.FHE_ACTIONS_TOKEN }}
-
-      - name: Check device is detected
-        if: ${{ !cancelled() }}
-        run: nvidia-smi
-
       - name: Should run benchmarks with all precisions
         if: inputs.all_precisions
         run: |
@@ -236,14 +244,23 @@ jobs:
           --commit-date "${{ env.COMMIT_DATE }}" \
           --bench-date "${{ env.BENCH_DATE }}" \
           --walk-subdirs \
-          --name-suffix avx512
+          --name-suffix avx512 \
+          --bench-type ${{ matrix.bench_type }}
 
       - name: Upload parsed results artifact
-        uses: actions/upload-artifact@b4b15b8c7c6ac21ea08fcf65892d2ee8f75cf882
+        uses: actions/upload-artifact@ea165f8d65b6e75b540449e92b4886f43607fa02
         with:
-          name: ${{ github.sha }}_${{ matrix.command }}_${{ matrix.op_flavor }}
+          name: ${{ github.sha }}_${{ matrix.command }}_${{ matrix.op_flavor }}_${{ inputs.profile }}
           path: ${{ env.RESULTS_FILENAME }}
 
+      - name: Checkout Slab repo
+        uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
+        with:
+          repository: zama-ai/slab
+          path: slab
+          persist-credentials: 'false'
+          token: ${{ secrets.REPO_CHECKOUT_TOKEN }}
+
       - name: Send data to Slab
         shell: bash
         run: |
@@ -265,13 +282,13 @@ jobs:
 
   teardown-instance:
     name: Teardown instance (cuda-${{ inputs.profile }}-benchmarks)
-    if: ${{ always() && needs.setup-instance.result != 'skipped' }}
+    if: ${{ always() && needs.setup-instance.outputs.remote-instance-outcome == 'success' }}
     needs: [ setup-instance, cuda-benchmarks, slack-notify ]
     runs-on: ubuntu-latest
     steps:
       - name: Stop instance
         id: stop-instance
-        uses: zama-ai/slab-github-runner@801df0b8db5ea2b06128b7476c652f5ed5f193a8
+        uses: zama-ai/slab-github-runner@79939325c3c429837c10d6041e4fd8589d328bac
         with:
           mode: stop
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}

.github/workflows/benchmark_gpu_core_crypto.yml

@@ -1,5 +1,5 @@
 # Run core crypto benchmarks on an instance with CUDA and return parsed results to Slab CI bot.
-name: Core crypto GPU benchmarks
+name: Cuda - Core crypto benchmarks
 
 on:
   workflow_dispatch:
@@ -23,11 +23,17 @@ jobs:
     if: github.event_name != 'schedule' ||
       (github.event_name == 'schedule' && github.repository == 'zama-ai/tfhe-rs')
     outputs:
-      runner-name: ${{ steps.start-instance.outputs.label }}
+      # Use permanent remote instance label first as on-demand remote instance label output is set before the end of start-remote-instance step.
+      # If the latter fails due to a failed GitHub action runner set up, we have to fallback on the permanent instance.
+      # Since the on-demand remote label is set before failure, we have to do the logical OR in this order,
+      # otherwise we'll try to run the next job on a non-existing on-demand instance.
+      runner-name: ${{ steps.use-permanent-instance.outputs.runner_group || steps.start-remote-instance.outputs.label }}
+      remote-instance-outcome: ${{ steps.start-remote-instance.outcome }}
     steps:
-      - name: Start instance
-        id: start-instance
-        uses: zama-ai/slab-github-runner@801df0b8db5ea2b06128b7476c652f5ed5f193a8
+      - name: Start remote instance
+        id: start-remote-instance
+        continue-on-error: true
+        uses: zama-ai/slab-github-runner@79939325c3c429837c10d6041e4fd8589d328bac
         with:
           mode: start
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}
@@ -36,6 +42,13 @@ jobs:
           backend: hyperstack
           profile: single-h100
 
+      # This will allow to fallback on permanent instances running on Hyperstack.
+      - name: Use permanent remote instance
+        id: use-permanent-instance
+        if: env.SECRETS_AVAILABLE == 'true' && steps.start-remote-instance.outcome == 'failure'
+        run: |
+          echo "runner_group=h100x1" >> "$GITHUB_OUTPUT"
+
   cuda-core-crypto-benchmarks:
     name: Execute GPU core crypto benchmarks
     needs: setup-instance
@@ -48,27 +61,20 @@ jobs:
           - os: ubuntu-22.04
             cuda: "12.2"
             gcc: 11
-    env:
-      CUDA_PATH: /usr/local/cuda-${{ matrix.cuda }}
-      CMAKE_VERSION: 3.29.6
     steps:
-      # Mandatory on hyperstack since a bootable volume is not re-usable yet.
-      - name: Install dependencies
-        run: |
-          sudo apt update
-          sudo apt install -y checkinstall zlib1g-dev libssl-dev libclang-dev
-          wget https://github.com/Kitware/CMake/releases/download/v${{ env.CMAKE_VERSION }}/cmake-${{ env.CMAKE_VERSION }}.tar.gz
-          tar -zxvf cmake-${{ env.CMAKE_VERSION }}.tar.gz
-          cd cmake-${{ env.CMAKE_VERSION }}
-          ./bootstrap
-          make -j"$(nproc)"
-          sudo make install
-
       - name: Checkout tfhe-rs repo with tags
         uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
         with:
           fetch-depth: 0
-          token: ${{ secrets.FHE_ACTIONS_TOKEN }}
+          persist-credentials: 'false'
+          token: ${{ secrets.REPO_CHECKOUT_TOKEN }}
+
+      - name: Setup Hyperstack dependencies
+        if: needs.setup-instance.outputs.remote-instance-outcome == 'success'
+        uses: ./.github/actions/gpu_setup
+        with:
+          cuda-version: ${{ matrix.cuda }}
+          gcc-version: ${{ matrix.gcc }}
 
       - name: Get benchmark details
         run: |
@@ -78,40 +84,16 @@ jobs:
             echo "COMMIT_HASH=$(git describe --tags --dirty)";
           } >> "${GITHUB_ENV}"
 
-      - name: Set up home
-        # "Install rust" step require root user to have a HOME directory which is not set.
-        run: |
-          echo "HOME=/home/ubuntu" >> "${GITHUB_ENV}"
-
       - name: Install rust
-        uses: dtolnay/rust-toolchain@7b1c307e0dcbda6122208f10795a713336a9b35a
+        uses: dtolnay/rust-toolchain@888c2e1ea69ab0d4330cbf0af1ecc7b68f368cc1
         with:
           toolchain: nightly
 
-      - name: Export CUDA variables
-        if: ${{ !cancelled() }}
-        run: |
-          {
-            echo "CUDA_PATH=$CUDA_PATH";
-            echo "LD_LIBRARY_PATH=$CUDA_PATH/lib:$LD_LIBRARY_PATH";
-            echo "CUDACXX=/usr/local/cuda-${{ matrix.cuda }}/bin/nvcc";
-          } >> "${GITHUB_ENV}"
-          echo "$CUDA_PATH/bin" >> "${GITHUB_PATH}"
-
-      # Specify the correct host compilers
-      - name: Export gcc and g++ variables
-        if: ${{ !cancelled() }}
-        run: |
-          {
-            echo "CC=/usr/bin/gcc-${{ matrix.gcc }}";
-            echo "CXX=/usr/bin/g++-${{ matrix.gcc }}";
-            echo "CUDAHOSTCXX=/usr/bin/g++-${{ matrix.gcc }}";
-            echo "HOME=/home/ubuntu";
-          } >> "${GITHUB_ENV}"
-
       - name: Run benchmarks with AVX512
         run: |
+          make bench_ks_pbs_gpu
           make bench_pbs_gpu
+          make bench_pbs128_gpu
           make bench_ks_gpu
 
       - name: Parse results
@@ -128,7 +110,7 @@ jobs:
           --walk-subdirs
 
       - name: Upload parsed results artifact
-        uses: actions/upload-artifact@b4b15b8c7c6ac21ea08fcf65892d2ee8f75cf882
+        uses: actions/upload-artifact@ea165f8d65b6e75b540449e92b4886f43607fa02
         with:
           name: ${{ github.sha }}_core_crypto
           path: ${{ env.RESULTS_FILENAME }}
@@ -138,7 +120,8 @@ jobs:
         with:
           repository: zama-ai/slab
           path: slab
-          token: ${{ secrets.FHE_ACTIONS_TOKEN }}
+          persist-credentials: 'false'
+          token: ${{ secrets.REPO_CHECKOUT_TOKEN }}
 
       - name: Send data to Slab
         shell: bash
@@ -161,13 +144,13 @@ jobs:
 
   teardown-instance:
     name: Teardown instance (cuda-integer-full-benchmarks)
-    if: ${{ always() && needs.setup-instance.result != 'skipped' }}
+    if: ${{ always() && needs.setup-instance.outputs.remote-instance-outcome == 'success' }}
     needs: [ setup-instance, cuda-core-crypto-benchmarks, slack-notify ]
     runs-on: ubuntu-latest
     steps:
       - name: Stop instance
         id: stop-instance
-        uses: zama-ai/slab-github-runner@801df0b8db5ea2b06128b7476c652f5ed5f193a8
+        uses: zama-ai/slab-github-runner@79939325c3c429837c10d6041e4fd8589d328bac
         with:
           mode: stop
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}

.github/workflows/benchmark_gpu_erc20.yml

@@ -12,7 +12,10 @@ on:
           - "l40 (n3-L40x1)"
           - "single-h100 (n3-H100x1)"
           - "2-h100 (n3-H100x2)"
+          - "4-h100 (n3-H100x4)"
           - "multi-h100 (n3-H100x8)"
+          - "multi-h100-nvlink (n3-H100x8-NVLink)"
+          - "multi-h100-sxm5 (n3-H100x8-SXM5)"
 
 jobs:
   parse-inputs:

.github/workflows/benchmark_gpu_erc20_common.yml

@@ -14,7 +14,7 @@ on:
         type: string
         required: true
     secrets:
-      FHE_ACTIONS_TOKEN:
+      REPO_CHECKOUT_TOKEN:
         required: true
       SLAB_ACTION_TOKEN:
         required: true
@@ -50,11 +50,17 @@ jobs:
     if:  github.event_name == 'workflow_dispatch' ||
       (github.event_name == 'schedule' && github.repository == 'zama-ai/tfhe-rs')
     outputs:
-      runner-name: ${{ steps.start-instance.outputs.label }}
+      # Use permanent remote instance label first as on-demand remote instance label output is set before the end of start-remote-instance step.
+      # If the latter fails due to a failed GitHub action runner set up, we have to fallback on the permanent instance.
+      # Since the on-demand remote label is set before failure, we have to do the logical OR in this order,
+      # otherwise we'll try to run the next job on a non-existing on-demand instance.
+      runner-name: ${{ steps.use-permanent-instance.outputs.runner_group || steps.start-remote-instance.outputs.label }}
+      remote-instance-outcome: ${{ steps.start-remote-instance.outcome }}
     steps:
-      - name: Start instance
-        id: start-instance
-        uses: zama-ai/slab-github-runner@801df0b8db5ea2b06128b7476c652f5ed5f193a8
+      - name: Start remote instance
+        id: start-remote-instance
+        continue-on-error: true
+        uses: zama-ai/slab-github-runner@79939325c3c429837c10d6041e4fd8589d328bac
         with:
           mode: start
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}
@@ -63,6 +69,15 @@ jobs:
           backend: ${{ inputs.backend }}
           profile: ${{ inputs.profile }}
 
+      # This will allow to fallback on permanent instances running on Hyperstack.
+      - name: Use permanent remote instance
+        id: use-permanent-instance
+        if: env.SECRETS_AVAILABLE == 'true' &&
+          steps.start-remote-instance.outcome == 'failure' &&
+          inputs.profile == 'single-h100'
+        run: |
+          echo "runner_group=h100x1" >> "$GITHUB_OUTPUT"
+
   cuda-erc20-benchmarks:
     name: Cuda ERC20 benchmarks (${{ inputs.profile }})
     needs: setup-instance
@@ -75,27 +90,20 @@ jobs:
           - os: ubuntu-22.04
             cuda: "12.2"
             gcc: 11
-    env:
-      CUDA_PATH: /usr/local/cuda-${{ matrix.cuda }}
-      CMAKE_VERSION: 3.29.6
     steps:
-      # Mandatory on hyperstack since a bootable volume is not re-usable yet.
-      - name: Install dependencies
-        run: |
-          sudo apt update
-          sudo apt install -y checkinstall zlib1g-dev libssl-dev
-          wget https://github.com/Kitware/CMake/releases/download/v${{ env.CMAKE_VERSION }}/cmake-${{ env.CMAKE_VERSION }}.tar.gz
-          tar -zxvf cmake-${{ env.CMAKE_VERSION }}.tar.gz
-          cd cmake-${{ env.CMAKE_VERSION }}
-          ./bootstrap
-          make -j"$(nproc)"
-          sudo make install
-
       - name: Checkout tfhe-rs repo with tags
         uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
         with:
           fetch-depth: 0
-          token: ${{ secrets.FHE_ACTIONS_TOKEN }}
+          persist-credentials: 'false'
+          token: ${{ secrets.REPO_CHECKOUT_TOKEN }}
+
+      - name: Setup Hyperstack dependencies
+        if: needs.setup-instance.outputs.remote-instance-outcome == 'success'
+        uses: ./.github/actions/gpu_setup
+        with:
+          cuda-version: ${{ matrix.cuda }}
+          gcc-version: ${{ matrix.gcc }}
 
       - name: Get benchmark details
         run: |
@@ -105,40 +113,11 @@ jobs:
             echo "COMMIT_HASH=$(git describe --tags --dirty)";
           } >> "${GITHUB_ENV}"
 
-      - name: Set up home
-        # "Install rust" step require root user to have a HOME directory which is not set.
-        run: |
-          echo "HOME=/home/ubuntu" >> "${GITHUB_ENV}"
-
       - name: Install rust
-        uses: dtolnay/rust-toolchain@7b1c307e0dcbda6122208f10795a713336a9b35a
+        uses: dtolnay/rust-toolchain@888c2e1ea69ab0d4330cbf0af1ecc7b68f368cc1
         with:
           toolchain: nightly
 
-      - name: Export CUDA variables
-        if: ${{ !cancelled() }}
-        run: |
-          {
-            echo "CUDA_PATH=$CUDA_PATH";
-            echo "LD_LIBRARY_PATH=$CUDA_PATH/lib:$LD_LIBRARY_PATH";
-            echo "CUDACXX=/usr/local/cuda-${{ matrix.cuda }}/bin/nvcc";
-          } >> "${GITHUB_ENV}"
-          echo "$CUDA_PATH/bin" >> "${GITHUB_PATH}"
-
-      # Specify the correct host compilers
-      - name: Export gcc and g++ variables
-        if: ${{ !cancelled() }}
-        run: |
-          {
-            echo "CC=/usr/bin/gcc-${{ matrix.gcc }}";
-            echo "CXX=/usr/bin/g++-${{ matrix.gcc }}";
-            echo "CUDAHOSTCXX=/usr/bin/g++-${{ matrix.gcc }}";
-          } >> "${GITHUB_ENV}"
-
-      - name: Check device is detected
-        if: ${{ !cancelled() }}
-        run: nvidia-smi
-
       - name: Run benchmarks
         run: |
           make bench_hlapi_erc20_gpu
@@ -157,9 +136,9 @@ jobs:
           --name-suffix avx512
 
       - name: Upload parsed results artifact
-        uses: actions/upload-artifact@b4b15b8c7c6ac21ea08fcf65892d2ee8f75cf882
+        uses: actions/upload-artifact@ea165f8d65b6e75b540449e92b4886f43607fa02
         with:
-          name: ${{ github.sha }}_erc20
+          name: ${{ github.sha }}_erc20_${{ inputs.profile }}
           path: ${{ env.RESULTS_FILENAME }}
 
       - name: Checkout Slab repo
@@ -167,7 +146,8 @@ jobs:
         with:
           repository: zama-ai/slab
           path: slab
-          token: ${{ secrets.FHE_ACTIONS_TOKEN }}
+          persist-credentials: 'false'
+          token: ${{ secrets.REPO_CHECKOUT_TOKEN }}
 
       - name: Send data to Slab
         shell: bash
@@ -190,13 +170,13 @@ jobs:
 
   teardown-instance:
     name: Teardown instance (cuda-erc20-${{ inputs.profile }}-benchmarks)
-    if: ${{ always() && needs.setup-instance.result != 'skipped' }}
+    if: ${{ always() && needs.setup-instance.outputs.remote-instance-outcome == 'success' }}
     needs: [ setup-instance, cuda-erc20-benchmarks, slack-notify ]
     runs-on: ubuntu-latest
     steps:
       - name: Stop instance
         id: stop-instance
-        uses: zama-ai/slab-github-runner@801df0b8db5ea2b06128b7476c652f5ed5f193a8
+        uses: zama-ai/slab-github-runner@79939325c3c429837c10d6041e4fd8589d328bac
         with:
           mode: stop
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}

.github/workflows/benchmark_gpu_weekly.yml

@@ -10,7 +10,7 @@ jobs:
   run-benchmarks-1-h100:
     name: Run benchmarks (1xH100)
     if: github.repository == 'zama-ai/tfhe-rs'
-    uses: ./.github/workflows/benchmark_gpu_integer_common.yml
+    uses: ./.github/workflows/benchmark_gpu_common.yml
     with:
       profile: single-h100
       hardware_name: n3-H100x1
@@ -23,7 +23,7 @@ jobs:
   run-benchmarks-2-h100:
     name: Run benchmarks (2xH100)
     if: github.repository == 'zama-ai/tfhe-rs'
-    uses: ./.github/workflows/benchmark_gpu_integer_common.yml
+    uses: ./.github/workflows/benchmark_gpu_common.yml
     with:
       profile: 2-h100
       hardware_name: n3-H100x2
@@ -36,7 +36,7 @@ jobs:
   run-benchmarks-8-h100:
     name: Run benchmarks (8xH100)
     if: github.repository == 'zama-ai/tfhe-rs'
-    uses: ./.github/workflows/benchmark_gpu_integer_common.yml
+    uses: ./.github/workflows/benchmark_gpu_common.yml
     with:
       profile: multi-h100
       hardware_name: n3-H100x8
@@ -49,7 +49,7 @@ jobs:
   run-benchmarks-l40:
     name: Run benchmarks (L40)
     if: github.repository == 'zama-ai/tfhe-rs'
-    uses: ./.github/workflows/benchmark_gpu_integer_common.yml
+    uses: ./.github/workflows/benchmark_gpu_common.yml
     with:
       profile: l40
       hardware_name: n3-L40x1

.github/workflows/benchmark_integer.yml

@@ -35,7 +35,6 @@ env:
   SLACK_USERNAME: ${{ secrets.BOT_USERNAME }}
   SLACK_WEBHOOK: ${{ secrets.SLACK_WEBHOOK }}
   FAST_BENCH: TRUE
-  BENCH_TYPE: latency
 
 jobs:
   prepare-matrix:
@@ -91,7 +90,7 @@ jobs:
     steps:
       - name: Start instance
         id: start-instance
-        uses: zama-ai/slab-github-runner@801df0b8db5ea2b06128b7476c652f5ed5f193a8
+        uses: zama-ai/slab-github-runner@79939325c3c429837c10d6041e4fd8589d328bac
         with:
           mode: start
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}
@@ -105,7 +104,7 @@ jobs:
     needs: [ prepare-matrix, setup-instance ]
     runs-on: ${{ needs.setup-instance.outputs.runner-name }}
     concurrency:
-      group: ${{ github.workflow }}_${{ github.ref }}
+      group: ${{ github.workflow_ref }}
       cancel-in-progress: ${{ github.ref != 'refs/heads/main' }}
     continue-on-error: true
     timeout-minutes: 1440  # 24 hours
@@ -120,7 +119,8 @@ jobs:
         uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
         with:
           fetch-depth: 0
-          token: ${{ secrets.FHE_ACTIONS_TOKEN }}
+          persist-credentials: 'false'
+          token: ${{ secrets.REPO_CHECKOUT_TOKEN }}
 
       - name: Get benchmark details
         run: |
@@ -131,7 +131,7 @@ jobs:
           } >> "${GITHUB_ENV}"
 
       - name: Install rust
-        uses: dtolnay/rust-toolchain@7b1c307e0dcbda6122208f10795a713336a9b35a
+        uses: dtolnay/rust-toolchain@888c2e1ea69ab0d4330cbf0af1ecc7b68f368cc1
         with:
           toolchain: nightly
 
@@ -140,7 +140,8 @@ jobs:
         with:
           repository: zama-ai/slab
           path: slab
-          token: ${{ secrets.FHE_ACTIONS_TOKEN }}
+          persist-credentials: 'false'
+          token: ${{ secrets.REPO_CHECKOUT_TOKEN }}
 
       - name: Should run benchmarks with all precisions
         if: inputs.all_precisions
@@ -168,10 +169,10 @@ jobs:
           --bench-date "${{ env.BENCH_DATE }}" \
           --walk-subdirs \
           --name-suffix avx512 \
-          --bench-type ${{ env.BENCH_TYPE }}
+          --bench-type ${{ matrix.bench_type }}
 
       - name: Upload parsed results artifact
-        uses: actions/upload-artifact@b4b15b8c7c6ac21ea08fcf65892d2ee8f75cf882
+        uses: actions/upload-artifact@ea165f8d65b6e75b540449e92b4886f43607fa02
         with:
           name: ${{ github.sha }}_${{ matrix.command }}_${{ matrix.op_flavor }}_${{ matrix.bench_type }}
           path: ${{ env.RESULTS_FILENAME }}
@@ -192,13 +193,13 @@ jobs:
 
   teardown-instance:
     name: Teardown instance (integer-benchmarks)
-    if: ${{ always() && needs.setup-instance.result != 'skipped' }}
+    if: ${{ always() && needs.setup-instance.result == 'success' }}
     needs: [ setup-instance, integer-benchmarks ]
     runs-on: ubuntu-latest
     steps:
       - name: Stop instance
         id: stop-instance
-        uses: zama-ai/slab-github-runner@801df0b8db5ea2b06128b7476c652f5ed5f193a8
+        uses: zama-ai/slab-github-runner@79939325c3c429837c10d6041e4fd8589d328bac
         with:
           mode: stop
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}

.github/workflows/benchmark_shortint.yml

@@ -56,7 +56,7 @@ jobs:
     steps:
       - name: Start instance
         id: start-instance
-        uses: zama-ai/slab-github-runner@801df0b8db5ea2b06128b7476c652f5ed5f193a8
+        uses: zama-ai/slab-github-runner@79939325c3c429837c10d6041e4fd8589d328bac
         with:
           mode: start
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}
@@ -70,7 +70,7 @@ jobs:
     needs: [ prepare-matrix, setup-instance ]
     runs-on: ${{ needs.setup-instance.outputs.runner-name }}
     concurrency:
-      group: ${{ github.workflow }}_${{ github.ref }}
+      group: ${{ github.workflow_ref }}
       cancel-in-progress: ${{ github.ref != 'refs/heads/main' }}
     continue-on-error: true
     strategy:
@@ -82,7 +82,8 @@ jobs:
         uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
         with:
           fetch-depth: 0
-          token: ${{ secrets.FHE_ACTIONS_TOKEN }}
+          persist-credentials: 'false'
+          token: ${{ secrets.REPO_CHECKOUT_TOKEN }}
 
       - name: Get benchmark details
         run: |
@@ -93,7 +94,7 @@ jobs:
           } >> "${GITHUB_ENV}"
 
       - name: Install rust
-        uses: dtolnay/rust-toolchain@7b1c307e0dcbda6122208f10795a713336a9b35a
+        uses: dtolnay/rust-toolchain@888c2e1ea69ab0d4330cbf0af1ecc7b68f368cc1
         with:
           toolchain: nightly
 
@@ -102,7 +103,8 @@ jobs:
         with:
           repository: zama-ai/slab
           path: slab
-          token: ${{ secrets.FHE_ACTIONS_TOKEN }}
+          persist-credentials: 'false'
+          token: ${{ secrets.REPO_CHECKOUT_TOKEN }}
 
       - name: Run benchmarks with AVX512
         run: |
@@ -136,7 +138,7 @@ jobs:
           --append-results
 
       - name: Upload parsed results artifact
-        uses: actions/upload-artifact@b4b15b8c7c6ac21ea08fcf65892d2ee8f75cf882
+        uses: actions/upload-artifact@ea165f8d65b6e75b540449e92b4886f43607fa02
         with:
           name: ${{ github.sha }}_shortint_${{ matrix.op_flavor }}
           path: ${{ env.RESULTS_FILENAME }}
@@ -157,13 +159,13 @@ jobs:
 
   teardown-instance:
     name: Teardown instance (shortint-benchmarks)
-    if: ${{ always() && needs.setup-instance.result != 'skipped' }}
+    if: ${{ always() && needs.setup-instance.result == 'success' }}
     needs: [ setup-instance, shortint-benchmarks ]
     runs-on: ubuntu-latest
     steps:
       - name: Stop instance
         id: stop-instance
-        uses: zama-ai/slab-github-runner@801df0b8db5ea2b06128b7476c652f5ed5f193a8
+        uses: zama-ai/slab-github-runner@79939325c3c429837c10d6041e4fd8589d328bac
         with:
           mode: stop
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}

.github/workflows/benchmark_signed_integer.yml

@@ -90,7 +90,7 @@ jobs:
     steps:
       - name: Start instance
         id: start-instance
-        uses: zama-ai/slab-github-runner@801df0b8db5ea2b06128b7476c652f5ed5f193a8
+        uses: zama-ai/slab-github-runner@79939325c3c429837c10d6041e4fd8589d328bac
         with:
           mode: start
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}
@@ -104,7 +104,7 @@ jobs:
     needs: [ prepare-matrix, setup-instance ]
     runs-on: ${{ needs.setup-instance.outputs.runner-name }}
     concurrency:
-      group: ${{ github.workflow }}_${{ github.ref }}
+      group: ${{ github.workflow_ref }}
       cancel-in-progress: ${{ github.ref != 'refs/heads/main' }}
     continue-on-error: true
     timeout-minutes: 1440  # 24 hours
@@ -119,7 +119,8 @@ jobs:
         uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
         with:
           fetch-depth: 0
-          token: ${{ secrets.FHE_ACTIONS_TOKEN }}
+          persist-credentials: 'false'
+          token: ${{ secrets.REPO_CHECKOUT_TOKEN }}
 
       - name: Get benchmark details
         run: |
@@ -130,7 +131,7 @@ jobs:
           } >> "${GITHUB_ENV}"
 
       - name: Install rust
-        uses: dtolnay/rust-toolchain@7b1c307e0dcbda6122208f10795a713336a9b35a
+        uses: dtolnay/rust-toolchain@888c2e1ea69ab0d4330cbf0af1ecc7b68f368cc1
         with:
           toolchain: nightly
 
@@ -139,7 +140,8 @@ jobs:
         with:
           repository: zama-ai/slab
           path: slab
-          token: ${{ secrets.FHE_ACTIONS_TOKEN }}
+          persist-credentials: 'false'
+          token: ${{ secrets.REPO_CHECKOUT_TOKEN }}
 
       - name: Should run benchmarks with all precisions
         if: inputs.all_precisions
@@ -164,7 +166,7 @@ jobs:
           --bench-type ${{ matrix.bench_type }}
 
       - name: Upload parsed results artifact
-        uses: actions/upload-artifact@b4b15b8c7c6ac21ea08fcf65892d2ee8f75cf882
+        uses: actions/upload-artifact@ea165f8d65b6e75b540449e92b4886f43607fa02
         with:
           name: ${{ github.sha }}_${{ matrix.command }}_${{ matrix.op_flavor }}_${{ matrix.bench_type }}
           path: ${{ env.RESULTS_FILENAME }}
@@ -185,13 +187,13 @@ jobs:
 
   teardown-instance:
     name: Teardown instance (integer-benchmarks)
-    if: ${{ always() && needs.setup-instance.result != 'skipped' }}
+    if: ${{ always() && needs.setup-instance.result == 'success' }}
     needs: [ setup-instance, signed-integer-benchmarks ]
     runs-on: ubuntu-latest
     steps:
       - name: Stop instance
         id: stop-instance
-        uses: zama-ai/slab-github-runner@801df0b8db5ea2b06128b7476c652f5ed5f193a8
+        uses: zama-ai/slab-github-runner@79939325c3c429837c10d6041e4fd8589d328bac
         with:
           mode: stop
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}

.github/workflows/benchmark_tfhe_fft.yml

@@ -16,6 +16,9 @@ on:
   push:
     branches:
       - "main"
+    paths:
+      - tfhe-fft/**
+      - .github/workflows/benchmark_tfhe_fft.yml
   schedule:
     # Job will be triggered each Thursday at 11p.m.
     - cron: '0 23 * * 4'
@@ -29,7 +32,7 @@ jobs:
     steps:
       - name: Start instance
         id: start-instance
-        uses: zama-ai/slab-github-runner@1dced74825027fe3d481392163ed8fc56813fb5d
+        uses: zama-ai/slab-github-runner@79939325c3c429837c10d6041e4fd8589d328bac
         with:
           mode: start
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}
@@ -42,12 +45,12 @@ jobs:
     name: Execute FFT benchmarks in EC2
     needs: setup-ec2
     concurrency:
-      group: ${{ github.workflow }}_${{ github.ref }}
+      group: ${{ github.workflow_ref }}
       cancel-in-progress: true
     runs-on: ${{ needs.setup-ec2.outputs.runner-name }}
     steps:
       - name: Checkout tfhe-rs repo with tags
-        uses: actions/checkout@ac593985615ec2ede58e132d2e21d2b1cbd6127c
+        uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
         with:
           fetch-depth: 0
 
@@ -81,17 +84,18 @@ jobs:
           --name-suffix avx512
 
       - name: Upload parsed results artifact
-        uses: actions/upload-artifact@0b7f8abb1508181956e8e162db84b466c27e18ce
+        uses: actions/upload-artifact@ea165f8d65b6e75b540449e92b4886f43607fa02
         with:
           name: ${{ github.sha }}_fft
           path: ${{ env.RESULTS_FILENAME }}
 
       - name: Checkout Slab repo
-        uses: actions/checkout@ac593985615ec2ede58e132d2e21d2b1cbd6127c
+        uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
         with:
           repository: zama-ai/slab
           path: slab
-          token: ${{ secrets.FHE_ACTIONS_TOKEN }}
+          persist-credentials: 'false'
+          token: ${{ secrets.REPO_CHECKOUT_TOKEN }}
 
       - name: Send data to Slab
         shell: bash
@@ -110,7 +114,7 @@ jobs:
       - name: Slack Notification
         if: ${{ failure() }}
         continue-on-error: true
-        uses: rtCamp/action-slack-notify@b24d75fe0e728a4bf9fc42ee217caa686d141ee8
+        uses: rtCamp/action-slack-notify@c33737706dea87cd7784c687dadc9adf1be59990
         env:
           SLACK_COLOR: ${{ job.status }}
           SLACK_MESSAGE: "tfhe-fft benchmarks failed. (${{ env.ACTION_RUN_URL }})"
@@ -123,7 +127,7 @@ jobs:
     steps:
       - name: Stop instance
         id: stop-instance
-        uses: zama-ai/slab-github-runner@1dced74825027fe3d481392163ed8fc56813fb5d
+        uses: zama-ai/slab-github-runner@79939325c3c429837c10d6041e4fd8589d328bac
         with:
           mode: stop
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}
@@ -134,7 +138,7 @@ jobs:
       - name: Slack Notification
         if: ${{ failure() }}
         continue-on-error: true
-        uses: rtCamp/action-slack-notify@b24d75fe0e728a4bf9fc42ee217caa686d141ee8
+        uses: rtCamp/action-slack-notify@c33737706dea87cd7784c687dadc9adf1be59990
         env:
           SLACK_COLOR: ${{ job.status }}
           SLACK_MESSAGE: "EC2 teardown (fft-benchmarks) failed. (${{ env.ACTION_RUN_URL }})"

.github/workflows/benchmark_tfhe_ntt.yml

@@ -16,6 +16,9 @@ on:
   push:
     branches:
       - "main"
+    paths:
+      - tfhe-ntt/**
+      - .github/workflows/benchmark_tfhe_ntt.yml
   schedule:
     # Job will be triggered each Friday at 11p.m.
     - cron: "0 23 * * 5"
@@ -29,7 +32,7 @@ jobs:
     steps:
       - name: Start instance
         id: start-instance
-        uses: zama-ai/slab-github-runner@1dced74825027fe3d481392163ed8fc56813fb5d
+        uses: zama-ai/slab-github-runner@79939325c3c429837c10d6041e4fd8589d328bac
         with:
           mode: start
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}
@@ -42,12 +45,12 @@ jobs:
     name: Execute NTT benchmarks in EC2
     needs: setup-ec2
     concurrency:
-      group: ${{ github.workflow }}_${{ github.ref }}
+      group: ${{ github.workflow_ref }}
       cancel-in-progress: true
     runs-on: ${{ needs.setup-ec2.outputs.runner-name }}
     steps:
       - name: Checkout tfhe-rs repo with tags
-        uses: actions/checkout@ac593985615ec2ede58e132d2e21d2b1cbd6127c
+        uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
         with:
           fetch-depth: 0
 
@@ -81,17 +84,18 @@ jobs:
           --name-suffix avx512
 
       - name: Upload parsed results artifact
-        uses: actions/upload-artifact@0b7f8abb1508181956e8e162db84b466c27e18ce
+        uses: actions/upload-artifact@ea165f8d65b6e75b540449e92b4886f43607fa02
         with:
           name: ${{ github.sha }}_ntt
           path: ${{ env.RESULTS_FILENAME }}
 
       - name: Checkout Slab repo
-        uses: actions/checkout@ac593985615ec2ede58e132d2e21d2b1cbd6127c
+        uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
         with:
           repository: zama-ai/slab
           path: slab
-          token: ${{ secrets.FHE_ACTIONS_TOKEN }}
+          persist-credentials: 'false'
+          token: ${{ secrets.REPO_CHECKOUT_TOKEN }}
 
       - name: Send data to Slab
         shell: bash
@@ -110,7 +114,7 @@ jobs:
       - name: Slack Notification
         if: ${{ failure() }}
         continue-on-error: true
-        uses: rtCamp/action-slack-notify@b24d75fe0e728a4bf9fc42ee217caa686d141ee8
+        uses: rtCamp/action-slack-notify@c33737706dea87cd7784c687dadc9adf1be59990
         env:
           SLACK_COLOR: ${{ job.status }}
           SLACK_MESSAGE: "tfhe-ntt benchmarks failed. (${{ env.ACTION_RUN_URL }})"
@@ -123,7 +127,7 @@ jobs:
     steps:
       - name: Stop instance
         id: stop-instance
-        uses: zama-ai/slab-github-runner@1dced74825027fe3d481392163ed8fc56813fb5d
+        uses: zama-ai/slab-github-runner@79939325c3c429837c10d6041e4fd8589d328bac
         with:
           mode: stop
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}
@@ -134,7 +138,7 @@ jobs:
       - name: Slack Notification
         if: ${{ failure() }}
         continue-on-error: true
-        uses: rtCamp/action-slack-notify@b24d75fe0e728a4bf9fc42ee217caa686d141ee8
+        uses: rtCamp/action-slack-notify@c33737706dea87cd7784c687dadc9adf1be59990
         env:
           SLACK_COLOR: ${{ job.status }}
           SLACK_MESSAGE: "EC2 teardown (ntt-benchmarks) failed. (${{ env.ACTION_RUN_URL }})"

.github/workflows/benchmark_tfhe_zk_pok.yml

@@ -3,6 +3,14 @@ name: tfhe-zk-pok benchmarks
 
 on:
   workflow_dispatch:
+    inputs:
+      bench_type:
+        description: "Benchmarks type"
+        type: choice
+        default: latency
+        options:
+          - latency
+          - throughput
   push:
     branches:
       - main
@@ -20,6 +28,7 @@ env:
   SLACK_ICON: https://pbs.twimg.com/profile_images/1274014582265298945/OjBKP9kn_400x400.png
   SLACK_USERNAME: ${{ secrets.BOT_USERNAME }}
   SLACK_WEBHOOK: ${{ secrets.SLACK_WEBHOOK }}
+  BENCH_TYPE: ${{ inputs.bench_type || 'latency' }}
 
 jobs:
   should-run:
@@ -36,9 +45,8 @@ jobs:
 
       - name: Check for file changes
         id: changed-files
-        uses: tj-actions/changed-files@c3a1bb2c992d77180ae65be6ae6c166cf40f857c
+        uses: tj-actions/changed-files@26a38635fc1173cc5820336ce97be6188d0de9f5 # v46.0.2
         with:
-          since_last_remote_commit: true
           files_yaml: |
             zk_pok:
               - tfhe-zk-pok/**
@@ -58,7 +66,7 @@ jobs:
     steps:
       - name: Start instance
         id: start-instance
-        uses: zama-ai/slab-github-runner@801df0b8db5ea2b06128b7476c652f5ed5f193a8
+        uses: zama-ai/slab-github-runner@79939325c3c429837c10d6041e4fd8589d328bac
         with:
           mode: start
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}
@@ -72,7 +80,7 @@ jobs:
     if: needs.setup-instance.result != 'skipped'
     needs: setup-instance
     concurrency:
-      group: ${{ github.workflow }}_${{github.event_name}}_${{ github.ref }}${{ github.ref == 'refs/heads/main' && github.sha || '' }}
+      group: ${{ github.workflow_ref }}_${{github.event_name}}${{ github.ref == 'refs/heads/main' && github.sha || '' }}
       cancel-in-progress: ${{ github.ref != 'refs/heads/main' }}
     runs-on: ${{ needs.setup-instance.outputs.runner-name }}
     steps:
@@ -80,7 +88,8 @@ jobs:
         uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
         with:
           fetch-depth: 0
-          token: ${{ secrets.FHE_ACTIONS_TOKEN }}
+          persist-credentials: 'false'
+          token: ${{ secrets.REPO_CHECKOUT_TOKEN }}
 
       - name: Get benchmark details
         run: |
@@ -91,7 +100,7 @@ jobs:
           } >> "${GITHUB_ENV}"
 
       - name: Install rust
-        uses: dtolnay/rust-toolchain@7b1c307e0dcbda6122208f10795a713336a9b35a
+        uses: dtolnay/rust-toolchain@888c2e1ea69ab0d4330cbf0af1ecc7b68f368cc1
         with:
           toolchain: nightly
 
@@ -100,11 +109,12 @@ jobs:
         with:
           repository: zama-ai/slab
           path: slab
-          token: ${{ secrets.FHE_ACTIONS_TOKEN }}
+          persist-credentials: 'false'
+          token: ${{ secrets.REPO_CHECKOUT_TOKEN }}
 
       - name: Run benchmarks
         run: |
-          make bench_tfhe_zk_pok
+          make BENCH_TYPE=${{ env.BENCH_TYPE }} bench_tfhe_zk_pok
 
       - name: Parse results
         run: |
@@ -119,10 +129,10 @@ jobs:
           --bench-date "${{ env.BENCH_DATE }}" \
           --walk-subdirs \
           --name-suffix avx512 \
-          --throughput
+          --bench-type ${{ env.BENCH_TYPE }}
 
       - name: Upload parsed results artifact
-        uses: actions/upload-artifact@b4b15b8c7c6ac21ea08fcf65892d2ee8f75cf882
+        uses: actions/upload-artifact@ea165f8d65b6e75b540449e92b4886f43607fa02
         with:
           name: ${{ github.sha }}_tfhe_zk_pok
           path: ${{ env.RESULTS_FILENAME }}
@@ -132,7 +142,8 @@ jobs:
         with:
           repository: zama-ai/slab
           path: slab
-          token: ${{ secrets.FHE_ACTIONS_TOKEN }}
+          persist-credentials: 'false'
+          token: ${{ secrets.REPO_CHECKOUT_TOKEN }}
 
       - name: Send data to Slab
         shell: bash
@@ -150,13 +161,13 @@ jobs:
 
   teardown-instance:
     name: Teardown instance (tfhe-zk-pok-benchmarks)
-    if: ${{ always() && needs.setup-instance.result != 'skipped' }}
+    if: ${{ always() && needs.setup-instance.result == 'success' }}
     needs: [ setup-instance, tfhe-zk-pok-benchmarks ]
     runs-on: ubuntu-latest
     steps:
       - name: Stop instance
         id: stop-instance
-        uses: zama-ai/slab-github-runner@801df0b8db5ea2b06128b7476c652f5ed5f193a8
+        uses: zama-ai/slab-github-runner@79939325c3c429837c10d6041e4fd8589d328bac
         with:
           mode: stop
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}

.github/workflows/benchmark_wasm_client.yml

@@ -28,7 +28,7 @@ jobs:
       (github.event_name == 'schedule' && github.repository == 'zama-ai/tfhe-rs') ||
       (github.event_name == 'push' && github.repository == 'zama-ai/tfhe-rs')
     permissions:
-      pull-requests: write
+      pull-requests: read
     outputs:
       wasm_bench: ${{ steps.changed-files.outputs.wasm_bench_any_changed }}
     steps:
@@ -36,12 +36,13 @@ jobs:
         uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
         with:
           fetch-depth: 0
+          persist-credentials: 'false'
+          token: ${{ secrets.REPO_CHECKOUT_TOKEN }}
 
       - name: Check for file changes
         id: changed-files
-        uses: tj-actions/changed-files@4edd678ac3f81e2dc578756871e4d00c19191daf
+        uses: tj-actions/changed-files@26a38635fc1173cc5820336ce97be6188d0de9f5 # v46.0.2
         with:
-          since_last_remote_commit: true
           files_yaml: |
             wasm_bench:
               - tfhe/Cargo.toml
@@ -64,7 +65,7 @@ jobs:
     steps:
       - name: Start instance
         id: start-instance
-        uses: zama-ai/slab-github-runner@801df0b8db5ea2b06128b7476c652f5ed5f193a8
+        uses: zama-ai/slab-github-runner@79939325c3c429837c10d6041e4fd8589d328bac
         with:
           mode: start
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}
@@ -87,7 +88,8 @@ jobs:
         uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
         with:
           fetch-depth: 0
-          token: ${{ secrets.FHE_ACTIONS_TOKEN }}
+          persist-credentials: 'false'
+          token: ${{ secrets.REPO_CHECKOUT_TOKEN }}
 
       - name: Get benchmark details
         run: |
@@ -98,7 +100,7 @@ jobs:
           } >> "${GITHUB_ENV}"
 
       - name: Install rust
-        uses: dtolnay/rust-toolchain@7b1c307e0dcbda6122208f10795a713336a9b35a
+        uses: dtolnay/rust-toolchain@888c2e1ea69ab0d4330cbf0af1ecc7b68f368cc1
         with:
           toolchain: nightly
 
@@ -108,7 +110,7 @@ jobs:
 
       - name: Node cache restoration
         id: node-cache
-        uses: actions/cache/restore@6849a6489940f00c2f30c0fb92c6274307ccb58a #v4.1.2
+        uses: actions/cache/restore@5a3ec84eff668545956fd18022155c47e93e2684 #v4.2.3
         with:
           path: |
             ~/.nvm
@@ -121,7 +123,7 @@ jobs:
           make install_node
 
       - name: Node cache save
-        uses: actions/cache/save@6849a6489940f00c2f30c0fb92c6274307ccb58a #v4.1.2
+        uses: actions/cache/save@5a3ec84eff668545956fd18022155c47e93e2684 #v4.2.3
         if: steps.node-cache.outputs.cache-hit != 'true'
         with:
           path: |
@@ -165,7 +167,7 @@ jobs:
           --append-results
 
       - name: Upload parsed results artifact
-        uses: actions/upload-artifact@b4b15b8c7c6ac21ea08fcf65892d2ee8f75cf882
+        uses: actions/upload-artifact@ea165f8d65b6e75b540449e92b4886f43607fa02
         with:
           name: ${{ github.sha }}_wasm_${{ matrix.browser }}
           path: ${{ env.RESULTS_FILENAME }}
@@ -175,7 +177,8 @@ jobs:
         with:
           repository: zama-ai/slab
           path: slab
-          token: ${{ secrets.FHE_ACTIONS_TOKEN }}
+          persist-credentials: 'false'
+          token: ${{ secrets.REPO_CHECKOUT_TOKEN }}
 
       - name: Send data to Slab
         shell: bash
@@ -193,13 +196,13 @@ jobs:
 
   teardown-instance:
     name: Teardown instance (wasm-client-benchmarks)
-    if: ${{ always() && needs.setup-instance.result != 'skipped' }}
+    if: ${{ always() && needs.setup-instance.result == 'success' }}
     needs: [ setup-instance, wasm-client-benchmarks ]
     runs-on: ubuntu-latest
     steps:
       - name: Stop instance
         id: stop-instance
-        uses: zama-ai/slab-github-runner@801df0b8db5ea2b06128b7476c652f5ed5f193a8
+        uses: zama-ai/slab-github-runner@79939325c3c429837c10d6041e4fd8589d328bac
         with:
           mode: stop
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}

.github/workflows/benchmark_zk_pke.yml

@@ -4,10 +4,14 @@ name: PKE ZK benchmarks
 on:
   workflow_dispatch:
     inputs:
-      run_throughput:
-        description: "Run throughput benchmarks"
-        type: boolean
-        default: false
+      bench_type:
+        description: "Benchmarks type"
+        type: choice
+        default: latency
+        options:
+          - latency
+          - throughput
+          - both
 
   push:
     branches:
@@ -26,7 +30,6 @@ env:
   SLACK_ICON: https://pbs.twimg.com/profile_images/1274014582265298945/OjBKP9kn_400x400.png
   SLACK_USERNAME: ${{ secrets.BOT_USERNAME }}
   SLACK_WEBHOOK: ${{ secrets.SLACK_WEBHOOK }}
-  BENCH_TYPE: latency
 
 jobs:
   should-run:
@@ -40,12 +43,13 @@ jobs:
         uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
         with:
           fetch-depth: 0
+          persist-credentials: 'false'
+          token: ${{ secrets.REPO_CHECKOUT_TOKEN }}
 
       - name: Check for file changes
         id: changed-files
-        uses: tj-actions/changed-files@4edd678ac3f81e2dc578756871e4d00c19191daf
+        uses: tj-actions/changed-files@26a38635fc1173cc5820336ce97be6188d0de9f5 # v46.0.2
         with:
-          since_last_remote_commit: true
           files_yaml: |
             zk_pok:
               - tfhe/Cargo.toml
@@ -59,10 +63,37 @@ jobs:
               - tfhe/benches/integer/zk_pke.rs
               - .github/workflows/zk_pke_benchmark.yml
 
+  prepare-matrix:
+    name: Prepare operations matrix
+    runs-on: ubuntu-latest
+    if: github.event_name != 'schedule' ||
+      (github.event_name == 'schedule' && github.repository == 'zama-ai/tfhe-rs')
+    outputs:
+      bench_type: ${{ steps.set_bench_type.outputs.bench_type }}
+    steps:
+      - name: Set benchmark types
+        if: github.event_name == 'workflow_dispatch'
+        run: |
+          if [[ "${{ inputs.bench_type }}" == "both" ]]; then
+            echo "BENCH_TYPE=[\"latency\", \"throughput\"]" >> "${GITHUB_ENV}"
+          else
+            echo "BENCH_TYPE=[\"${{ inputs.bench_type }}\"]" >> "${GITHUB_ENV}"
+          fi
+
+      - name: Default benchmark type
+        if: github.event_name != 'workflow_dispatch'
+        run: |
+          echo "BENCH_TYPE=[\"latency\"]" >> "${GITHUB_ENV}"
+
+      - name: Set benchmark types output
+        id: set_bench_type
+        run: |
+          echo "bench_type=${{ toJSON(env.BENCH_TYPE) }}" >> "${GITHUB_OUTPUT}"
+
   setup-instance:
     name: Setup instance (pke-zk-benchmarks)
     runs-on: ubuntu-latest
-    needs: should-run
+    needs: [ should-run, prepare-matrix ]
     if: github.event_name == 'workflow_dispatch' ||
       (github.event_name == 'schedule' && github.repository == 'zama-ai/tfhe-rs') ||
       (github.event_name == 'push' &&
@@ -73,7 +104,7 @@ jobs:
     steps:
       - name: Start instance
         id: start-instance
-        uses: zama-ai/slab-github-runner@801df0b8db5ea2b06128b7476c652f5ed5f193a8
+        uses: zama-ai/slab-github-runner@79939325c3c429837c10d6041e4fd8589d328bac
         with:
           mode: start
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}
@@ -85,17 +116,22 @@ jobs:
   pke-zk-benchmarks:
     name: Execute PKE ZK benchmarks
     if: needs.setup-instance.result != 'skipped'
-    needs: setup-instance
+    needs: [ prepare-matrix, setup-instance ]
     concurrency:
-      group: ${{ github.workflow }}_${{github.event_name}}_${{ github.ref }}${{ github.ref == 'refs/heads/main' && github.sha || '' }}
+      group: ${{ github.workflow_ref }}_${{github.event_name}}${{ github.ref == 'refs/heads/main' && github.sha || '' }}
       cancel-in-progress: ${{ github.ref != 'refs/heads/main' }}
     runs-on: ${{ needs.setup-instance.outputs.runner-name }}
+    strategy:
+      max-parallel: 1
+      matrix:
+        bench_type: ${{ fromJSON(needs.prepare-matrix.outputs.bench_type) }}
     steps:
       - name: Checkout tfhe-rs repo with tags
         uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
         with:
           fetch-depth: 0
-          token: ${{ secrets.FHE_ACTIONS_TOKEN }}
+          persist-credentials: 'false'
+          token: ${{ secrets.REPO_CHECKOUT_TOKEN }}
 
       - name: Get benchmark details
         run: |
@@ -106,7 +142,7 @@ jobs:
           } >> "${GITHUB_ENV}"
 
       - name: Install rust
-        uses: dtolnay/rust-toolchain@7b1c307e0dcbda6122208f10795a713336a9b35a
+        uses: dtolnay/rust-toolchain@888c2e1ea69ab0d4330cbf0af1ecc7b68f368cc1
         with:
           toolchain: nightly
 
@@ -115,16 +151,12 @@ jobs:
         with:
           repository: zama-ai/slab
           path: slab
-          token: ${{ secrets.FHE_ACTIONS_TOKEN }}
-
-      - name: Should run throughput benchmarks
-        if: inputs.run_throughput
-        run: |
-          echo "BENCH_TYPE=throughput" >> "${GITHUB_ENV}"
+          persist-credentials: 'false'
+          token: ${{ secrets.REPO_CHECKOUT_TOKEN }}
 
       - name: Run benchmarks with AVX512
         run: |
-          make bench_integer_zk
+          make BENCH_TYPE=${{ matrix.bench_type }} bench_integer_zk
 
       - name: Parse results
         run: |
@@ -138,7 +170,7 @@ jobs:
           --bench-date "${{ env.BENCH_DATE }}" \
           --walk-subdirs \
           --name-suffix avx512 \
-          --bench-type ${{ env.BENCH_TYPE }}
+          --bench-type ${{ matrix.bench_type }}
 
       - name: Parse CRS sizes results
         run: |
@@ -147,7 +179,7 @@ jobs:
           --append-results
 
       - name: Upload parsed results artifact
-        uses: actions/upload-artifact@b4b15b8c7c6ac21ea08fcf65892d2ee8f75cf882
+        uses: actions/upload-artifact@ea165f8d65b6e75b540449e92b4886f43607fa02
         with:
           name: ${{ github.sha }}_integer_zk
           path: ${{ env.RESULTS_FILENAME }}
@@ -157,7 +189,8 @@ jobs:
         with:
           repository: zama-ai/slab
           path: slab
-          token: ${{ secrets.FHE_ACTIONS_TOKEN }}
+          persist-credentials: 'false'
+          token: ${{ secrets.REPO_CHECKOUT_TOKEN }}
 
       - name: Send data to Slab
         shell: bash
@@ -175,13 +208,13 @@ jobs:
 
   teardown-instance:
     name: Teardown instance (pke-zk-benchmarks)
-    if: ${{ always() && needs.setup-instance.result != 'skipped' }}
+    if: ${{ always() && needs.setup-instance.result == 'success' }}
     needs: [ setup-instance, pke-zk-benchmarks ]
     runs-on: ubuntu-latest
     steps:
       - name: Stop instance
         id: stop-instance
-        uses: zama-ai/slab-github-runner@801df0b8db5ea2b06128b7476c652f5ed5f193a8
+        uses: zama-ai/slab-github-runner@79939325c3c429837c10d6041e4fd8589d328bac
         with:
           mode: stop
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}

.github/workflows/cargo_build.yml

@@ -28,7 +28,7 @@ jobs:
       - uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
 
       - name: Install latest stable
-        uses: dtolnay/rust-toolchain@7b1c307e0dcbda6122208f10795a713336a9b35a
+        uses: dtolnay/rust-toolchain@888c2e1ea69ab0d4330cbf0af1ecc7b68f368cc1
         with:
           toolchain: stable
 

.github/workflows/cargo_build_tfhe_fft.yml

@@ -21,7 +21,7 @@ jobs:
       fail-fast: false
 
     steps:
-      - uses: actions/checkout@ac593985615ec2ede58e132d2e21d2b1cbd6127c
+      - uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
 
       - name: Install Rust
         uses: actions-rs/toolchain@16499b5e05bf2e26879000db0c1d13f7e13fa3af

.github/workflows/cargo_build_tfhe_ntt.yml

@@ -12,14 +12,14 @@ concurrency:
   cancel-in-progress: true
 
 jobs:
-  cargo-builds:
+  cargo-builds-ntt:
     runs-on: ${{ matrix.os }}
     strategy:
       matrix:
         os: [ubuntu-latest, macos-latest, windows-latest]
       fail-fast: false
     steps:
-      - uses: actions/checkout@ac593985615ec2ede58e132d2e21d2b1cbd6127c
+      - uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
 
       - name: Install Rust
         uses: actions-rs/toolchain@16499b5e05bf2e26879000db0c1d13f7e13fa3af

.github/workflows/cargo_test_fft.yml

@@ -3,23 +3,53 @@ name: Cargo Test tfhe-fft
 
 on:
   pull_request:
+  push:
+    branches:
+      - main
 
 env:
   CARGO_TERM_COLOR: always
+  IS_PULL_REQUEST: ${{ github.event_name == 'pull_request' }}
 
 concurrency:
   group: ${{ github.workflow }}-${{ github.head_ref }}
   cancel-in-progress: true
 
 jobs:
-  cargo-tests:
+  should-run:
+    runs-on: ubuntu-latest
+    permissions:
+      pull-requests: read
+    outputs:
+      fft_test: ${{ env.IS_PULL_REQUEST == 'false' || steps.changed-files.outputs.fft_any_changed }}
+    steps:
+      - name: Checkout tfhe-rs
+        uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
+        with:
+          fetch-depth: 0
+          persist-credentials: 'false'
+
+      - name: Check for file changes
+        id: changed-files
+        uses: tj-actions/changed-files@26a38635fc1173cc5820336ce97be6188d0de9f5 # v46.0.2
+        with:
+          files_yaml: |
+            fft:
+              - tfhe/Cargo.toml
+              - Makefile
+              - tfhe-fft/**
+              - '.github/workflows/cargo_test_fft.yml'
+
+  cargo-tests-fft:
+    needs: should-run
+    if: needs.should-run.outputs.fft_test == 'true'
     runs-on: ${{ matrix.runner_type }}
     strategy:
       matrix:
         runner_type: [ubuntu-latest, macos-latest, windows-latest]
       fail-fast: false
     steps:
-      - uses: actions/checkout@ac593985615ec2ede58e132d2e21d2b1cbd6127c
+      - uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
 
       - name: Install Rust
         uses: actions-rs/toolchain@16499b5e05bf2e26879000db0c1d13f7e13fa3af
@@ -38,13 +68,15 @@ jobs:
         run: |
           make test_fft_no_std
 
-  cargo-tests-nightly:
+  cargo-tests-fft-nightly:
+    needs: should-run
+    if: needs.should-run.outputs.fft_test == 'true'
     runs-on: ${{ matrix.runner_type }}
     strategy:
       matrix:
         runner_type: [ubuntu-latest, macos-latest, windows-latest]
     steps:
-      - uses: actions/checkout@ac593985615ec2ede58e132d2e21d2b1cbd6127c
+      - uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
 
       - name: Install Rust
         uses: actions-rs/toolchain@16499b5e05bf2e26879000db0c1d13f7e13fa3af
@@ -60,12 +92,41 @@ jobs:
         run: |
           make test_fft_no_std_nightly
 
-  cargo-tests-node-js:
-    runs-on: "ubuntu-latest"
+  cargo-tests-fft-node-js:
+    needs: should-run
+    if: needs.should-run.outputs.fft_test == 'true'
+    runs-on: ubuntu-latest
     steps:
-      - uses: actions/checkout@ac593985615ec2ede58e132d2e21d2b1cbd6127c
+      - uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
 
       - name: Test node js
         run: |
           make install_node
           make test_fft_node_js_ci
+
+  cargo-tests-fft-successful:
+    needs: [ should-run, cargo-tests-fft, cargo-tests-fft-nightly, cargo-tests-fft-node-js ]
+    if: ${{ always() }}
+    runs-on: ubuntu-latest
+    steps:
+      - name: Tests do not need to run
+        if: needs.should-run.outputs.fft_test == 'false'
+        run: |
+          echo "tfhe-fft files haven't changed tests don't need to run"
+
+      - name: Check all tests passed
+        if: needs.should-run.outputs.fft_test == 'true' &&
+          needs.cargo-tests-fft.result == 'success' &&
+          needs.cargo-tests-fft-nightly.result == 'success' &&
+          needs.cargo-tests-fft-node-js.result == 'success'
+        run: |
+          echo "All tfhe-fft test passed"
+
+      - name: Check tests failure
+        if: needs.should-run.outputs.fft_test == 'true' &&
+          (needs.cargo-tests-fft.result != 'success' ||
+          needs.cargo-tests-fft-nightly.result != 'success' ||
+          needs.cargo-tests-fft-node-js.result != 'success')
+        run: |
+          echo "Some tfhe-fft tests failed"
+          exit 1

.github/workflows/cargo_test_ntt.yml

@@ -3,23 +3,53 @@ name: Cargo Test tfhe-ntt
 
 on:
   pull_request:
+  push:
+    branches:
+      - main
 
 env:
   CARGO_TERM_COLOR: always
+  IS_PULL_REQUEST: ${{ github.event_name == 'pull_request' }}
 
 concurrency:
   group: ${{ github.workflow }}-${{ github.head_ref }}
   cancel-in-progress: true
 
-jobs:
-  cargo-tests:
+jobs:  
+  should-run:
+    runs-on: ubuntu-latest
+    permissions:
+      pull-requests: read
+    outputs:
+      ntt_test: ${{ env.IS_PULL_REQUEST == 'false' || steps.changed-files.outputs.ntt_any_changed }}
+    steps:
+      - name: Checkout tfhe-rs
+        uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
+        with:
+          fetch-depth: 0
+          persist-credentials: 'false'
+
+      - name: Check for file changes
+        id: changed-files
+        uses: tj-actions/changed-files@26a38635fc1173cc5820336ce97be6188d0de9f5 # v46.0.2
+        with:
+          files_yaml: |
+            ntt:
+              - tfhe/Cargo.toml
+              - Makefile
+              - tfhe-ntt/**
+              - '.github/workflows/cargo_test_ntt.yml'
+
+  cargo-tests-ntt:
+    needs: should-run
+    if: needs.should-run.outputs.ntt_test == 'true'
     runs-on: ${{ matrix.os }}
     strategy:
       matrix:
         os: [ubuntu-latest, macos-latest, windows-latest]
       fail-fast: false
     steps:
-      - uses: actions/checkout@ac593985615ec2ede58e132d2e21d2b1cbd6127c
+      - uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
 
       - name: Install Rust
         uses: actions-rs/toolchain@16499b5e05bf2e26879000db0c1d13f7e13fa3af
@@ -33,13 +63,15 @@ jobs:
       - name: Test no-std
         run: make test_ntt_no_std
 
-  cargo-tests-nightly:
+  cargo-tests-ntt-nightly:
+    needs: should-run
+    if: needs.should-run.outputs.ntt_test == 'true'
     runs-on: ${{ matrix.os }}
     strategy:
       matrix:
         os: [ubuntu-latest, macos-latest, windows-latest]
     steps:
-      - uses: actions/checkout@ac593985615ec2ede58e132d2e21d2b1cbd6127c
+      - uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
 
       - name: Install Rust
         uses: actions-rs/toolchain@16499b5e05bf2e26879000db0c1d13f7e13fa3af
@@ -52,3 +84,28 @@ jobs:
 
       - name: Test no-std nightly
         run: make test_ntt_no_std_nightly
+
+  cargo-tests-ntt-successful:
+    needs: [ should-run, cargo-tests-ntt, cargo-tests-ntt-nightly ]
+    if: ${{ always() }}
+    runs-on: ubuntu-latest
+    steps:
+      - name: Tests do not need to run
+        if: needs.should-run.outputs.ntt_test == 'false'
+        run: |
+          echo "tfhe-ntt files haven't changed tests don't need to run"
+
+      - name: Check all tests success
+        if: needs.should-run.outputs.ntt_test == 'true' &&
+          needs.cargo-tests-ntt.result == 'success' &&
+          needs.cargo-tests-ntt-nightly.result == 'success'
+        run: |
+          echo "All tfhe-ntt tests passed"
+
+      - name: Check tests failure
+        if: needs.should-run.outputs.ntt_test == 'true' &&
+          (needs.cargo-tests-ntt.result != 'success' ||
+          needs.cargo-tests-ntt-nightly.result != 'success')
+        run: |
+          echo "Some tfhe-ntt tests failed"
+          exit 1

.github/workflows/check_commit.yml

@@ -2,6 +2,7 @@
 name: Check commit and PR compliance
 on:
   pull_request:
+
 jobs:
   check-commit-pr:
     name: Check commit and PR

.github/workflows/ci_lint.yml

@@ -6,6 +6,7 @@ on:
 
 env:
   ACTIONLINT_VERSION: 1.6.27
+  CHECKOUT_TOKEN: ${{ secrets.REPO_CHECKOUT_TOKEN || secrets.GITHUB_TOKEN }}
 
 jobs:
   lint-check:
@@ -14,6 +15,9 @@ jobs:
     steps:
       - name: Checkout tfhe-rs
         uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
+        with:
+          persist-credentials: 'false'
+          token: ${{ env.CHECKOUT_TOKEN }}
 
       - name: Get actionlint
         run: |
@@ -27,7 +31,8 @@ jobs:
           make lint_workflow
 
       - name: Ensure SHA pinned actions
-        uses: zgosalvez/github-actions-ensure-sha-pinned-actions@38608ef4fb69adae7f1eac6eeb88e67b7d083bfd # v3.0.16
+        uses: zgosalvez/github-actions-ensure-sha-pinned-actions@25ed13d0628a1601b4b44048e63cc4328ed03633 # v3.0.22
         with:
           allowlist: |
             slsa-framework/slsa-github-generator
+            ./

.github/workflows/code_coverage.yml

@@ -25,7 +25,7 @@ jobs:
     steps:
       - name: Start instance
         id: start-instance
-        uses: zama-ai/slab-github-runner@801df0b8db5ea2b06128b7476c652f5ed5f193a8
+        uses: zama-ai/slab-github-runner@79939325c3c429837c10d6041e4fd8589d328bac
         with:
           mode: start
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}
@@ -38,7 +38,7 @@ jobs:
     name: Code coverage tests
     needs: setup-instance
     concurrency:
-      group: ${{ github.workflow }}_${{ github.event_name }}_${{ github.ref }}
+      group: ${{ github.workflow_ref }}_${{ github.event_name }}
       cancel-in-progress: true
     runs-on: ${{ needs.setup-instance.outputs.runner-name }}
     timeout-minutes: 5760 # 4 days
@@ -47,13 +47,13 @@ jobs:
         uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
 
       - name: Install latest stable
-        uses: dtolnay/rust-toolchain@7b1c307e0dcbda6122208f10795a713336a9b35a
+        uses: dtolnay/rust-toolchain@888c2e1ea69ab0d4330cbf0af1ecc7b68f368cc1
         with:
           toolchain: stable
 
       - name: Check for file changes
         id: changed-files
-        uses: tj-actions/changed-files@4edd678ac3f81e2dc578756871e4d00c19191daf
+        uses: tj-actions/changed-files@26a38635fc1173cc5820336ce97be6188d0de9f5 # v46.0.2
         with:
           files_yaml: |
             tfhe:
@@ -83,7 +83,7 @@ jobs:
           make test_shortint_cov
 
       - name: Upload tfhe coverage to Codecov
-        uses: codecov/codecov-action@5c47607acb93fed5485fdbf7232e8a31425f672a
+        uses: codecov/codecov-action@0565863a31f2c772f9f0395002a31e3f06189574
         if: steps.changed-files.outputs.tfhe_any_changed == 'true'
         with:
           token: ${{ secrets.CODECOV_TOKEN }}
@@ -97,7 +97,7 @@ jobs:
           make test_integer_cov
 
       - name: Upload tfhe coverage to Codecov
-        uses: codecov/codecov-action@5c47607acb93fed5485fdbf7232e8a31425f672a
+        uses: codecov/codecov-action@0565863a31f2c772f9f0395002a31e3f06189574
         if: steps.changed-files.outputs.tfhe_any_changed == 'true'
         with:
           token: ${{ secrets.CODECOV_TOKEN }}
@@ -115,13 +115,13 @@ jobs:
 
   teardown-instance:
     name: Teardown instance (code-coverage)
-    if: ${{ always() && needs.setup-instance.result != 'skipped' }}
+    if: ${{ always() && needs.setup-instance.result == 'success' }}
     needs: [ setup-instance, code-coverage ]
     runs-on: ubuntu-latest
     steps:
       - name: Stop instance
         id: stop-instance
-        uses: zama-ai/slab-github-runner@801df0b8db5ea2b06128b7476c652f5ed5f193a8
+        uses: zama-ai/slab-github-runner@79939325c3c429837c10d6041e4fd8589d328bac
         with:
           mode: stop
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}

.github/workflows/csprng_randomness_tests.yml

@@ -10,6 +10,10 @@ env:
   SLACK_ICON: https://pbs.twimg.com/profile_images/1274014582265298945/OjBKP9kn_400x400.png
   SLACK_USERNAME: ${{ secrets.BOT_USERNAME }}
   SLACK_WEBHOOK: ${{ secrets.SLACK_WEBHOOK }}
+  CHECKOUT_TOKEN: ${{ secrets.REPO_CHECKOUT_TOKEN || secrets.GITHUB_TOKEN }}
+  # Secrets will be available only to zama-ai organization members
+  SECRETS_AVAILABLE: ${{ secrets.JOB_SECRET != '' }}
+  EXTERNAL_CONTRIBUTION_RUNNER: "large_ubuntu_16"
 
 on:
   # Allows you to run this workflow manually from the Actions tab as an alternative.
@@ -23,11 +27,12 @@ jobs:
     if: ${{ github.event_name == 'workflow_dispatch' || contains(github.event.label.name, 'approved') }}
     runs-on: ubuntu-latest
     outputs:
-      runner-name: ${{ steps.start-instance.outputs.label }}
+      runner-name: ${{ steps.start-remote-instance.outputs.label || steps.start-github-instance.outputs.runner_group }}
     steps:
-      - name: Start instance
-        id: start-instance
-        uses: zama-ai/slab-github-runner@801df0b8db5ea2b06128b7476c652f5ed5f193a8
+      - name: Start remote instance
+        id: start-remote-instance
+        if: env.SECRETS_AVAILABLE == 'true'
+        uses: zama-ai/slab-github-runner@79939325c3c429837c10d6041e4fd8589d328bac
         with:
           mode: start
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}
@@ -36,11 +41,18 @@ jobs:
           backend: aws
           profile: cpu-small
 
+      # This instance will be spawned especially for pull-request from forked repository
+      - name: Start GitHub instance
+        id: start-github-instance
+        if: env.SECRETS_AVAILABLE == 'false'
+        run: |
+          echo "runner_group=${{ env.EXTERNAL_CONTRIBUTION_RUNNER }}" >> "$GITHUB_OUTPUT"
+
   csprng-randomness-tests:
     name: CSPRNG randomness tests
     needs: setup-instance
     concurrency:
-      group: ${{ github.workflow }}_${{ github.ref }}
+      group: ${{ github.workflow_ref }}
       cancel-in-progress: true
     runs-on: ${{ needs.setup-instance.outputs.runner-name }}
     steps:
@@ -48,10 +60,10 @@ jobs:
         uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
         with:
           persist-credentials: 'false'
-          token: ${{ secrets.FHE_ACTIONS_TOKEN }}
+          token: ${{ env.CHECKOUT_TOKEN }}
 
       - name: Install latest stable
-        uses: dtolnay/rust-toolchain@7b1c307e0dcbda6122208f10795a713336a9b35a
+        uses: dtolnay/rust-toolchain@888c2e1ea69ab0d4330cbf0af1ecc7b68f368cc1
         with:
           toolchain: stable
 
@@ -69,13 +81,14 @@ jobs:
 
   teardown-instance:
     name: Teardown instance (csprng-randomness-tests)
-    if: ${{ always() && needs.setup-instance.result != 'skipped' }}
+    if: ${{ always() && needs.setup-instance.result == 'success' }}
     needs: [ setup-instance, csprng-randomness-tests ]
     runs-on: ubuntu-latest
     steps:
-      - name: Stop instance
+      - name: Stop remote instance
         id: stop-instance
-        uses: zama-ai/slab-github-runner@801df0b8db5ea2b06128b7476c652f5ed5f193a8
+        if: env.SECRETS_AVAILABLE == 'true'
+        uses: zama-ai/slab-github-runner@79939325c3c429837c10d6041e4fd8589d328bac
         with:
           mode: stop
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}

.github/workflows/data_pr_close.yml

@@ -8,7 +8,7 @@ env:
   SLACK_ICON: https://pbs.twimg.com/profile_images/1274014582265298945/OjBKP9kn_400x400.png
   SLACK_USERNAME: ${{ secrets.BOT_USERNAME }}
   SLACK_WEBHOOK: ${{ secrets.SLACK_WEBHOOK }}
-  PR_BRANCH: ${{ github.head_ref || github.ref_name }}
+  PR_BRANCH: ${{ github.ref_name }}
   CLOSE_TYPE: ${{ github.event.pull_request.merged && 'merge' || 'close' }}
 
 # only trigger on pull request closed events

.github/workflows/gpu_4090_tests.yml

@@ -1,5 +1,5 @@
 # Compile and test tfhe-cuda-backend on an RTX 4090 machine
-name: TFHE Cuda Backend - 4090 full tests
+name: Cuda - 4090 full tests
 
 env:
   CARGO_TERM_COLOR: always
@@ -11,6 +11,7 @@ env:
   SLACK_ICON: https://pbs.twimg.com/profile_images/1274014582265298945/OjBKP9kn_400x400.png
   SLACK_USERNAME: ${{ secrets.BOT_USERNAME }}
   SLACK_WEBHOOK: ${{ secrets.SLACK_WEBHOOK }}
+  CHECKOUT_TOKEN: ${{ secrets.REPO_CHECKOUT_TOKEN || secrets.GITHUB_TOKEN }}
 
 on:
   # Allows you to run this workflow manually from the Actions tab as an alternative.
@@ -28,19 +29,20 @@ jobs:
       contains(github.event.label.name, '4090_test') ||
       (github.event_name == 'schedule' &&  github.repository == 'zama-ai/tfhe-rs')
     concurrency:
-      group: ${{ github.workflow }}_${{ github.ref }}
+      group: ${{ github.workflow_ref }}
       cancel-in-progress: true
     runs-on: ["self-hosted", "4090-desktop"]
+    timeout-minutes: 1440 # 24 hours
 
     steps:
       - name: Checkout tfhe-rs
         uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
         with:
           persist-credentials: 'false'
-          token: ${{ secrets.FHE_ACTIONS_TOKEN }}
+          token: ${{ env.CHECKOUT_TOKEN }}
 
       - name: Install latest stable
-        uses: dtolnay/rust-toolchain@7b1c307e0dcbda6122208f10795a713336a9b35a
+        uses: dtolnay/rust-toolchain@888c2e1ea69ab0d4330cbf0af1ecc7b68f368cc1
         with:
           toolchain: stable
 

.github/workflows/gpu_fast_h100_tests.yml

@@ -1,5 +1,5 @@
 # Compile and test tfhe-cuda-backend on an H100 VM on hyperstack
-name: TFHE Cuda Backend - Fast tests on H100
+name: Cuda - Fast tests on H100
 
 env:
   CARGO_TERM_COLOR: always
@@ -11,19 +11,25 @@ env:
   SLACK_ICON: https://pbs.twimg.com/profile_images/1274014582265298945/OjBKP9kn_400x400.png
   SLACK_USERNAME: ${{ secrets.BOT_USERNAME }}
   SLACK_WEBHOOK: ${{ secrets.SLACK_WEBHOOK }}
+  SLACKIFY_MARKDOWN: true
   IS_PULL_REQUEST: ${{ github.event_name == 'pull_request' }}
+  PULL_REQUEST_MD_LINK: ""
+  CHECKOUT_TOKEN: ${{ secrets.REPO_CHECKOUT_TOKEN || secrets.GITHUB_TOKEN }}
+  # Secrets will be available only to zama-ai organization members
+  SECRETS_AVAILABLE: ${{ secrets.JOB_SECRET != '' }}
+  EXTERNAL_CONTRIBUTION_RUNNER: "gpu_ubuntu-22.04"
 
 on:
   # Allows you to run this workflow manually from the Actions tab as an alternative.
   workflow_dispatch:
   pull_request:
-      types: [ labeled ]
+    types: [ labeled ]
 
 jobs:
   should-run:
     runs-on: ubuntu-latest
     permissions:
-      pull-requests: write
+      pull-requests: read
     outputs:
       gpu_test: ${{ env.IS_PULL_REQUEST == 'false' || steps.changed-files.outputs.gpu_any_changed }}
     steps:
@@ -31,12 +37,13 @@ jobs:
         uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
         with:
           fetch-depth: 0
+          persist-credentials: 'false'
+          token: ${{ env.CHECKOUT_TOKEN }}
 
       - name: Check for file changes
         id: changed-files
-        uses: tj-actions/changed-files@4edd678ac3f81e2dc578756871e4d00c19191daf
+        uses: tj-actions/changed-files@26a38635fc1173cc5820336ce97be6188d0de9f5 # v46.0.2
         with:
-          since_last_remote_commit: true
           files_yaml: |
             gpu:
               - tfhe/Cargo.toml
@@ -50,7 +57,7 @@ jobs:
               - tfhe/src/shortint/parameters/**
               - tfhe/src/high_level_api/**
               - tfhe/src/c_api/**
-              - 'tfhe/docs/**.md'
+              - 'tfhe/docs/**/**.md'
               - '.github/workflows/gpu_fast_h100_tests.yml'
               - scripts/integer-tests.sh
               - ci/slab.toml
@@ -63,11 +70,18 @@ jobs:
       (github.event.action == 'labeled' && github.event.label.name == 'approved' && needs.should-run.outputs.gpu_test == 'true')
     runs-on: ubuntu-latest
     outputs:
-      runner-name: ${{ steps.start-instance.outputs.label }}
+      # Use permanent remote instance label first as on-demand remote instance label output is set before the end of start-remote-instance step.
+      # If the latter fails due to a failed GitHub action runner set up, we have to fallback on the permanent instance.
+      # Since the on-demand remote label is set before failure, we have to do the logical OR in this order,
+      # otherwise we'll try to run the next job on a non-existing on-demand instance.
+      runner-name: ${{ steps.use-permanent-instance.outputs.runner_group || steps.start-remote-instance.outputs.label || steps.start-github-instance.outputs.runner_group }}
+      remote-instance-outcome: ${{ steps.start-remote-instance.outcome }}
     steps:
-      - name: Start instance
-        id: start-instance
-        uses: zama-ai/slab-github-runner@801df0b8db5ea2b06128b7476c652f5ed5f193a8
+      - name: Start remote instance
+        id: start-remote-instance
+        if: env.SECRETS_AVAILABLE == 'true'
+        continue-on-error: true
+        uses: zama-ai/slab-github-runner@79939325c3c429837c10d6041e4fd8589d328bac
         with:
           mode: start
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}
@@ -76,13 +90,27 @@ jobs:
           backend: hyperstack
           profile: single-h100
 
+      # This will allow to fallback on permanent instances running on Hyperstack.
+      - name: Use permanent remote instance
+        id: use-permanent-instance
+        if: env.SECRETS_AVAILABLE == 'true' && steps.start-remote-instance.outcome == 'failure'
+        run: |
+          echo "runner_group=h100x1" >> "$GITHUB_OUTPUT"
+
+      # This instance will be spawned especially for pull-request from forked repository
+      - name: Start GitHub instance
+        id: start-github-instance
+        if: env.SECRETS_AVAILABLE == 'false'
+        run: |
+          echo "runner_group=${{ env.EXTERNAL_CONTRIBUTION_RUNNER }}" >> "$GITHUB_OUTPUT"
+
   cuda-tests-linux:
     name: CUDA H100 tests
     needs: [ should-run, setup-instance ]
     if: github.event_name != 'pull_request' ||
       (github.event_name == 'pull_request' && needs.setup-instance.result != 'skipped')
     concurrency:
-      group: ${{ github.workflow }}_${{ github.ref }}
+      group: ${{ github.workflow_ref }}
       cancel-in-progress: ${{ github.ref != 'refs/heads/main' }}
     runs-on: ${{ needs.setup-instance.outputs.runner-name }}
     strategy:
@@ -93,60 +121,26 @@ jobs:
           - os: ubuntu-22.04
             cuda: "12.2"
             gcc: 11 
-    env:
-      CUDA_PATH: /usr/local/cuda-${{ matrix.cuda }}
-      CMAKE_VERSION: 3.29.6
     steps:
-      # Mandatory on hyperstack since a bootable volume is not re-usable yet.
-      - name: Install dependencies
-        run: |
-          sudo apt update
-          sudo apt install -y checkinstall zlib1g-dev libssl-dev libclang-dev
-          wget https://github.com/Kitware/CMake/releases/download/v${{ env.CMAKE_VERSION }}/cmake-${{ env.CMAKE_VERSION }}.tar.gz
-          tar -zxvf cmake-${{ env.CMAKE_VERSION }}.tar.gz
-          cd cmake-${{ env.CMAKE_VERSION }}
-          ./bootstrap
-          make -j"$(nproc)"
-          sudo make install
-
       - name: Checkout tfhe-rs
         uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
         with:
           persist-credentials: 'false'
-          token: ${{ secrets.FHE_ACTIONS_TOKEN }}
+          token: ${{ env.CHECKOUT_TOKEN }}
 
-      - name: Set up home
-        run: |
-          echo "HOME=/home/ubuntu" >> "${GITHUB_ENV}"
+      - name: Setup Hyperstack dependencies
+        if: needs.setup-instance.outputs.remote-instance-outcome == 'success'
+        uses: ./.github/actions/gpu_setup
+        with:
+          cuda-version: ${{ matrix.cuda }}
+          gcc-version: ${{ matrix.gcc }}
+          github-instance: ${{ env.SECRETS_AVAILABLE == 'false' }}
 
       - name: Install latest stable
-        uses: dtolnay/rust-toolchain@7b1c307e0dcbda6122208f10795a713336a9b35a
+        uses: dtolnay/rust-toolchain@888c2e1ea69ab0d4330cbf0af1ecc7b68f368cc1
         with:
           toolchain: stable
 
-      - name: Export CUDA variables
-        if: ${{ !cancelled() }}
-        run: |
-          echo "CUDA_PATH=$CUDA_PATH" >> "${GITHUB_ENV}"
-          echo "$CUDA_PATH/bin" >> "${GITHUB_PATH}"
-          echo "LD_LIBRARY_PATH=$CUDA_PATH/lib:$LD_LIBRARY_PATH" >> "${GITHUB_ENV}"
-          echo "CUDACXX=/usr/local/cuda-${{ matrix.cuda }}/bin/nvcc" >> "${GITHUB_ENV}"
-
-      # Specify the correct host compilers
-      - name: Export gcc and g++ variables
-        if: ${{ !cancelled() }}
-        run: |
-          {
-            echo "CC=/usr/bin/gcc-${{ matrix.gcc }}";
-            echo "CXX=/usr/bin/g++-${{ matrix.gcc }}";
-            echo "CUDAHOSTCXX=/usr/bin/g++-${{ matrix.gcc }}";
-            echo "HOME=/home/ubuntu";
-          } >> "${GITHUB_ENV}"
-
-      - name: Check device is detected
-        if: ${{ !cancelled() }}
-        run: nvidia-smi
-
       - name: Run core crypto and internal CUDA backend tests
         run: |
           BIG_TESTS_INSTANCE=TRUE make test_core_crypto_gpu
@@ -172,21 +166,28 @@ jobs:
     if: ${{ always() && needs.cuda-tests-linux.result != 'skipped' && failure() }}
     continue-on-error: true
     steps:
+      - name: Set pull-request URL
+        if: env.SECRETS_AVAILABLE == 'true' && github.event_name == 'pull_request'
+        run: |
+          echo "PULL_REQUEST_MD_LINK=[pull-request](${{ vars.PR_BASE_URL }}${{ github.event.pull_request.number }}), "  >> "${GITHUB_ENV}"
+
       - name: Send message
+        if: env.SECRETS_AVAILABLE == 'true'
         uses: rtCamp/action-slack-notify@c33737706dea87cd7784c687dadc9adf1be59990
         env:
           SLACK_COLOR: ${{ needs.cuda-tests-linux.result }}
-          SLACK_MESSAGE: "Fast H100 tests finished with status: ${{ needs.cuda-tests-linux.result }}. (${{ env.ACTION_RUN_URL }})"
+          SLACK_MESSAGE: "Fast H100 tests finished with status: ${{ needs.cuda-tests-linux.result }}. (${{ env.PULL_REQUEST_MD_LINK }}[action run](${{ env.ACTION_RUN_URL }}))"
 
   teardown-instance:
     name: Teardown instance (cuda-h100-tests)
-    if: ${{ always() && needs.setup-instance.result != 'skipped' }}
+    if: ${{ always() && needs.setup-instance.outputs.remote-instance-outcome == 'success' }}
     needs: [ setup-instance, cuda-tests-linux ]
     runs-on: ubuntu-latest
     steps:
-      - name: Stop instance
+      - name: Stop remote instance
         id: stop-instance
-        uses: zama-ai/slab-github-runner@801df0b8db5ea2b06128b7476c652f5ed5f193a8
+        if: env.SECRETS_AVAILABLE == 'true'
+        uses: zama-ai/slab-github-runner@79939325c3c429837c10d6041e4fd8589d328bac
         with:
           mode: stop
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}

.github/workflows/gpu_fast_tests.yml

@@ -1,5 +1,5 @@
 # Compile and test tfhe-cuda-backend on an AWS instance
-name: TFHE Cuda Backend - Fast tests
+name: Cuda - Fast tests
 
 env:
   CARGO_TERM_COLOR: always
@@ -11,7 +11,13 @@ env:
   SLACK_ICON: https://pbs.twimg.com/profile_images/1274014582265298945/OjBKP9kn_400x400.png
   SLACK_USERNAME: ${{ secrets.BOT_USERNAME }}
   SLACK_WEBHOOK: ${{ secrets.SLACK_WEBHOOK }}
+  SLACKIFY_MARKDOWN: true
   IS_PULL_REQUEST: ${{ github.event_name == 'pull_request' }}
+  PULL_REQUEST_MD_LINK: ""
+  CHECKOUT_TOKEN: ${{ secrets.REPO_CHECKOUT_TOKEN || secrets.GITHUB_TOKEN }}
+  # Secrets will be available only to zama-ai organization members
+  SECRETS_AVAILABLE: ${{ secrets.JOB_SECRET != '' }}
+  EXTERNAL_CONTRIBUTION_RUNNER: "gpu_ubuntu-22.04"
 
 on:
   # Allows you to run this workflow manually from the Actions tab as an alternative.
@@ -22,7 +28,7 @@ jobs:
   should-run:
     runs-on: ubuntu-latest
     permissions:
-      pull-requests: write
+      pull-requests: read
     outputs:
       gpu_test: ${{ env.IS_PULL_REQUEST == 'false' || steps.changed-files.outputs.gpu_any_changed }}
     steps:
@@ -30,12 +36,13 @@ jobs:
         uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
         with:
           fetch-depth: 0
+          persist-credentials: 'false'
+          token: ${{ env.CHECKOUT_TOKEN }}
 
       - name: Check for file changes
         id: changed-files
-        uses: tj-actions/changed-files@4edd678ac3f81e2dc578756871e4d00c19191daf
+        uses: tj-actions/changed-files@26a38635fc1173cc5820336ce97be6188d0de9f5 # v46.0.2
         with:
-          since_last_remote_commit: true
           files_yaml: |
             gpu:
               - tfhe/Cargo.toml
@@ -49,7 +56,7 @@ jobs:
               - tfhe/src/shortint/parameters/**
               - tfhe/src/high_level_api/**
               - tfhe/src/c_api/**
-              - 'tfhe/docs/**.md'
+              - 'tfhe/docs/**/**.md'
               - '.github/workflows/gpu_fast_tests.yml'
               - scripts/integer-tests.sh
               - ci/slab.toml
@@ -57,15 +64,16 @@ jobs:
   setup-instance:
     name: Setup instance (cuda-tests)
     needs: should-run
-    if: github.event_name != 'pull_request' ||
+    if: github.event_name == 'workflow_dispatch' ||
       needs.should-run.outputs.gpu_test == 'true'
     runs-on: ubuntu-latest
     outputs:
-      runner-name: ${{ steps.start-instance.outputs.label }}
+      runner-name: ${{ steps.start-remote-instance.outputs.label || steps.start-github-instance.outputs.runner_group }}
     steps:
-      - name: Start instance
-        id: start-instance
-        uses: zama-ai/slab-github-runner@801df0b8db5ea2b06128b7476c652f5ed5f193a8
+      - name: Start remote instance
+        id: start-remote-instance
+        if: env.SECRETS_AVAILABLE == 'true'
+        uses: zama-ai/slab-github-runner@79939325c3c429837c10d6041e4fd8589d328bac
         with:
           mode: start
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}
@@ -74,13 +82,20 @@ jobs:
           backend: hyperstack
           profile: gpu-test
 
+      # This instance will be spawned especially for pull-request from forked repository
+      - name: Start GitHub instance
+        id: start-github-instance
+        if: env.SECRETS_AVAILABLE == 'false'
+        run: |
+          echo "runner_group=${{ env.EXTERNAL_CONTRIBUTION_RUNNER }}" >> "$GITHUB_OUTPUT"
+
   cuda-tests-linux:
     name: CUDA tests
     needs: [ should-run, setup-instance ]
     if: github.event_name != 'pull_request' ||
       (github.event_name == 'pull_request' && needs.setup-instance.result != 'skipped')
     concurrency:
-      group: ${{ github.workflow }}_${{ github.ref }}
+      group: ${{ github.workflow_ref }}
       cancel-in-progress: ${{ github.ref != 'refs/heads/main' }}
     runs-on: ${{ needs.setup-instance.outputs.runner-name }}
     strategy:
@@ -91,60 +106,25 @@ jobs:
           - os: ubuntu-22.04
             cuda: "12.2"
             gcc: 11 
-    env:
-      CUDA_PATH: /usr/local/cuda-${{ matrix.cuda }}
-      CMAKE_VERSION: 3.29.6
     steps:
-      # Mandatory on hyperstack since a bootable volume is not re-usable yet.
-      - name: Install dependencies
-        run: |
-          sudo apt update
-          sudo apt install -y checkinstall zlib1g-dev libssl-dev libclang-dev
-          wget https://github.com/Kitware/CMake/releases/download/v${{ env.CMAKE_VERSION }}/cmake-${{ env.CMAKE_VERSION }}.tar.gz
-          tar -zxvf cmake-${{ env.CMAKE_VERSION }}.tar.gz
-          cd cmake-${{ env.CMAKE_VERSION }}
-          ./bootstrap
-          make -j"$(nproc)"
-          sudo make install
-
       - name: Checkout tfhe-rs
         uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
         with:
           persist-credentials: 'false'
-          token: ${{ secrets.FHE_ACTIONS_TOKEN }}
+          token: ${{ env.CHECKOUT_TOKEN }}
 
-      - name: Set up home
-        run: |
-          echo "HOME=/home/ubuntu" >> "${GITHUB_ENV}"
+      - name: Setup Hyperstack dependencies
+        uses: ./.github/actions/gpu_setup
+        with:
+          cuda-version: ${{ matrix.cuda }}
+          gcc-version: ${{ matrix.gcc }}
+          github-instance: ${{ env.SECRETS_AVAILABLE == 'false' }}
 
       - name: Install latest stable
-        uses: dtolnay/rust-toolchain@7b1c307e0dcbda6122208f10795a713336a9b35a
+        uses: dtolnay/rust-toolchain@888c2e1ea69ab0d4330cbf0af1ecc7b68f368cc1
         with:
           toolchain: stable
 
-      - name: Export CUDA variables
-        if: ${{ !cancelled() }}
-        run: |
-          echo "CUDA_PATH=$CUDA_PATH" >> "${GITHUB_ENV}"
-          echo "$CUDA_PATH/bin" >> "${GITHUB_PATH}"
-          echo "LD_LIBRARY_PATH=$CUDA_PATH/lib:$LD_LIBRARY_PATH" >> "${GITHUB_ENV}"
-          echo "CUDACXX=/usr/local/cuda-${{ matrix.cuda }}/bin/nvcc" >> "${GITHUB_ENV}"
-
-      # Specify the correct host compilers
-      - name: Export gcc and g++ variables
-        if: ${{ !cancelled() }}
-        run: |
-          {
-            echo "CC=/usr/bin/gcc-${{ matrix.gcc }}";
-            echo "CXX=/usr/bin/g++-${{ matrix.gcc }}";
-            echo "CUDAHOSTCXX=/usr/bin/g++-${{ matrix.gcc }}";
-            echo "HOME=/home/ubuntu";
-          } >> "${GITHUB_ENV}"
-
-      - name: Check device is detected
-        if: ${{ !cancelled() }}
-        run: nvidia-smi
-
       - name: Run core crypto and internal CUDA backend tests
         run: |
           make test_core_crypto_gpu
@@ -170,21 +150,28 @@ jobs:
     if: ${{ always() && needs.cuda-tests-linux.result != 'skipped' && failure() }}
     continue-on-error: true
     steps:
+      - name: Set pull-request URL
+        if: env.SECRETS_AVAILABLE == 'true' && github.event_name == 'pull_request'
+        run: |
+          echo "PULL_REQUEST_MD_LINK=[pull-request](${{ vars.PR_BASE_URL }}${{ github.event.pull_request.number }}), "  >> "${GITHUB_ENV}"
+
       - name: Send message
+        if: env.SECRETS_AVAILABLE == 'true'
         uses: rtCamp/action-slack-notify@c33737706dea87cd7784c687dadc9adf1be59990
         env:
           SLACK_COLOR: ${{ needs.cuda-tests-linux.result }}
-          SLACK_MESSAGE: "Base GPU tests finished with status: ${{ needs.cuda-tests-linux.result }}. (${{ env.ACTION_RUN_URL }})"
+          SLACK_MESSAGE: "Base GPU tests finished with status: ${{ needs.cuda-tests-linux.result }}. (${{ env.PULL_REQUEST_MD_LINK }}[action run](${{ env.ACTION_RUN_URL }}))"
 
   teardown-instance:
     name: Teardown instance (cuda-tests)
-    if: ${{ always() && needs.setup-instance.result != 'skipped' }}
+    if: ${{ always() && needs.setup-instance.result == 'success' }}
     needs: [ setup-instance, cuda-tests-linux ]
     runs-on: ubuntu-latest
     steps:
-      - name: Stop instance
+      - name: Stop remote instance
         id: stop-instance
-        uses: zama-ai/slab-github-runner@801df0b8db5ea2b06128b7476c652f5ed5f193a8
+        if: env.SECRETS_AVAILABLE == 'true'
+        uses: zama-ai/slab-github-runner@79939325c3c429837c10d6041e4fd8589d328bac
         with:
           mode: stop
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}

.github/workflows/gpu_full_h100_tests.yml

@@ -1,5 +1,5 @@
 # Compile and test tfhe-cuda-backend on an H100 VM on hyperstack
-name: TFHE Cuda Backend - Full tests on H100
+name: Cuda - Full tests on H100
 
 env:
   CARGO_TERM_COLOR: always
@@ -11,7 +11,6 @@ env:
   SLACK_ICON: https://pbs.twimg.com/profile_images/1274014582265298945/OjBKP9kn_400x400.png
   SLACK_USERNAME: ${{ secrets.BOT_USERNAME }}
   SLACK_WEBHOOK: ${{ secrets.SLACK_WEBHOOK }}
-  IS_PULL_REQUEST: ${{ github.event_name == 'pull_request' }}
 
 on:
   workflow_dispatch:
@@ -21,11 +20,17 @@ jobs:
     name: Setup instance (cuda-h100-tests)
     runs-on: ubuntu-latest
     outputs:
-      runner-name: ${{ steps.start-instance.outputs.label }}
+      # Use permanent remote instance label first as on-demand remote instance label output is set before the end of start-remote-instance step.
+      # If the latter fails due to a failed GitHub action runner set up, we have to fallback on the permanent instance.
+      # Since the on-demand remote label is set before failure, we have to do the logical OR in this order,
+      # otherwise we'll try to run the next job on a non-existing on-demand instance.
+      runner-name: ${{ steps.use-permanent-instance.outputs.runner_group || steps.start-remote-instance.outputs.label }}
+      remote-instance-outcome: ${{ steps.start-remote-instance.outcome }}
     steps:
-      - name: Start instance
-        id: start-instance
-        uses: zama-ai/slab-github-runner@801df0b8db5ea2b06128b7476c652f5ed5f193a8
+      - name: Start remote instance
+        id: start-remote-instance
+        continue-on-error: true
+        uses: zama-ai/slab-github-runner@79939325c3c429837c10d6041e4fd8589d328bac
         with:
           mode: start
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}
@@ -34,11 +39,18 @@ jobs:
           backend: hyperstack
           profile: single-h100
 
+      # This will allow to fallback on permanent instances running on Hyperstack.
+      - name: Use permanent remote instance
+        id: use-permanent-instance
+        if: env.SECRETS_AVAILABLE == 'true' && steps.start-remote-instance.outcome == 'failure'
+        run: |
+          echo "runner_group=h100x1" >> "$GITHUB_OUTPUT"
+
   cuda-tests-linux:
     name: CUDA H100 tests
     needs: [ setup-instance ]
     concurrency:
-      group: ${{ github.workflow }}_${{ github.ref }}
+      group: ${{ github.workflow_ref }}
       cancel-in-progress: ${{ github.ref != 'refs/heads/main' }}
     runs-on: ${{ needs.setup-instance.outputs.runner-name }}
     strategy:
@@ -49,9 +61,6 @@ jobs:
           - os: ubuntu-22.04
             cuda: "12.2"
             gcc: 11 
-    env:
-      CUDA_PATH: /usr/local/cuda-${{ matrix.cuda }}
-      CMAKE_VERSION: 3.29.6
     steps:
       # Mandatory on hyperstack since a bootable volume is not re-usable yet.
       - name: Install dependencies
@@ -69,40 +78,20 @@ jobs:
         uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
         with:
           persist-credentials: 'false'
-          token: ${{ secrets.FHE_ACTIONS_TOKEN }}
+          token: ${{ secrets.REPO_CHECKOUT_TOKEN }}
 
-      - name: Set up home
-        run: |
-          echo "HOME=/home/ubuntu" >> "${GITHUB_ENV}"
+      - name: Setup Hyperstack dependencies
+        if: needs.setup-instance.outputs.remote-instance-outcome == 'success'
+        uses: ./.github/actions/gpu_setup
+        with:
+          cuda-version: ${{ matrix.cuda }}
+          gcc-version: ${{ matrix.gcc }}
 
       - name: Install latest stable
-        uses: dtolnay/rust-toolchain@7b1c307e0dcbda6122208f10795a713336a9b35a
+        uses: dtolnay/rust-toolchain@888c2e1ea69ab0d4330cbf0af1ecc7b68f368cc1
         with:
           toolchain: stable
 
-      - name: Export CUDA variables
-        if: ${{ !cancelled() }}
-        run: |
-          echo "CUDA_PATH=$CUDA_PATH" >> "${GITHUB_ENV}"
-          echo "$CUDA_PATH/bin" >> "${GITHUB_PATH}"
-          echo "LD_LIBRARY_PATH=$CUDA_PATH/lib:$LD_LIBRARY_PATH" >> "${GITHUB_ENV}"
-          echo "CUDACXX=/usr/local/cuda-${{ matrix.cuda }}/bin/nvcc" >> "${GITHUB_ENV}"
-
-      # Specify the correct host compilers
-      - name: Export gcc and g++ variables
-        if: ${{ !cancelled() }}
-        run: |
-          {
-            echo "CC=/usr/bin/gcc-${{ matrix.gcc }}";
-            echo "CXX=/usr/bin/g++-${{ matrix.gcc }}";
-            echo "CUDAHOSTCXX=/usr/bin/g++-${{ matrix.gcc }}";
-            echo "HOME=/home/ubuntu";
-          } >> "${GITHUB_ENV}"
-
-      - name: Check device is detected
-        if: ${{ !cancelled() }}
-        run: nvidia-smi
-
       - name: Run core crypto, integer and internal CUDA backend tests
         run: |
           make test_gpu
@@ -134,12 +123,13 @@ jobs:
 
   teardown-instance:
     name: Teardown instance (cuda-h100-tests)
+    if: ${{ always() && needs.setup-instance.outputs.remote-instance-outcome == 'success' }}
     needs: [ setup-instance, cuda-tests-linux ]
     runs-on: ubuntu-latest
     steps:
       - name: Stop instance
         id: stop-instance
-        uses: zama-ai/slab-github-runner@801df0b8db5ea2b06128b7476c652f5ed5f193a8
+        uses: zama-ai/slab-github-runner@79939325c3c429837c10d6041e4fd8589d328bac
         with:
           mode: stop
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}

.github/workflows/gpu_full_multi_gpu_tests.yml

@@ -1,5 +1,5 @@
 # Compile and test tfhe-cuda-backend on an AWS instance
-name: TFHE Cuda Backend - Full tests multi-GPU
+name: Cuda - Full tests multi-GPU
 
 env:
   CARGO_TERM_COLOR: always
@@ -11,7 +11,13 @@ env:
   SLACK_ICON: https://pbs.twimg.com/profile_images/1274014582265298945/OjBKP9kn_400x400.png
   SLACK_USERNAME: ${{ secrets.BOT_USERNAME }}
   SLACK_WEBHOOK: ${{ secrets.SLACK_WEBHOOK }}
+  SLACKIFY_MARKDOWN: true
   IS_PULL_REQUEST: ${{ github.event_name == 'pull_request' }}
+  PULL_REQUEST_MD_LINK: ""
+  CHECKOUT_TOKEN: ${{ secrets.REPO_CHECKOUT_TOKEN || secrets.GITHUB_TOKEN }}
+  # Secrets will be available only to zama-ai organization members
+  SECRETS_AVAILABLE: ${{ secrets.JOB_SECRET != '' }}
+  EXTERNAL_CONTRIBUTION_RUNNER: "gpu_ubuntu-22.04"
 
 on:
   # Allows you to run this workflow manually from the Actions tab as an alternative.
@@ -23,7 +29,7 @@ jobs:
   should-run:
     runs-on: ubuntu-latest
     permissions:
-      pull-requests: write
+      pull-requests: read
     outputs:
       gpu_test: ${{ env.IS_PULL_REQUEST == 'false' || steps.changed-files.outputs.gpu_any_changed }}
     steps:
@@ -31,12 +37,13 @@ jobs:
         uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
         with:
           fetch-depth: 0
+          persist-credentials: 'false'
+          token: ${{ env.CHECKOUT_TOKEN }}
 
       - name: Check for file changes
         id: changed-files
-        uses: tj-actions/changed-files@4edd678ac3f81e2dc578756871e4d00c19191daf
+        uses: tj-actions/changed-files@26a38635fc1173cc5820336ce97be6188d0de9f5 # v46.0.2
         with:
-          since_last_remote_commit: true
           files_yaml: |
             gpu:
               - tfhe/Cargo.toml
@@ -50,7 +57,7 @@ jobs:
               - tfhe/src/shortint/parameters/**
               - tfhe/src/high_level_api/**
               - tfhe/src/c_api/**
-              - 'tfhe/docs/**.md'
+              - 'tfhe/docs/**/**.md'
               - '.github/workflows/**_multi_gpu_tests.yml'
               - scripts/integer-tests.sh
               - ci/slab.toml
@@ -63,11 +70,12 @@ jobs:
       (github.event.action == 'labeled' && github.event.label.name == 'approved' && needs.should-run.outputs.gpu_test == 'true')
     runs-on: ubuntu-latest
     outputs:
-      runner-name: ${{ steps.start-instance.outputs.label }}
+      runner-name: ${{ steps.start-remote-instance.outputs.label || steps.start-github-instance.outputs.runner_group }}
     steps:
-      - name: Start instance
-        id: start-instance
-        uses: zama-ai/slab-github-runner@801df0b8db5ea2b06128b7476c652f5ed5f193a8
+      - name: Start remote instance
+        id: start-remote-instance
+        if: env.SECRETS_AVAILABLE == 'true'
+        uses: zama-ai/slab-github-runner@79939325c3c429837c10d6041e4fd8589d328bac
         with:
           mode: start
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}
@@ -76,13 +84,20 @@ jobs:
           backend: hyperstack
           profile: multi-gpu-test
 
+      # This instance will be spawned especially for pull-request from forked repository
+      - name: Start GitHub instance
+        id: start-github-instance
+        if: env.SECRETS_AVAILABLE == 'false'
+        run: |
+          echo "runner_group=${{ env.EXTERNAL_CONTRIBUTION_RUNNER }}" >> "$GITHUB_OUTPUT"
+
   cuda-tests-linux:
     name: CUDA multi-GPU tests
     needs: [ should-run, setup-instance ]
     if: github.event_name != 'pull_request' ||
       (github.event_name == 'pull_request' && needs.setup-instance.result != 'skipped')
     concurrency:
-      group: ${{ github.workflow }}_${{ github.ref }}
+      group: ${{ github.workflow_ref }}
       cancel-in-progress: ${{ github.ref != 'refs/heads/main' }}
     runs-on: ${{ needs.setup-instance.outputs.runner-name }}
     strategy:
@@ -93,60 +108,25 @@ jobs:
           - os: ubuntu-22.04
             cuda: "12.2"
             gcc: 11 
-    env:
-      CUDA_PATH: /usr/local/cuda-${{ matrix.cuda }}
-      CMAKE_VERSION: 3.29.6
     steps:
-      # Mandatory on hyperstack since a bootable volume is not re-usable yet.
-      - name: Install dependencies
-        run: |
-          sudo apt update
-          sudo apt install -y checkinstall zlib1g-dev libssl-dev libclang-dev
-          wget https://github.com/Kitware/CMake/releases/download/v${{ env.CMAKE_VERSION }}/cmake-${{ env.CMAKE_VERSION }}.tar.gz
-          tar -zxvf cmake-${{ env.CMAKE_VERSION }}.tar.gz
-          cd cmake-${{ env.CMAKE_VERSION }}
-          ./bootstrap
-          make -j"$(nproc)"
-          sudo make install
-
       - name: Checkout tfhe-rs
         uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
         with:
           persist-credentials: 'false'
-          token: ${{ secrets.FHE_ACTIONS_TOKEN }}
+          token: ${{ env.CHECKOUT_TOKEN }}
 
-      - name: Set up home
-        run: |
-          echo "HOME=/home/ubuntu" >> "${GITHUB_ENV}"
+      - name: Setup Hyperstack dependencies
+        uses: ./.github/actions/gpu_setup
+        with:
+          cuda-version: ${{ matrix.cuda }}
+          gcc-version: ${{ matrix.gcc }}
+          github-instance: ${{ env.SECRETS_AVAILABLE == 'false' }}
 
       - name: Install latest stable
-        uses: dtolnay/rust-toolchain@7b1c307e0dcbda6122208f10795a713336a9b35a
+        uses: dtolnay/rust-toolchain@888c2e1ea69ab0d4330cbf0af1ecc7b68f368cc1
         with:
           toolchain: stable
 
-      - name: Export CUDA variables
-        if: ${{ !cancelled() }}
-        run: |
-          echo "CUDA_PATH=$CUDA_PATH" >> "${GITHUB_ENV}"
-          echo "$CUDA_PATH/bin" >> "${GITHUB_PATH}"
-          echo "LD_LIBRARY_PATH=$CUDA_PATH/lib:$LD_LIBRARY_PATH" >> "${GITHUB_ENV}"
-          echo "CUDACXX=/usr/local/cuda-${{ matrix.cuda }}/bin/nvcc" >> "${GITHUB_ENV}"
-
-      # Specify the correct host compilers
-      - name: Export gcc and g++ variables
-        if: ${{ !cancelled() }}
-        run: |
-          {
-            echo "CC=/usr/bin/gcc-${{ matrix.gcc }}";
-            echo "CXX=/usr/bin/g++-${{ matrix.gcc }}";
-            echo "CUDAHOSTCXX=/usr/bin/g++-${{ matrix.gcc }}";
-            echo "HOME=/home/ubuntu";
-          } >> "${GITHUB_ENV}"
-
-      - name: Check device is detected
-        if: ${{ !cancelled() }}
-        run: nvidia-smi
-
       - name: Run multi-bit CUDA integer compression tests
         run: |
           BIG_TESTS_INSTANCE=TRUE make test_integer_compression_gpu
@@ -175,21 +155,28 @@ jobs:
     if: ${{ always() && needs.cuda-tests-linux.result != 'skipped' && failure() }}
     continue-on-error: true
     steps:
+      - name: Set pull-request URL
+        if: env.SECRETS_AVAILABLE == 'true' && github.event_name == 'pull_request'
+        run: |
+          echo "PULL_REQUEST_MD_LINK=[pull-request](${{ vars.PR_BASE_URL }}${{ github.event.pull_request.number }}), "  >> "${GITHUB_ENV}"
+
       - name: Send message
+        if: env.SECRETS_AVAILABLE == 'true'
         uses: rtCamp/action-slack-notify@c33737706dea87cd7784c687dadc9adf1be59990
         env:
           SLACK_COLOR: ${{ needs.cuda-tests-linux.result }}
-          SLACK_MESSAGE: "Multi-GPU tests finished with status: ${{ needs.cuda-tests-linux.result }}. (${{ env.ACTION_RUN_URL }})"
+          SLACK_MESSAGE: "Multi-GPU tests finished with status: ${{ needs.cuda-tests-linux.result }}. (${{ env.PULL_REQUEST_MD_LINK }}[action run](${{ env.ACTION_RUN_URL }}))"
 
   teardown-instance:
     name: Teardown instance (cuda-tests-multi-gpu)
-    if: ${{ always() && needs.setup-instance.result != 'skipped' }}
+    if: ${{ always() && needs.setup-instance.result == 'success' }}
     needs: [ setup-instance, cuda-tests-linux ]
     runs-on: ubuntu-latest
     steps:
-      - name: Stop instance
+      - name: Stop remote instance
         id: stop-instance
-        uses: zama-ai/slab-github-runner@801df0b8db5ea2b06128b7476c652f5ed5f193a8
+        if: env.SECRETS_AVAILABLE == 'true'
+        uses: zama-ai/slab-github-runner@79939325c3c429837c10d6041e4fd8589d328bac
         with:
           mode: stop
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}

.github/workflows/gpu_integer_long_run_tests.yml

@@ -0,0 +1,111 @@
+name: Cuda - Long Run Tests on GPU
+
+env:
+  CARGO_TERM_COLOR: always
+  ACTION_RUN_URL: ${{ github.server_url }}/${{ github.repository }}/actions/runs/${{ github.run_id }}
+  RUSTFLAGS: "-C target-cpu=native"
+  RUST_BACKTRACE: "full"
+  RUST_MIN_STACK: "8388608"
+  SLACK_CHANNEL: ${{ secrets.SLACK_CHANNEL }}
+  SLACK_ICON: https://pbs.twimg.com/profile_images/1274014582265298945/OjBKP9kn_400x400.png
+  SLACK_USERNAME: ${{ secrets.BOT_USERNAME }}
+  SLACK_WEBHOOK: ${{ secrets.SLACK_WEBHOOK }}
+
+on:
+  # Allows you to run this workflow manually from the Actions tab as an alternative.
+  workflow_dispatch:
+  schedule:
+    # Nightly tests will be triggered each evening 8p.m.
+    - cron: "0 20 * * *"
+
+jobs:
+  setup-instance:
+    name: Setup instance (gpu-tests)
+    if: github.event_name != 'schedule' ||
+      (github.event_name == 'schedule' && github.repository == 'zama-ai/tfhe-rs')
+    runs-on: ubuntu-latest
+    outputs:
+      runner-name: ${{ steps.start-instance.outputs.label }}
+    steps:
+      - name: Start instance
+        id: start-instance
+        uses: zama-ai/slab-github-runner@79939325c3c429837c10d6041e4fd8589d328bac
+        with:
+          mode: start
+          github-token: ${{ secrets.SLAB_ACTION_TOKEN }}
+          slab-url: ${{ secrets.SLAB_BASE_URL }}
+          job-secret: ${{ secrets.JOB_SECRET }}
+          backend: hyperstack
+          profile: multi-gpu-test
+
+  cuda-tests:
+    name: Long run GPU tests
+    needs: [ setup-instance ]
+    concurrency:
+      group: ${{ github.workflow_ref }}_${{github.event_name}}
+      cancel-in-progress: true
+    runs-on: ${{ needs.setup-instance.outputs.runner-name }}
+    strategy:
+      fail-fast: false
+      # explicit include-based build matrix, of known valid options
+      matrix:
+        include:
+          - os: ubuntu-22.04
+            cuda: "12.2"
+            gcc: 11 
+    timeout-minutes: 4320 # 72 hours
+    steps:
+      - name: Checkout tfhe-rs
+        uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
+
+      - name: Setup Hyperstack dependencies
+        uses: ./.github/actions/gpu_setup
+        with:
+          cuda-version: ${{ matrix.cuda }}
+          gcc-version: ${{ matrix.gcc }}
+
+      - name: Install latest stable
+        uses: dtolnay/rust-toolchain@888c2e1ea69ab0d4330cbf0af1ecc7b68f368cc1
+        with:
+          toolchain: stable
+
+      - name: Run tests
+        run: |
+          make test_integer_long_run_gpu
+
+  slack-notify:
+    name: Slack Notification
+    needs: [ setup-instance, cuda-tests ]
+    runs-on: ubuntu-latest
+    if: ${{ always() && needs.cuda-tests.result != 'skipped' && failure() }}
+    continue-on-error: true
+    steps:
+      - name: Send message
+        uses: rtCamp/action-slack-notify@c33737706dea87cd7784c687dadc9adf1be59990
+        env:
+          SLACK_COLOR: ${{ needs.cuda-tests.result }}
+          SLACK_MESSAGE: "Integer GPU long run tests finished with status: ${{ needs.cuda-tests.result }}. (${{ env.ACTION_RUN_URL }})"
+
+  teardown-instance:
+    name: Teardown instance (gpu-tests)
+    if: ${{ always() && needs.setup-instance.result == 'success' }}
+    needs: [ setup-instance, cuda-tests ]
+    runs-on: ubuntu-latest
+    steps:
+      - name: Stop instance
+        id: stop-instance
+        uses: zama-ai/slab-github-runner@79939325c3c429837c10d6041e4fd8589d328bac
+        with:
+          mode: stop
+          github-token: ${{ secrets.SLAB_ACTION_TOKEN }}
+          slab-url: ${{ secrets.SLAB_BASE_URL }}
+          job-secret: ${{ secrets.JOB_SECRET }}
+          label: ${{ needs.setup-instance.outputs.runner-name }}
+
+      - name: Slack Notification
+        if: ${{ failure() }}
+        continue-on-error: true
+        uses: rtCamp/action-slack-notify@c33737706dea87cd7784c687dadc9adf1be59990
+        env:
+          SLACK_COLOR: ${{ job.status }}
+          SLACK_MESSAGE: "Instance teardown (gpu-long-run-tests) finished with status: ${{ job.status }}. (${{ env.ACTION_RUN_URL }})"

.github/workflows/gpu_pcc.yml

@@ -1,5 +1,5 @@
 # Perfom tfhe-cuda-backend post-commit checks on an AWS instance
-name: TFHE Cuda Backend - Post-commit Checks
+name: Cuda - Post-commit Checks
 
 env:
   CARGO_TERM_COLOR: always
@@ -11,6 +11,12 @@ env:
   SLACK_ICON: https://pbs.twimg.com/profile_images/1274014582265298945/OjBKP9kn_400x400.png
   SLACK_USERNAME: ${{ secrets.BOT_USERNAME }}
   SLACK_WEBHOOK: ${{ secrets.SLACK_WEBHOOK }}
+  SLACKIFY_MARKDOWN: true
+  PULL_REQUEST_MD_LINK: ""
+  CHECKOUT_TOKEN: ${{ secrets.REPO_CHECKOUT_TOKEN || secrets.GITHUB_TOKEN }}
+  # Secrets will be available only to zama-ai organization members
+  SECRETS_AVAILABLE: ${{ secrets.JOB_SECRET != '' }}
+  EXTERNAL_CONTRIBUTION_RUNNER: "large_ubuntu_16-22.04"
 
 on:
   pull_request:
@@ -20,11 +26,12 @@ jobs:
     name: Setup instance (cuda-pcc)
     runs-on: ubuntu-latest
     outputs:
-      runner-name: ${{ steps.start-instance.outputs.label }}
+      runner-name: ${{ steps.start-remote-instance.outputs.label || steps.start-github-instance.outputs.runner_group }}
     steps:
-      - name: Start instance
-        id: start-instance
-        uses: zama-ai/slab-github-runner@801df0b8db5ea2b06128b7476c652f5ed5f193a8
+      - name: Start remote instance
+        id: start-remote-instance
+        if: env.SECRETS_AVAILABLE == 'true'
+        uses: zama-ai/slab-github-runner@79939325c3c429837c10d6041e4fd8589d328bac
         with:
           mode: start
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}
@@ -33,11 +40,18 @@ jobs:
           backend: aws
           profile: gpu-build
 
+      # This instance will be spawned especially for pull-request from forked repository
+      - name: Start GitHub instance
+        id: start-github-instance
+        if: env.SECRETS_AVAILABLE == 'false'
+        run: |
+          echo "runner_group=${{ env.EXTERNAL_CONTRIBUTION_RUNNER }}" >> "$GITHUB_OUTPUT"
+
   cuda-pcc:
     name: CUDA post-commit checks
     needs: setup-instance
     concurrency:
-      group: ${{ github.workflow }}_${{ github.ref }}
+      group: ${{ github.workflow_ref }}
       cancel-in-progress: true
     runs-on: ${{ needs.setup-instance.outputs.runner-name }}
     strategy:
@@ -56,14 +70,20 @@ jobs:
         uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
         with:
           persist-credentials: 'false'
-          token: ${{ secrets.FHE_ACTIONS_TOKEN }}
+          token: ${{ env.CHECKOUT_TOKEN }}
 
-      - name: Set up home
+      - name: Install CUDA
+        if: env.SECRETS_AVAILABLE == 'false'
+        shell: bash
         run: |
-          echo "HOME=/home/ubuntu" >> "${GITHUB_ENV}"
+          TOOLKIT_VERSION="$(echo ${{ matrix.cuda }} | sed 's/\(.*\)\.\(.*\)/\1-\2/')"
+          wget https://developer.download.nvidia.com/compute/cuda/repos/ubuntu2204/x86_64/cuda-keyring_1.1-1_all.deb
+          sudo dpkg -i cuda-keyring_1.1-1_all.deb
+          sudo apt update
+          sudo apt -y install "cuda-toolkit-${TOOLKIT_VERSION}" cmake-format
 
       - name: Install latest stable
-        uses: dtolnay/rust-toolchain@7b1c307e0dcbda6122208f10795a713336a9b35a
+        uses: dtolnay/rust-toolchain@888c2e1ea69ab0d4330cbf0af1ecc7b68f368cc1
         with:
           toolchain: stable
 
@@ -83,7 +103,6 @@ jobs:
             echo "CC=/usr/bin/gcc-${{ matrix.gcc }}";
             echo "CXX=/usr/bin/g++-${{ matrix.gcc }}";
             echo "CUDAHOSTCXX=/usr/bin/g++-${{ matrix.gcc }}";
-            echo "HOME=/home/ubuntu";
           } >> "${GITHUB_ENV}"
 
       - name: Run fmt checks
@@ -94,23 +113,29 @@ jobs:
         run: |
           make pcc_gpu
 
+      - name: Set pull-request URL
+        if: ${{ failure() && github.event_name == 'pull_request' }}
+        run: |
+          echo "PULL_REQUEST_MD_LINK=[pull-request](${{ vars.PR_BASE_URL }}${{ github.event.pull_request.number }}), "  >> "${GITHUB_ENV}"
+
       - name: Slack Notification
-        if: ${{ failure() }}
+        if: ${{ failure() && env.SECRETS_AVAILABLE == 'true' }}
         continue-on-error: true
         uses: rtCamp/action-slack-notify@c33737706dea87cd7784c687dadc9adf1be59990
         env:
           SLACK_COLOR: ${{ job.status }}
-          SLACK_MESSAGE: "CUDA AWS post-commit checks finished with status: ${{ job.status }}. (${{ env.ACTION_RUN_URL }})"
+          SLACK_MESSAGE: "CUDA AWS post-commit checks finished with status: ${{ job.status }}. (${{ env.PULL_REQUEST_MD_LINK }}[action run](${{ env.ACTION_RUN_URL }}))"
 
   teardown-instance:
     name: Teardown instance (cuda-pcc)
-    if: ${{ always() && needs.setup-instance.result != 'skipped' }}
+    if: ${{ always() && needs.setup-instance.result == 'success' }}
     needs: [ setup-instance, cuda-pcc ]
     runs-on: ubuntu-latest
     steps:
-      - name: Stop instance
+      - name: Stop remote instance
         id: stop-instance
-        uses: zama-ai/slab-github-runner@801df0b8db5ea2b06128b7476c652f5ed5f193a8
+        if: env.SECRETS_AVAILABLE == 'true'
+        uses: zama-ai/slab-github-runner@79939325c3c429837c10d6041e4fd8589d328bac
         with:
           mode: stop
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}

.github/workflows/gpu_signed_integer_classic_tests.yml

@@ -1,5 +1,5 @@
 # Signed integer GPU tests on an RTXA6000 VM on hyperstack with classical PBS
-name: TFHE Cuda Backend - Signed integer tests with classical PBS
+name: Cuda - Signed integer tests with classical PBS
 
 env:
   CARGO_TERM_COLOR: always
@@ -11,19 +11,25 @@ env:
   SLACK_ICON: https://pbs.twimg.com/profile_images/1274014582265298945/OjBKP9kn_400x400.png
   SLACK_USERNAME: ${{ secrets.BOT_USERNAME }}
   SLACK_WEBHOOK: ${{ secrets.SLACK_WEBHOOK }}
+  SLACKIFY_MARKDOWN: true
   IS_PULL_REQUEST: ${{ github.event_name == 'pull_request' }}
+  PULL_REQUEST_MD_LINK: ""
+  CHECKOUT_TOKEN: ${{ secrets.REPO_CHECKOUT_TOKEN || secrets.GITHUB_TOKEN }}
+  # Secrets will be available only to zama-ai organization members
+  SECRETS_AVAILABLE: ${{ secrets.JOB_SECRET != '' }}
+  EXTERNAL_CONTRIBUTION_RUNNER: "gpu_ubuntu-22.04"
 
 on:
   # Allows you to run this workflow manually from the Actions tab as an alternative.
   workflow_dispatch:
   pull_request:
-      types: [ labeled ]
+    types: [ labeled ]
 
 jobs:
   should-run:
     runs-on: ubuntu-latest
     permissions:
-      pull-requests: write
+      pull-requests: read
     outputs:
       gpu_test: ${{ env.IS_PULL_REQUEST == 'false' || steps.changed-files.outputs.gpu_any_changed }}
     steps:
@@ -31,13 +37,13 @@ jobs:
         uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
         with:
           fetch-depth: 0
-          token: ${{ secrets.FHE_ACTIONS_TOKEN }}
+          persist-credentials: 'false'
+          token: ${{ env.CHECKOUT_TOKEN }}
 
       - name: Check for file changes
         id: changed-files
-        uses: tj-actions/changed-files@4edd678ac3f81e2dc578756871e4d00c19191daf
+        uses: tj-actions/changed-files@26a38635fc1173cc5820336ce97be6188d0de9f5 # v46.0.2
         with:
-          since_last_remote_commit: true
           files_yaml: |
             gpu:
               - tfhe/Cargo.toml
@@ -51,7 +57,7 @@ jobs:
               - tfhe/src/shortint/parameters/**
               - tfhe/src/high_level_api/**
               - tfhe/src/c_api/**
-              - 'tfhe/docs/**.md'
+              - 'tfhe/docs/**/**.md'
               - '.github/workflows/gpu_signed_integer_classic_tests.yml'
               - scripts/integer-tests.sh
               - ci/slab.toml
@@ -64,11 +70,12 @@ jobs:
       (github.event.action == 'labeled' && github.event.label.name == 'approved' && needs.should-run.outputs.gpu_test == 'true')
     runs-on: ubuntu-latest
     outputs:
-      runner-name: ${{ steps.start-instance.outputs.label }}
+      runner-name: ${{ steps.start-remote-instance.outputs.label || steps.start-github-instance.outputs.runner_group }}
     steps:
-      - name: Start instance
-        id: start-instance
-        uses: zama-ai/slab-github-runner@801df0b8db5ea2b06128b7476c652f5ed5f193a8
+      - name: Start remote instance
+        id: start-remote-instance
+        if: env.SECRETS_AVAILABLE == 'true'
+        uses: zama-ai/slab-github-runner@79939325c3c429837c10d6041e4fd8589d328bac
         with:
           mode: start
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}
@@ -77,13 +84,20 @@ jobs:
           backend: hyperstack
           profile: gpu-test
 
+      # This instance will be spawned especially for pull-request from forked repository
+      - name: Start GitHub instance
+        id: start-github-instance
+        if: env.SECRETS_AVAILABLE == 'false'
+        run: |
+          echo "runner_group=${{ env.EXTERNAL_CONTRIBUTION_RUNNER }}" >> "$GITHUB_OUTPUT"
+
   cuda-tests-linux:
     name: CUDA signed integer tests with classical PBS
     needs: [ should-run, setup-instance ]
     if: github.event_name != 'pull_request' ||
       (github.event_name == 'pull_request' && needs.setup-instance.result != 'skipped')
     concurrency:
-      group: ${{ github.workflow }}_${{ github.ref }}
+      group: ${{ github.workflow_ref }}
       cancel-in-progress: ${{ github.ref != 'refs/heads/main' }}
     runs-on: ${{ needs.setup-instance.outputs.runner-name }}
     strategy:
@@ -94,58 +108,25 @@ jobs:
           - os: ubuntu-22.04
             cuda: "12.2"
             gcc: 11 
-    env:
-      CUDA_PATH: /usr/local/cuda-${{ matrix.cuda }}
-      CMAKE_VERSION: 3.29.6
     steps:
-      # Mandatory on hyperstack since a bootable volume is not re-usable yet.
-      - name: Install dependencies
-        run: |
-          sudo apt update
-          sudo apt install -y checkinstall zlib1g-dev libssl-dev libclang-dev
-          wget https://github.com/Kitware/CMake/releases/download/v${{ env.CMAKE_VERSION }}/cmake-${{ env.CMAKE_VERSION }}.tar.gz
-          tar -zxvf cmake-${{ env.CMAKE_VERSION }}.tar.gz
-          cd cmake-${{ env.CMAKE_VERSION }}
-          ./bootstrap
-          make -j"$(nproc)"
-          sudo make install
-
-
       - name: Checkout tfhe-rs
         uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
+        with:
+          persist-credentials: 'false'
+          token: ${{ env.CHECKOUT_TOKEN }}
 
-      - name: Set up home
-        run: |
-          echo "HOME=/home/ubuntu" >> "${GITHUB_ENV}"
+      - name: Setup Hyperstack dependencies
+        uses: ./.github/actions/gpu_setup
+        with:
+          cuda-version: ${{ matrix.cuda }}
+          gcc-version: ${{ matrix.gcc }}
+          github-instance: ${{ env.SECRETS_AVAILABLE == 'false' }}
 
       - name: Install latest stable
-        uses: dtolnay/rust-toolchain@7b1c307e0dcbda6122208f10795a713336a9b35a
+        uses: dtolnay/rust-toolchain@888c2e1ea69ab0d4330cbf0af1ecc7b68f368cc1
         with:
           toolchain: stable
 
-      - name: Export CUDA variables
-        if: ${{ !cancelled() }}
-        run: |
-          echo "CUDA_PATH=$CUDA_PATH" >> "${GITHUB_ENV}"
-          echo "$CUDA_PATH/bin" >> "${GITHUB_PATH}"
-          echo "LD_LIBRARY_PATH=$CUDA_PATH/lib:$LD_LIBRARY_PATH" >> "${GITHUB_ENV}"
-          echo "CUDACXX=/usr/local/cuda-${{ matrix.cuda }}/bin/nvcc" >> "${GITHUB_ENV}"
-
-      # Specify the correct host compilers
-      - name: Export gcc and g++ variables
-        if: ${{ !cancelled() }}
-        run: |
-          {
-            echo "CC=/usr/bin/gcc-${{ matrix.gcc }}";
-            echo "CXX=/usr/bin/g++-${{ matrix.gcc }}";
-            echo "CUDAHOSTCXX=/usr/bin/g++-${{ matrix.gcc }}";
-            echo "HOME=/home/ubuntu";
-          } >> "${GITHUB_ENV}"
-
-      - name: Check device is detected
-        if: ${{ !cancelled() }}
-        run: nvidia-smi
-
       - name: Run signed integer tests
         run: |
           BIG_TESTS_INSTANCE=TRUE make test_signed_integer_gpu_ci
@@ -157,21 +138,28 @@ jobs:
     if: ${{ always() && needs.cuda-tests-linux.result != 'skipped' && failure() }}
     continue-on-error: true
     steps:
+      - name: Set pull-request URL
+        if: env.SECRETS_AVAILABLE == 'true' && github.event_name == 'pull_request'
+        run: |
+          echo "PULL_REQUEST_MD_LINK=[pull-request](${{ vars.PR_BASE_URL }}${{ github.event.pull_request.number }}), "  >> "${GITHUB_ENV}"
+
       - name: Send message
+        if: env.SECRETS_AVAILABLE == 'true'
         uses: rtCamp/action-slack-notify@c33737706dea87cd7784c687dadc9adf1be59990
         env:
           SLACK_COLOR: ${{ needs.cuda-tests-linux.result }}
-          SLACK_MESSAGE: "Integer GPU signed integer tests with classical PBS finished with status: ${{ needs.cuda-tests-linux.result }}. (${{ env.ACTION_RUN_URL }})"
+          SLACK_MESSAGE: "Integer GPU signed integer tests with classical PBS finished with status: ${{ needs.cuda-tests-linux.result }}. (${{ env.PULL_REQUEST_MD_LINK }}[action run](${{ env.ACTION_RUN_URL }}))"
 
   teardown-instance:
     name: Teardown instance (cuda-signed-classic-tests)
-    if: ${{ always() && needs.setup-instance.result != 'skipped' }}
+    if: ${{ always() && needs.setup-instance.result == 'success' }}
     needs: [ setup-instance, cuda-tests-linux ]
     runs-on: ubuntu-latest
     steps:
-      - name: Stop instance
+      - name: Stop remote instance
         id: stop-instance
-        uses: zama-ai/slab-github-runner@801df0b8db5ea2b06128b7476c652f5ed5f193a8
+        if: env.SECRETS_AVAILABLE == 'true'
+        uses: zama-ai/slab-github-runner@79939325c3c429837c10d6041e4fd8589d328bac
         with:
           mode: stop
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}

.github/workflows/gpu_signed_integer_h100_tests.yml

@@ -1,5 +1,5 @@
 # Signed integer GPU tests on an H100 VM on hyperstack
-name: TFHE Cuda Backend - Signed integer tests on H100
+name: Cuda - Signed integer tests on H100
 
 env:
   CARGO_TERM_COLOR: always
@@ -11,19 +11,26 @@ env:
   SLACK_ICON: https://pbs.twimg.com/profile_images/1274014582265298945/OjBKP9kn_400x400.png
   SLACK_USERNAME: ${{ secrets.BOT_USERNAME }}
   SLACK_WEBHOOK: ${{ secrets.SLACK_WEBHOOK }}
+  SLACKIFY_MARKDOWN: true
   IS_PULL_REQUEST: ${{ github.event_name == 'pull_request' }}
+  PULL_REQUEST_MD_LINK: ""
+  CHECKOUT_TOKEN: ${{ secrets.REPO_CHECKOUT_TOKEN || secrets.GITHUB_TOKEN }}
+  # Secrets will be available only to zama-ai organization members
+  SECRETS_AVAILABLE: ${{ secrets.JOB_SECRET != '' }}
+  EXTERNAL_CONTRIBUTION_RUNNER: "gpu_ubuntu-22.04"
 
 on:
   # Allows you to run this workflow manually from the Actions tab as an alternative.
   workflow_dispatch:
   pull_request:
-      types: [ labeled ]
+    types: [ labeled ]
+
 
 jobs:
   should-run:
     runs-on: ubuntu-latest
     permissions:
-      pull-requests: write
+      pull-requests: read
     outputs:
       gpu_test: ${{ env.IS_PULL_REQUEST == 'false' || steps.changed-files.outputs.gpu_any_changed }}
     steps:
@@ -31,13 +38,13 @@ jobs:
         uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
         with:
           fetch-depth: 0
-          token: ${{ secrets.FHE_ACTIONS_TOKEN }}
+          persist-credentials: 'false'
+          token: ${{ env.CHECKOUT_TOKEN }}
 
       - name: Check for file changes
         id: changed-files
-        uses: tj-actions/changed-files@4edd678ac3f81e2dc578756871e4d00c19191daf
+        uses: tj-actions/changed-files@26a38635fc1173cc5820336ce97be6188d0de9f5 # v46.0.2
         with:
-          since_last_remote_commit: true
           files_yaml: |
             gpu:
               - tfhe/Cargo.toml
@@ -51,7 +58,7 @@ jobs:
               - tfhe/src/shortint/parameters/**
               - tfhe/src/high_level_api/**
               - tfhe/src/c_api/**
-              - 'tfhe/docs/**.md'
+              - 'tfhe/docs/**/**.md'
               - '.github/workflows/gpu_signed_integer_h100_tests.yml'
               - scripts/integer-tests.sh
               - ci/slab.toml
@@ -64,11 +71,18 @@ jobs:
       (github.event.action == 'labeled' && github.event.label.name == 'approved' && needs.should-run.outputs.gpu_test == 'true')
     runs-on: ubuntu-latest
     outputs:
-      runner-name: ${{ steps.start-instance.outputs.label }}
+      # Use permanent remote instance label first as on-demand remote instance label output is set before the end of start-remote-instance step.
+      # If the latter fails due to a failed GitHub action runner set up, we have to fallback on the permanent instance.
+      # Since the on-demand remote label is set before failure, we have to do the logical OR in this order,
+      # otherwise we'll try to run the next job on a non-existing on-demand instance.
+      runner-name: ${{ steps.use-permanent-instance.outputs.runner_group || steps.start-remote-instance.outputs.label || steps.start-github-instance.outputs.runner_group }}
+      remote-instance-outcome: ${{ steps.start-remote-instance.outcome }}
     steps:
-      - name: Start instance
-        id: start-instance
-        uses: zama-ai/slab-github-runner@801df0b8db5ea2b06128b7476c652f5ed5f193a8
+      - name: Start remote instance
+        id: start-remote-instance
+        if: env.SECRETS_AVAILABLE == 'true'
+        continue-on-error: true
+        uses: zama-ai/slab-github-runner@79939325c3c429837c10d6041e4fd8589d328bac
         with:
           mode: start
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}
@@ -77,13 +91,27 @@ jobs:
           backend: hyperstack
           profile: single-h100
 
+      # This will allow to fallback on permanent instances running on Hyperstack.
+      - name: Use permanent remote instance
+        id: use-permanent-instance
+        if: env.SECRETS_AVAILABLE == 'true' && steps.start-remote-instance.outcome == 'failure'
+        run: |
+          echo "runner_group=h100x1" >> "$GITHUB_OUTPUT"
+
+      # This instance will be spawned especially for pull-request from forked repository
+      - name: Start GitHub instance
+        id: start-github-instance
+        if: env.SECRETS_AVAILABLE == 'false'
+        run: |
+          echo "runner_group=${{ env.EXTERNAL_CONTRIBUTION_RUNNER }}" >> "$GITHUB_OUTPUT"
+
   cuda-tests-linux:
     name: CUDA H100 signed integer tests
     needs: [ should-run, setup-instance ]
     if: github.event_name != 'pull_request' ||
       (github.event_name == 'pull_request' && needs.setup-instance.result != 'skipped')
     concurrency:
-      group: ${{ github.workflow }}_${{ github.ref }}
+      group: ${{ github.workflow_ref }}
       cancel-in-progress: ${{ github.ref != 'refs/heads/main' }}
     runs-on: ${{ needs.setup-instance.outputs.runner-name }}
     strategy:
@@ -94,58 +122,26 @@ jobs:
           - os: ubuntu-22.04
             cuda: "12.2"
             gcc: 11 
-    env:
-      CUDA_PATH: /usr/local/cuda-${{ matrix.cuda }}
-      CMAKE_VERSION: 3.29.6
     steps:
-      # Mandatory on hyperstack since a bootable volume is not re-usable yet.
-      - name: Install dependencies
-        run: |
-          sudo apt update
-          sudo apt install -y checkinstall zlib1g-dev libssl-dev libclang-dev
-          wget https://github.com/Kitware/CMake/releases/download/v${{ env.CMAKE_VERSION }}/cmake-${{ env.CMAKE_VERSION }}.tar.gz
-          tar -zxvf cmake-${{ env.CMAKE_VERSION }}.tar.gz
-          cd cmake-${{ env.CMAKE_VERSION }}
-          ./bootstrap
-          make -j"$(nproc)"
-          sudo make install
-
-
       - name: Checkout tfhe-rs
         uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
+        with:
+          persist-credentials: 'false'
+          token: ${{ env.CHECKOUT_TOKEN }}
 
-      - name: Set up home
-        run: |
-          echo "HOME=/home/ubuntu" >> "${GITHUB_ENV}"
+      - name: Setup Hyperstack dependencies
+        if: needs.setup-instance.outputs.remote-instance-outcome == 'success'
+        uses: ./.github/actions/gpu_setup
+        with:
+          cuda-version: ${{ matrix.cuda }}
+          gcc-version: ${{ matrix.gcc }}
+          github-instance: ${{ env.SECRETS_AVAILABLE == 'false' }}
 
       - name: Install latest stable
-        uses: dtolnay/rust-toolchain@7b1c307e0dcbda6122208f10795a713336a9b35a
+        uses: dtolnay/rust-toolchain@888c2e1ea69ab0d4330cbf0af1ecc7b68f368cc1
         with:
           toolchain: stable
 
-      - name: Export CUDA variables
-        if: ${{ !cancelled() }}
-        run: |
-          echo "CUDA_PATH=$CUDA_PATH" >> "${GITHUB_ENV}"
-          echo "$CUDA_PATH/bin" >> "${GITHUB_PATH}"
-          echo "LD_LIBRARY_PATH=$CUDA_PATH/lib:$LD_LIBRARY_PATH" >> "${GITHUB_ENV}"
-          echo "CUDACXX=/usr/local/cuda-${{ matrix.cuda }}/bin/nvcc" >> "${GITHUB_ENV}"
-
-      # Specify the correct host compilers
-      - name: Export gcc and g++ variables
-        if: ${{ !cancelled() }}
-        run: |
-          {
-            echo "CC=/usr/bin/gcc-${{ matrix.gcc }}";
-            echo "CXX=/usr/bin/g++-${{ matrix.gcc }}";
-            echo "CUDAHOSTCXX=/usr/bin/g++-${{ matrix.gcc }}";
-            echo "HOME=/home/ubuntu";
-          } >> "${GITHUB_ENV}"
-
-      - name: Check device is detected
-        if: ${{ !cancelled() }}
-        run: nvidia-smi
-
       - name: Run signed integer multi-bit tests
         run: |
           BIG_TESTS_INSTANCE=TRUE make test_signed_integer_multi_bit_gpu_ci
@@ -157,21 +153,28 @@ jobs:
     if: ${{ always() && needs.cuda-tests-linux.result != 'skipped' && failure() }}
     continue-on-error: true
     steps:
+      - name: Set pull-request URL
+        if: env.SECRETS_AVAILABLE == 'true' && github.event_name == 'pull_request'
+        run: |
+          echo "PULL_REQUEST_MD_LINK=[pull-request](${{ vars.PR_BASE_URL }}${{ github.event.pull_request.number }}), "  >> "${GITHUB_ENV}"
+
       - name: Send message
+        if: env.SECRETS_AVAILABLE == 'true'
         uses: rtCamp/action-slack-notify@c33737706dea87cd7784c687dadc9adf1be59990
         env:
           SLACK_COLOR: ${{ needs.cuda-tests-linux.result }}
-          SLACK_MESSAGE: "Integer GPU H100 tests finished with status: ${{ needs.cuda-tests-linux.result }}. (${{ env.ACTION_RUN_URL }})"
+          SLACK_MESSAGE: "Integer GPU H100 tests finished with status: ${{ needs.cuda-tests-linux.result }}. (${{ env.PULL_REQUEST_MD_LINK }}[action run](${{ env.ACTION_RUN_URL }}))"
 
   teardown-instance:
     name: Teardown instance (cuda-h100-tests)
-    if: ${{ always() && needs.setup-instance.result != 'skipped' }}
+    if: ${{ always() && needs.setup-instance.outputs.remote-instance-outcome == 'success' }}
     needs: [ setup-instance, cuda-tests-linux ]
     runs-on: ubuntu-latest
     steps:
-      - name: Stop instance
+      - name: Stop remote instance
         id: stop-instance
-        uses: zama-ai/slab-github-runner@801df0b8db5ea2b06128b7476c652f5ed5f193a8
+        if: env.SECRETS_AVAILABLE == 'true'
+        uses: zama-ai/slab-github-runner@79939325c3c429837c10d6041e4fd8589d328bac
         with:
           mode: stop
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}

.github/workflows/gpu_signed_integer_tests.yml

@@ -1,5 +1,5 @@
 # Compile and test tfhe-cuda-backend signed integer on an AWS instance
-name: TFHE Cuda Backend - Signed integer tests
+name: Cuda - Signed integer tests
 
 env:
   CARGO_TERM_COLOR: always
@@ -11,17 +11,20 @@ env:
   SLACK_ICON: https://pbs.twimg.com/profile_images/1274014582265298945/OjBKP9kn_400x400.png
   SLACK_USERNAME: ${{ secrets.BOT_USERNAME }}
   SLACK_WEBHOOK: ${{ secrets.SLACK_WEBHOOK }}
+  SLACKIFY_MARKDOWN: true
   FAST_TESTS: TRUE
   NIGHTLY_TESTS: FALSE
   IS_PULL_REQUEST: ${{ github.event_name == 'pull_request' }}
+  PULL_REQUEST_MD_LINK: ""
+  CHECKOUT_TOKEN: ${{ secrets.REPO_CHECKOUT_TOKEN || secrets.GITHUB_TOKEN }}
+  # Secrets will be available only to zama-ai organization members
+  SECRETS_AVAILABLE: ${{ secrets.JOB_SECRET != '' }}
+  EXTERNAL_CONTRIBUTION_RUNNER: "gpu_ubuntu-22.04"
 
 on:
   # Allows you to run this workflow manually from the Actions tab as an alternative.
   workflow_dispatch:
   pull_request:
-    types:
-      - opened
-      - synchronize
   schedule:
     # Nightly tests @ 1AM after each work day
     - cron: "0 1 * * MON-FRI"
@@ -30,7 +33,7 @@ jobs:
   should-run:
     runs-on: ubuntu-latest
     permissions:
-      pull-requests: write
+      pull-requests: read
     outputs:
       gpu_test: ${{ env.IS_PULL_REQUEST == 'false' || steps.changed-files.outputs.gpu_any_changed }}
     steps:
@@ -38,12 +41,13 @@ jobs:
         uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
         with:
           fetch-depth: 0
+          persist-credentials: 'false'
+          token: ${{ env.CHECKOUT_TOKEN }}
 
       - name: Check for file changes
         id: changed-files
-        uses: tj-actions/changed-files@4edd678ac3f81e2dc578756871e4d00c19191daf
+        uses: tj-actions/changed-files@26a38635fc1173cc5820336ce97be6188d0de9f5 # v46.0.2
         with:
-          since_last_remote_commit: true
           files_yaml: |
             gpu:
               - tfhe/Cargo.toml
@@ -57,7 +61,7 @@ jobs:
               - tfhe/src/shortint/parameters/**
               - tfhe/src/high_level_api/**
               - tfhe/src/c_api/**
-              - 'tfhe/docs/**.md'
+              - 'tfhe/docs/**/**.md'
               - '.github/workflows/gpu_signed_integer_tests.yml'
               - scripts/integer-tests.sh
               - ci/slab.toml
@@ -70,11 +74,12 @@ jobs:
       github.event_name == 'workflow_dispatch' ||
       needs.should-run.outputs.gpu_test == 'true'
     outputs:
-      runner-name: ${{ steps.start-instance.outputs.label }}
+      runner-name: ${{ steps.start-remote-instance.outputs.label || steps.start-github-instance.outputs.runner_group }}
     steps:
-      - name: Start instance
-        id: start-instance
-        uses: zama-ai/slab-github-runner@801df0b8db5ea2b06128b7476c652f5ed5f193a8
+      - name: Start remote instance
+        id: start-remote-instance
+        if: env.SECRETS_AVAILABLE == 'true'
+        uses: zama-ai/slab-github-runner@79939325c3c429837c10d6041e4fd8589d328bac
         with:
           mode: start
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}
@@ -83,13 +88,20 @@ jobs:
           backend: hyperstack
           profile: gpu-test
 
+      # This instance will be spawned especially for pull-request from forked repository
+      - name: Start GitHub instance
+        id: start-github-instance
+        if: env.SECRETS_AVAILABLE == 'false'
+        run: |
+          echo "runner_group=${{ env.EXTERNAL_CONTRIBUTION_RUNNER }}" >> "$GITHUB_OUTPUT"
+
   cuda-signed-integer-tests:
     name: CUDA signed integer tests
     needs: [ should-run, setup-instance ]
     if: github.event_name != 'pull_request' ||
       (github.event_name == 'pull_request' && needs.setup-instance.result != 'skipped')
     concurrency:
-      group: ${{ github.workflow }}_${{ github.ref }}
+      group: ${{ github.workflow_ref }}
       cancel-in-progress: ${{ github.ref != 'refs/heads/main' }}
     runs-on: ${{ needs.setup-instance.outputs.runner-name }}
     strategy:
@@ -99,58 +111,26 @@ jobs:
         include:
           - os: ubuntu-22.04
             cuda: "12.2"
-            gcc: 11 
-    env:
-      CUDA_PATH: /usr/local/cuda-${{ matrix.cuda }}
-      CMAKE_VERSION: 3.29.6
+            gcc: 11
     steps:
-      # Mandatory on hyperstack since a bootable volume is not re-usable yet.
-      - name: Install dependencies
-        run: |
-          sudo apt update
-          sudo apt install -y checkinstall zlib1g-dev libssl-dev libclang-dev
-          wget https://github.com/Kitware/CMake/releases/download/v${{ env.CMAKE_VERSION }}/cmake-${{ env.CMAKE_VERSION }}.tar.gz
-          tar -zxvf cmake-${{ env.CMAKE_VERSION }}.tar.gz
-          cd cmake-${{ env.CMAKE_VERSION }}
-          ./bootstrap
-          make -j"$(nproc)"
-          sudo make install
-
-
       - name: Checkout tfhe-rs
         uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
         with:
           persist-credentials: 'false'
-          token: ${{ secrets.FHE_ACTIONS_TOKEN }}
+          token: ${{ env.CHECKOUT_TOKEN }}
 
-      - name: Set up home
-        run: |
-          echo "HOME=/home/ubuntu" >> "${GITHUB_ENV}"
+      - name: Setup Hyperstack dependencies
+        uses: ./.github/actions/gpu_setup
+        with:
+          cuda-version: ${{ matrix.cuda }}
+          gcc-version: ${{ matrix.gcc }}
+          github-instance: ${{ env.SECRETS_AVAILABLE == 'false' }}
 
       - name: Install latest stable
-        uses: dtolnay/rust-toolchain@7b1c307e0dcbda6122208f10795a713336a9b35a
+        uses: dtolnay/rust-toolchain@888c2e1ea69ab0d4330cbf0af1ecc7b68f368cc1
         with:
           toolchain: stable
 
-      - name: Export CUDA variables
-        if: ${{ !cancelled() }}
-        run: |
-          echo "CUDA_PATH=$CUDA_PATH" >> "${GITHUB_ENV}"
-          echo "$CUDA_PATH/bin" >> "${GITHUB_PATH}"
-          echo "LD_LIBRARY_PATH=$CUDA_PATH/lib:$LD_LIBRARY_PATH" >> "${GITHUB_ENV}"
-          echo "CUDACXX=/usr/local/cuda-${{ matrix.cuda }}/bin/nvcc" >> "${GITHUB_ENV}"
-
-      # Specify the correct host compilers
-      - name: Export gcc and g++ variables
-        if: ${{ !cancelled() }}
-        run: |
-          {
-            echo "CC=/usr/bin/gcc-${{ matrix.gcc }}";
-            echo "CXX=/usr/bin/g++-${{ matrix.gcc }}";
-            echo "CUDAHOSTCXX=/usr/bin/g++-${{ matrix.gcc }}";
-            echo "HOME=/home/ubuntu";
-          } >> "${GITHUB_ENV}"
-
       - name: Should run nightly tests
         if: github.event_name == 'schedule'
         run: |
@@ -159,10 +139,6 @@ jobs:
             echo "NIGHTLY_TESTS=TRUE";
           } >> "${GITHUB_ENV}"
 
-      - name: Check device is detected
-        if: ${{ !cancelled() }}
-        run: nvidia-smi
-
       - name: Run signed integer multi-bit tests
         run: |
           make test_signed_integer_multi_bit_gpu_ci
@@ -174,21 +150,28 @@ jobs:
     if: ${{ always() && needs.cuda-signed-integer-tests.result != 'skipped' && failure() }}
     continue-on-error: true
     steps:
+      - name: Set pull-request URL
+        if: env.SECRETS_AVAILABLE == 'true' && github.event_name == 'pull_request'
+        run: |
+          echo "PULL_REQUEST_MD_LINK=[pull-request](${{ vars.PR_BASE_URL }}${{ github.event.pull_request.number }}), "  >> "${GITHUB_ENV}"
+
       - name: Send message
+        if: env.SECRETS_AVAILABLE == 'true'
         uses: rtCamp/action-slack-notify@c33737706dea87cd7784c687dadc9adf1be59990
         env:
           SLACK_COLOR: ${{ needs.cuda-signed-integer-tests.result }}
-          SLACK_MESSAGE: "Base GPU tests finished with status: ${{ needs.cuda-signed-integer-tests.result }}. (${{ env.ACTION_RUN_URL }})"
+          SLACK_MESSAGE: "Signed GPU tests finished with status: ${{ needs.cuda-signed-integer-tests.result }}. (${{ env.PULL_REQUEST_MD_LINK }}[action run](${{ env.ACTION_RUN_URL }}))"
 
   teardown-instance:
     name: Teardown instance (cuda-tests)
-    if: ${{ always() && needs.setup-instance.result != 'skipped' }}
+    if: ${{ always() && needs.setup-instance.result == 'success' }}
     needs: [ setup-instance, cuda-signed-integer-tests ]
     runs-on: ubuntu-latest
     steps:
-      - name: Stop instance
+      - name: Stop remote instance
         id: stop-instance
-        uses: zama-ai/slab-github-runner@801df0b8db5ea2b06128b7476c652f5ed5f193a8
+        if: env.SECRETS_AVAILABLE == 'true'
+        uses: zama-ai/slab-github-runner@79939325c3c429837c10d6041e4fd8589d328bac
         with:
           mode: stop
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}

.github/workflows/gpu_unsigned_integer_classic_tests.yml

@@ -1,5 +1,5 @@
 # Test unsigned integers on an RTXA6000 VM on hyperstack with the classical PBS
-name: TFHE Cuda Backend - Unsigned integer tests with classical PBS
+name: Cuda - Unsigned integer tests with classical PBS
 
 env:
   CARGO_TERM_COLOR: always
@@ -11,19 +11,26 @@ env:
   SLACK_ICON: https://pbs.twimg.com/profile_images/1274014582265298945/OjBKP9kn_400x400.png
   SLACK_USERNAME: ${{ secrets.BOT_USERNAME }}
   SLACK_WEBHOOK: ${{ secrets.SLACK_WEBHOOK }}
+  SLACKIFY_MARKDOWN: true
   IS_PULL_REQUEST: ${{ github.event_name == 'pull_request' }}
+  PULL_REQUEST_MD_LINK: ""
+  CHECKOUT_TOKEN: ${{ secrets.REPO_CHECKOUT_TOKEN || secrets.GITHUB_TOKEN }}
+  # Secrets will be available only to zama-ai organization members
+  SECRETS_AVAILABLE: ${{ secrets.JOB_SECRET != '' }}
+  EXTERNAL_CONTRIBUTION_RUNNER: "gpu_ubuntu-22.04"
 
 on:
   # Allows you to run this workflow manually from the Actions tab as an alternative.
   workflow_dispatch:
   pull_request:
-      types: [ labeled ]
+    types: [ labeled ]
+
 
 jobs:
   should-run:
     runs-on: ubuntu-latest
     permissions:
-      pull-requests: write
+      pull-requests: read
     outputs:
       gpu_test: ${{ env.IS_PULL_REQUEST == 'false' || steps.changed-files.outputs.gpu_any_changed }}
     steps:
@@ -31,13 +38,13 @@ jobs:
         uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
         with:
           fetch-depth: 0
-          token: ${{ secrets.FHE_ACTIONS_TOKEN }}
+          persist-credentials: 'false'
+          token: ${{ env.CHECKOUT_TOKEN }}
 
       - name: Check for file changes
         id: changed-files
-        uses: tj-actions/changed-files@4edd678ac3f81e2dc578756871e4d00c19191daf
+        uses: tj-actions/changed-files@26a38635fc1173cc5820336ce97be6188d0de9f5 # v46.0.2
         with:
-          since_last_remote_commit: true
           files_yaml: |
             gpu:
               - tfhe/Cargo.toml
@@ -51,7 +58,7 @@ jobs:
               - tfhe/src/shortint/parameters/**
               - tfhe/src/high_level_api/**
               - tfhe/src/c_api/**
-              - 'tfhe/docs/**.md'
+              - 'tfhe/docs/**/**.md'
               - '.github/workflows/gpu_unsigned_integer_classic_tests.yml'
               - scripts/integer-tests.sh
               - ci/slab.toml
@@ -59,16 +66,17 @@ jobs:
   setup-instance:
     name: Setup instance (cuda-unsigned-classic-tests)
     needs: should-run
-    if: github.event_name != 'pull_request' ||
+    if: github.event_name == 'workflow_dispatch' ||
       (github.event.action != 'labeled' && needs.should-run.outputs.gpu_test == 'true') ||
       (github.event.action == 'labeled' && github.event.label.name == 'approved' && needs.should-run.outputs.gpu_test == 'true')
     runs-on: ubuntu-latest
     outputs:
-      runner-name: ${{ steps.start-instance.outputs.label }}
+      runner-name: ${{ steps.start-remote-instance.outputs.label || steps.start-github-instance.outputs.runner_group }}
     steps:
-      - name: Start instance
-        id: start-instance
-        uses: zama-ai/slab-github-runner@801df0b8db5ea2b06128b7476c652f5ed5f193a8
+      - name: Start remote instance
+        id: start-remote-instance
+        if: env.SECRETS_AVAILABLE == 'true'
+        uses: zama-ai/slab-github-runner@79939325c3c429837c10d6041e4fd8589d328bac
         with:
           mode: start
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}
@@ -77,13 +85,20 @@ jobs:
           backend: hyperstack
           profile: gpu-test
 
+      # This instance will be spawned especially for pull-request from forked repository
+      - name: Start GitHub instance
+        id: start-github-instance
+        if: env.SECRETS_AVAILABLE == 'false'
+        run: |
+          echo "runner_group=${{ env.EXTERNAL_CONTRIBUTION_RUNNER }}" >> "$GITHUB_OUTPUT"
+
   cuda-tests-linux:
     name: CUDA unsigned integer tests with classical PBS
     needs: [ should-run, setup-instance ]
     if: github.event_name != 'pull_request' ||
       (github.event_name == 'pull_request' && needs.setup-instance.result != 'skipped')
     concurrency:
-      group: ${{ github.workflow }}_${{ github.ref }}
+      group: ${{ github.workflow_ref }}
       cancel-in-progress: ${{ github.ref != 'refs/heads/main' }}
     runs-on: ${{ needs.setup-instance.outputs.runner-name }}
     strategy:
@@ -94,58 +109,25 @@ jobs:
           - os: ubuntu-22.04
             cuda: "12.2"
             gcc: 11 
-    env:
-      CUDA_PATH: /usr/local/cuda-${{ matrix.cuda }}
-      CMAKE_VERSION: 3.29.6
     steps:
-      # Mandatory on hyperstack since a bootable volume is not re-usable yet.
-      - name: Install dependencies
-        run: |
-          sudo apt update
-          sudo apt install -y checkinstall zlib1g-dev libssl-dev libclang-dev
-          wget https://github.com/Kitware/CMake/releases/download/v${{ env.CMAKE_VERSION }}/cmake-${{ env.CMAKE_VERSION }}.tar.gz
-          tar -zxvf cmake-${{ env.CMAKE_VERSION }}.tar.gz
-          cd cmake-${{ env.CMAKE_VERSION }}
-          ./bootstrap
-          make -j"$(nproc)"
-          sudo make install
-
-
       - name: Checkout tfhe-rs
         uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
+        with:
+          persist-credentials: 'false'
+          token: ${{ env.CHECKOUT_TOKEN }}
 
-      - name: Set up home
-        run: |
-          echo "HOME=/home/ubuntu" >> "${GITHUB_ENV}"
+      - name: Setup Hyperstack dependencies
+        uses: ./.github/actions/gpu_setup
+        with:
+          cuda-version: ${{ matrix.cuda }}
+          gcc-version: ${{ matrix.gcc }}
+          github-instance: ${{ env.SECRETS_AVAILABLE == 'false' }}
 
       - name: Install latest stable
-        uses: dtolnay/rust-toolchain@7b1c307e0dcbda6122208f10795a713336a9b35a
+        uses: dtolnay/rust-toolchain@888c2e1ea69ab0d4330cbf0af1ecc7b68f368cc1
         with:
           toolchain: stable
 
-      - name: Export CUDA variables
-        if: ${{ !cancelled() }}
-        run: |
-          echo "CUDA_PATH=$CUDA_PATH" >> "${GITHUB_ENV}"
-          echo "$CUDA_PATH/bin" >> "${GITHUB_PATH}"
-          echo "LD_LIBRARY_PATH=$CUDA_PATH/lib:$LD_LIBRARY_PATH" >> "${GITHUB_ENV}"
-          echo "CUDACXX=/usr/local/cuda-${{ matrix.cuda }}/bin/nvcc" >> "${GITHUB_ENV}"
-
-      # Specify the correct host compilers
-      - name: Export gcc and g++ variables
-        if: ${{ !cancelled() }}
-        run: |
-          {
-            echo "CC=/usr/bin/gcc-${{ matrix.gcc }}";
-            echo "CXX=/usr/bin/g++-${{ matrix.gcc }}";
-            echo "CUDAHOSTCXX=/usr/bin/g++-${{ matrix.gcc }}";
-            echo "HOME=/home/ubuntu";
-          } >> "${GITHUB_ENV}"
-
-      - name: Check device is detected
-        if: ${{ !cancelled() }}
-        run: nvidia-smi
-
       - name: Run unsigned integer tests
         run: |
           BIG_TESTS_INSTANCE=TRUE make test_unsigned_integer_gpu_ci
@@ -157,21 +139,28 @@ jobs:
     if: ${{ always() && needs.cuda-tests-linux.result != 'skipped' && failure() }}
     continue-on-error: true
     steps:
+      - name: Set pull-request URL
+        if: env.SECRETS_AVAILABLE == 'true' && github.event_name == 'pull_request'
+        run: |
+          echo "PULL_REQUEST_MD_LINK=[pull-request](${{ vars.PR_BASE_URL }}${{ github.event.pull_request.number }}), "  >> "${GITHUB_ENV}"
+
       - name: Send message
+        if: env.SECRETS_AVAILABLE == 'true'
         uses: rtCamp/action-slack-notify@c33737706dea87cd7784c687dadc9adf1be59990
         env:
           SLACK_COLOR: ${{ needs.cuda-tests-linux.result }}
-          SLACK_MESSAGE: "Unsigned integer GPU classic tests finished with status: ${{ needs.cuda-tests-linux.result }}. (${{ env.ACTION_RUN_URL }})"
+          SLACK_MESSAGE: "Unsigned integer GPU classic tests finished with status: ${{ needs.cuda-tests-linux.result }}. (${{ env.PULL_REQUEST_MD_LINK }}[action run](${{ env.ACTION_RUN_URL }}))"
 
   teardown-instance:
     name: Teardown instance (cuda-unsigned-classic-tests)
-    if: ${{ always() && needs.setup-instance.result != 'skipped' }}
+    if: ${{ always() && needs.setup-instance.result == 'success' }}
     needs: [ setup-instance, cuda-tests-linux ]
     runs-on: ubuntu-latest
     steps:
-      - name: Stop instance
+      - name: Stop remote instance
         id: stop-instance
-        uses: zama-ai/slab-github-runner@801df0b8db5ea2b06128b7476c652f5ed5f193a8
+        if: env.SECRETS_AVAILABLE == 'true'
+        uses: zama-ai/slab-github-runner@79939325c3c429837c10d6041e4fd8589d328bac
         with:
           mode: stop
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}

.github/workflows/gpu_unsigned_integer_h100_tests.yml

@@ -1,5 +1,5 @@
 # Test unsigned integers on an H100 VM on hyperstack
-name: TFHE Cuda Backend - Unsigned integer tests on H100
+name: Cuda - Unsigned integer tests on H100
 
 env:
   CARGO_TERM_COLOR: always
@@ -11,19 +11,25 @@ env:
   SLACK_ICON: https://pbs.twimg.com/profile_images/1274014582265298945/OjBKP9kn_400x400.png
   SLACK_USERNAME: ${{ secrets.BOT_USERNAME }}
   SLACK_WEBHOOK: ${{ secrets.SLACK_WEBHOOK }}
+  SLACKIFY_MARKDOWN: true
   IS_PULL_REQUEST: ${{ github.event_name == 'pull_request' }}
+  PULL_REQUEST_MD_LINK: ""
+  CHECKOUT_TOKEN: ${{ secrets.REPO_CHECKOUT_TOKEN || secrets.GITHUB_TOKEN }}
+  # Secrets will be available only to zama-ai organization members
+  SECRETS_AVAILABLE: ${{ secrets.JOB_SECRET != '' }}
+  EXTERNAL_CONTRIBUTION_RUNNER: "gpu_ubuntu-22.04"
 
 on:
   # Allows you to run this workflow manually from the Actions tab as an alternative.
   workflow_dispatch:
   pull_request:
-      types: [ labeled ]
+    types: [ labeled ]
 
 jobs:
   should-run:
     runs-on: ubuntu-latest
     permissions:
-      pull-requests: write
+      pull-requests: read
     outputs:
       gpu_test: ${{ env.IS_PULL_REQUEST == 'false' || steps.changed-files.outputs.gpu_any_changed }}
     steps:
@@ -31,13 +37,13 @@ jobs:
         uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
         with:
           fetch-depth: 0
-          token: ${{ secrets.FHE_ACTIONS_TOKEN }}
+          persist-credentials: 'false'
+          token: ${{ env.CHECKOUT_TOKEN }}
 
       - name: Check for file changes
         id: changed-files
-        uses: tj-actions/changed-files@4edd678ac3f81e2dc578756871e4d00c19191daf
+        uses: tj-actions/changed-files@26a38635fc1173cc5820336ce97be6188d0de9f5 # v46.0.2
         with:
-          since_last_remote_commit: true
           files_yaml: |
             gpu:
               - tfhe/Cargo.toml
@@ -51,7 +57,7 @@ jobs:
               - tfhe/src/shortint/parameters/**
               - tfhe/src/high_level_api/**
               - tfhe/src/c_api/**
-              - 'tfhe/docs/**.md'
+              - 'tfhe/docs/**/**.md'
               - '.github/workflows/gpu_unsigned_integer_h100_tests.yml'
               - scripts/integer-tests.sh
               - ci/slab.toml
@@ -59,16 +65,23 @@ jobs:
   setup-instance:
     name: Setup instance (cuda-h100-tests)
     needs: should-run
-    if: github.event_name != 'pull_request' ||
+    if: github.event_name == 'workflow_dispatch' ||
       (github.event.action != 'labeled' && needs.should-run.outputs.gpu_test == 'true') ||
       (github.event.action == 'labeled' && github.event.label.name == 'approved' && needs.should-run.outputs.gpu_test == 'true')
     runs-on: ubuntu-latest
     outputs:
-      runner-name: ${{ steps.start-instance.outputs.label }}
+      # Use permanent remote instance label first as on-demand remote instance label output is set before the end of start-remote-instance step.
+      # If the latter fails due to a failed GitHub action runner set up, we have to fallback on the permanent instance.
+      # Since the on-demand remote label is set before failure, we have to do the logical OR in this order,
+      # otherwise we'll try to run the next job on a non-existing on-demand instance.
+      runner-name: ${{ steps.use-permanent-instance.outputs.runner_group || steps.start-remote-instance.outputs.label || steps.start-github-instance.outputs.runner_group }}
+      remote-instance-outcome: ${{ steps.start-remote-instance.outcome }}
     steps:
-      - name: Start instance
-        id: start-instance
-        uses: zama-ai/slab-github-runner@801df0b8db5ea2b06128b7476c652f5ed5f193a8
+      - name: Start remote instance
+        id: start-remote-instance
+        if: env.SECRETS_AVAILABLE == 'true'
+        continue-on-error: true
+        uses: zama-ai/slab-github-runner@79939325c3c429837c10d6041e4fd8589d328bac
         with:
           mode: start
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}
@@ -77,13 +90,27 @@ jobs:
           backend: hyperstack
           profile: single-h100
 
+      # This will allow to fallback on permanent instances running on Hyperstack.
+      - name: Use permanent remote instance
+        id: use-permanent-instance
+        if: env.SECRETS_AVAILABLE == 'true' && steps.start-remote-instance.outcome == 'failure'
+        run: |
+          echo "runner_group=h100x1" >> "$GITHUB_OUTPUT"
+
+      # This instance will be spawned especially for pull-request from forked repository
+      - name: Start GitHub instance
+        id: start-github-instance
+        if: env.SECRETS_AVAILABLE == 'false'
+        run: |
+          echo "runner_group=${{ env.EXTERNAL_CONTRIBUTION_RUNNER }}" >> "$GITHUB_OUTPUT"
+
   cuda-tests-linux:
     name: CUDA H100 unsigned integer tests
     needs: [ should-run, setup-instance ]
     if: github.event_name != 'pull_request' ||
       (github.event_name == 'pull_request' && needs.setup-instance.result != 'skipped')
     concurrency:
-      group: ${{ github.workflow }}_${{ github.ref }}
+      group: ${{ github.workflow_ref }}
       cancel-in-progress: ${{ github.ref != 'refs/heads/main' }}
     runs-on: ${{ needs.setup-instance.outputs.runner-name }}
     strategy:
@@ -94,58 +121,26 @@ jobs:
           - os: ubuntu-22.04
             cuda: "12.2"
             gcc: 11 
-    env:
-      CUDA_PATH: /usr/local/cuda-${{ matrix.cuda }}
-      CMAKE_VERSION: 3.29.6
     steps:
-      # Mandatory on hyperstack since a bootable volume is not re-usable yet.
-      - name: Install dependencies
-        run: |
-          sudo apt update
-          sudo apt install -y checkinstall zlib1g-dev libssl-dev libclang-dev
-          wget https://github.com/Kitware/CMake/releases/download/v${{ env.CMAKE_VERSION }}/cmake-${{ env.CMAKE_VERSION }}.tar.gz
-          tar -zxvf cmake-${{ env.CMAKE_VERSION }}.tar.gz
-          cd cmake-${{ env.CMAKE_VERSION }}
-          ./bootstrap
-          make -j"$(nproc)"
-          sudo make install
-
-
       - name: Checkout tfhe-rs
         uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
+        with:
+          persist-credentials: 'false'
+          token: ${{ env.CHECKOUT_TOKEN }}
 
-      - name: Set up home
-        run: |
-          echo "HOME=/home/ubuntu" >> "${GITHUB_ENV}"
+      - name: Setup Hyperstack dependencies
+        if: needs.setup-instance.outputs.remote-instance-outcome == 'success'
+        uses: ./.github/actions/gpu_setup
+        with:
+          cuda-version: ${{ matrix.cuda }}
+          gcc-version: ${{ matrix.gcc }}
+          github-instance: ${{ env.SECRETS_AVAILABLE == 'false' }}
 
       - name: Install latest stable
-        uses: dtolnay/rust-toolchain@7b1c307e0dcbda6122208f10795a713336a9b35a
+        uses: dtolnay/rust-toolchain@888c2e1ea69ab0d4330cbf0af1ecc7b68f368cc1
         with:
           toolchain: stable
 
-      - name: Export CUDA variables
-        if: ${{ !cancelled() }}
-        run: |
-          echo "CUDA_PATH=$CUDA_PATH" >> "${GITHUB_ENV}"
-          echo "$CUDA_PATH/bin" >> "${GITHUB_PATH}"
-          echo "LD_LIBRARY_PATH=$CUDA_PATH/lib:$LD_LIBRARY_PATH" >> "${GITHUB_ENV}"
-          echo "CUDACXX=/usr/local/cuda-${{ matrix.cuda }}/bin/nvcc" >> "${GITHUB_ENV}"
-
-      # Specify the correct host compilers
-      - name: Export gcc and g++ variables
-        if: ${{ !cancelled() }}
-        run: |
-          {
-            echo "CC=/usr/bin/gcc-${{ matrix.gcc }}";
-            echo "CXX=/usr/bin/g++-${{ matrix.gcc }}";
-            echo "CUDAHOSTCXX=/usr/bin/g++-${{ matrix.gcc }}";
-            echo "HOME=/home/ubuntu";
-          } >> "${GITHUB_ENV}"
-
-      - name: Check device is detected
-        if: ${{ !cancelled() }}
-        run: nvidia-smi
-
       - name: Run unsigned integer multi-bit tests
         run: |
           BIG_TESTS_INSTANCE=TRUE make test_unsigned_integer_multi_bit_gpu_ci
@@ -157,21 +152,28 @@ jobs:
     if: ${{ always() && needs.cuda-tests-linux.result != 'skipped' && failure() }}
     continue-on-error: true
     steps:
+      - name: Set pull-request URL
+        if: env.SECRETS_AVAILABLE == 'true' && github.event_name == 'pull_request'
+        run: |
+          echo "PULL_REQUEST_MD_LINK=[pull-request](${{ vars.PR_BASE_URL }}${{ github.event.pull_request.number }}), "  >> "${GITHUB_ENV}"
+
       - name: Send message
+        if: env.SECRETS_AVAILABLE == 'true'
         uses: rtCamp/action-slack-notify@c33737706dea87cd7784c687dadc9adf1be59990
         env:
           SLACK_COLOR: ${{ needs.cuda-tests-linux.result }}
-          SLACK_MESSAGE: "Unsigned integer GPU H100 tests finished with status: ${{ needs.cuda-tests-linux.result }}. (${{ env.ACTION_RUN_URL }})"
+          SLACK_MESSAGE: "Unsigned integer GPU H100 tests finished with status: ${{ needs.cuda-tests-linux.result }}. (${{ env.PULL_REQUEST_MD_LINK }}[action run](${{ env.ACTION_RUN_URL }}))"
 
   teardown-instance:
     name: Teardown instance (cuda-h100-tests)
-    if: ${{ always() && needs.setup-instance.result != 'skipped' }}
+    if: ${{ always() && needs.setup-instance.outputs.remote-instance-outcome == 'success' }}
     needs: [ setup-instance, cuda-tests-linux ]
     runs-on: ubuntu-latest
     steps:
-      - name: Stop instance
+      - name: Stop remote instance
         id: stop-instance
-        uses: zama-ai/slab-github-runner@801df0b8db5ea2b06128b7476c652f5ed5f193a8
+        if: env.SECRETS_AVAILABLE == 'true'
+        uses: zama-ai/slab-github-runner@79939325c3c429837c10d6041e4fd8589d328bac
         with:
           mode: stop
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}

.github/workflows/gpu_unsigned_integer_tests.yml

@@ -1,5 +1,5 @@
 # Compile and test tfhe-cuda-backend unsigned integer on an AWS instance
-name: TFHE Cuda Backend - Unsigned integer tests
+name: Cuda - Unsigned integer tests
 
 env:
   CARGO_TERM_COLOR: always
@@ -11,16 +11,20 @@ env:
   SLACK_ICON: https://pbs.twimg.com/profile_images/1274014582265298945/OjBKP9kn_400x400.png
   SLACK_USERNAME: ${{ secrets.BOT_USERNAME }}
   SLACK_WEBHOOK: ${{ secrets.SLACK_WEBHOOK }}
+  SLACKIFY_MARKDOWN: true
   FAST_TESTS: TRUE
   NIGHTLY_TESTS: FALSE
+  IS_PULL_REQUEST: ${{ github.event_name == 'pull_request' }}
+  PULL_REQUEST_MD_LINK: ""
+  CHECKOUT_TOKEN: ${{ secrets.REPO_CHECKOUT_TOKEN || secrets.GITHUB_TOKEN }}
+  # Secrets will be available only to zama-ai organization members
+  SECRETS_AVAILABLE: ${{ secrets.JOB_SECRET != '' }}
+  EXTERNAL_CONTRIBUTION_RUNNER: "gpu_ubuntu-22.04"
 
 on:
   # Allows you to run this workflow manually from the Actions tab as an alternative.
   workflow_dispatch:
   pull_request:
-    types:
-      - opened
-      - synchronize
   schedule:
     # Nightly tests @ 1AM after each work day
     - cron: "0 1 * * MON-FRI"
@@ -29,7 +33,7 @@ jobs:
   should-run:
     runs-on: ubuntu-latest
     permissions:
-      pull-requests: write
+      pull-requests: read
     outputs:
       gpu_test: ${{ env.IS_PULL_REQUEST == 'false' || steps.changed-files.outputs.gpu_any_changed }}
     steps:
@@ -37,13 +41,13 @@ jobs:
         uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
         with:
           fetch-depth: 0
-          token: ${{ secrets.FHE_ACTIONS_TOKEN }}
+          persist-credentials: 'false'
+          token: ${{ env.CHECKOUT_TOKEN }}
 
       - name: Check for file changes
         id: changed-files
-        uses: tj-actions/changed-files@4edd678ac3f81e2dc578756871e4d00c19191daf
+        uses: tj-actions/changed-files@26a38635fc1173cc5820336ce97be6188d0de9f5 # v46.0.2
         with:
-          since_last_remote_commit: true
           files_yaml: |
             gpu:
               - tfhe/Cargo.toml
@@ -57,24 +61,25 @@ jobs:
               - tfhe/src/shortint/parameters/**
               - tfhe/src/high_level_api/**
               - tfhe/src/c_api/**
-              - 'tfhe/docs/**.md'
+              - 'tfhe/docs/**/**.md'
               - '.github/workflows/gpu_unsigned_integer_tests.yml'
               - scripts/integer-tests.sh
               - ci/slab.toml
 
   setup-instance:
     name: Setup instance (cuda-unsigned-integer-tests)
+    runs-on: ubuntu-latest
     needs: should-run
     if: (github.event_name == 'schedule' && github.repository == 'zama-ai/tfhe-rs') ||
       github.event_name == 'workflow_dispatch' ||
       needs.should-run.outputs.gpu_test == 'true'
-    runs-on: ubuntu-latest
     outputs:
-      runner-name: ${{ steps.start-instance.outputs.label }}
+      runner-name: ${{ steps.start-remote-instance.outputs.label || steps.start-github-instance.outputs.runner_group }}
     steps:
-      - name: Start instance
-        id: start-instance
-        uses: zama-ai/slab-github-runner@801df0b8db5ea2b06128b7476c652f5ed5f193a8
+      - name: Start remote instance
+        id: start-remote-instance
+        if: env.SECRETS_AVAILABLE == 'true'
+        uses: zama-ai/slab-github-runner@79939325c3c429837c10d6041e4fd8589d328bac
         with:
           mode: start
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}
@@ -83,13 +88,20 @@ jobs:
           backend: hyperstack
           profile: gpu-test
 
+      # This instance will be spawned especially for pull-request from forked repository
+      - name: Start GitHub instance
+        id: start-github-instance
+        if: env.SECRETS_AVAILABLE == 'false'
+        run: |
+          echo "runner_group=${{ env.EXTERNAL_CONTRIBUTION_RUNNER }}" >> "$GITHUB_OUTPUT"
+
   cuda-unsigned-integer-tests:
     name: CUDA unsigned integer tests
     needs: [ should-run, setup-instance ]
     if: github.event_name != 'pull_request' ||
       (github.event_name == 'pull_request' && needs.setup-instance.result != 'skipped')
     concurrency:
-      group: ${{ github.workflow }}_${{ github.ref }}
+      group: ${{ github.workflow_ref }}
       cancel-in-progress: ${{ github.ref != 'refs/heads/main' }}
     runs-on: ${{ needs.setup-instance.outputs.runner-name }}
     strategy:
@@ -100,54 +112,25 @@ jobs:
           - os: ubuntu-22.04
             cuda: "12.2"
             gcc: 11
-    env:
-      CUDA_PATH: /usr/local/cuda-${{ matrix.cuda }}
-      CMAKE_VERSION: 3.29.6
     steps:
-      # Mandatory on hyperstack since a bootable volume is not re-usable yet.
-      - name: Install dependencies
-        run: |
-          sudo apt update
-          sudo apt install -y checkinstall zlib1g-dev libssl-dev libclang-dev
-          wget https://github.com/Kitware/CMake/releases/download/v${{ env.CMAKE_VERSION }}/cmake-${{ env.CMAKE_VERSION }}.tar.gz
-          tar -zxvf cmake-${{ env.CMAKE_VERSION }}.tar.gz
-          cd cmake-${{ env.CMAKE_VERSION }}
-          ./bootstrap
-          make -j"$(nproc)"
-          sudo make install
-
-
       - name: Checkout tfhe-rs
         uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
+        with:
+          persist-credentials: 'false'
+          token: ${{ env.CHECKOUT_TOKEN }}
 
-      - name: Set up home
-        run: |
-          echo "HOME=/home/ubuntu" >> "${GITHUB_ENV}"
+      - name: Setup Hyperstack dependencies
+        uses: ./.github/actions/gpu_setup
+        with:
+          cuda-version: ${{ matrix.cuda }}
+          gcc-version: ${{ matrix.gcc }}
+          github-instance: ${{ env.SECRETS_AVAILABLE == 'false' }}
 
       - name: Install latest stable
-        uses: dtolnay/rust-toolchain@7b1c307e0dcbda6122208f10795a713336a9b35a
+        uses: dtolnay/rust-toolchain@888c2e1ea69ab0d4330cbf0af1ecc7b68f368cc1
         with:
           toolchain: stable
 
-      - name: Export CUDA variables
-        if: ${{ !cancelled() }}
-        run: |
-          echo "CUDA_PATH=$CUDA_PATH" >> "${GITHUB_ENV}"
-          echo "$CUDA_PATH/bin" >> "${GITHUB_PATH}"
-          echo "LD_LIBRARY_PATH=$CUDA_PATH/lib:$LD_LIBRARY_PATH" >> "${GITHUB_ENV}"
-          echo "CUDACXX=/usr/local/cuda-${{ matrix.cuda }}/bin/nvcc" >> "${GITHUB_ENV}"
-
-      # Specify the correct host compilers
-      - name: Export gcc and g++ variables
-        if: ${{ !cancelled() }}
-        run: |
-          {
-            echo "CC=/usr/bin/gcc-${{ matrix.gcc }}";
-            echo "CXX=/usr/bin/g++-${{ matrix.gcc }}";
-            echo "CUDAHOSTCXX=/usr/bin/g++-${{ matrix.gcc }}";
-            echo "HOME=/home/ubuntu";
-          } >> "${GITHUB_ENV}"
-
       - name: Should run nightly tests
         if: github.event_name == 'schedule'
         run: |
@@ -156,10 +139,6 @@ jobs:
             echo "NIGHTLY_TESTS=TRUE";
           } >> "${GITHUB_ENV}"
 
-      - name: Check device is detected
-        if: ${{ !cancelled() }}
-        run: nvidia-smi
-
       - name: Run unsigned integer multi-bit tests
         run: |
           make test_unsigned_integer_multi_bit_gpu_ci
@@ -171,21 +150,28 @@ jobs:
     if: ${{ always() && needs.cuda-unsigned-integer-tests.result != 'skipped' && failure() }}
     continue-on-error: true
     steps:
+      - name: Set pull-request URL
+        if: env.SECRETS_AVAILABLE == 'true' && github.event_name == 'pull_request'
+        run: |
+          echo "PULL_REQUEST_MD_LINK=[pull-request](${{ vars.PR_BASE_URL }}${{ github.event.pull_request.number }}), "  >> "${GITHUB_ENV}"
+
       - name: Send message
+        if: env.SECRETS_AVAILABLE == 'true'
         uses: rtCamp/action-slack-notify@c33737706dea87cd7784c687dadc9adf1be59990
         env:
           SLACK_COLOR: ${{ needs.cuda-unsigned-integer-tests.result }}
-          SLACK_MESSAGE: "Unsigned integer GPU tests finished with status: ${{ needs.cuda-unsigned-integer-tests.result }}. (${{ env.ACTION_RUN_URL }})"
+          SLACK_MESSAGE: "Unsigned integer GPU tests finished with status: ${{ needs.cuda-unsigned-integer-tests.result }}. (${{ env.PULL_REQUEST_MD_LINK }}[action run](${{ env.ACTION_RUN_URL }}))"
 
   teardown-instance:
     name: Teardown instance (cuda-tests)
-    if: ${{ always() && needs.setup-instance.result != 'skipped' }}
+    if: ${{ always() && needs.setup-instance.result == 'success' }}
     needs: [ setup-instance, cuda-unsigned-integer-tests ]
     runs-on: ubuntu-latest
     steps:
       - name: Stop instance
         id: stop-instance
-        uses: zama-ai/slab-github-runner@801df0b8db5ea2b06128b7476c652f5ed5f193a8
+        if: env.SECRETS_AVAILABLE == 'true'
+        uses: zama-ai/slab-github-runner@79939325c3c429837c10d6041e4fd8589d328bac
         with:
           mode: stop
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}

.github/workflows/integer_long_run_tests.yml

@@ -0,0 +1,95 @@
+name: AWS Long Run Tests on CPU
+
+env:
+  CARGO_TERM_COLOR: always
+  ACTION_RUN_URL: ${{ github.server_url }}/${{ github.repository }}/actions/runs/${{ github.run_id }}
+  RUSTFLAGS: "-C target-cpu=native"
+  RUST_BACKTRACE: "full"
+  RUST_MIN_STACK: "8388608"
+  SLACK_CHANNEL: ${{ secrets.SLACK_CHANNEL }}
+  SLACK_ICON: https://pbs.twimg.com/profile_images/1274014582265298945/OjBKP9kn_400x400.png
+  SLACK_USERNAME: ${{ secrets.BOT_USERNAME }}
+  SLACK_WEBHOOK: ${{ secrets.SLACK_WEBHOOK }}
+
+on:
+  # Allows you to run this workflow manually from the Actions tab as an alternative.
+  workflow_dispatch:
+  schedule:
+    # Weekly tests will be triggered each Friday at 9p.m.
+    - cron: "0 21 * * 5"
+
+jobs:
+  setup-instance:
+    name: Setup instance (cpu-tests)
+    if: github.event_name != 'schedule' ||
+      (github.event_name == 'schedule' && github.repository == 'zama-ai/tfhe-rs')
+    runs-on: ubuntu-latest
+    outputs:
+      runner-name: ${{ steps.start-instance.outputs.label }}
+    steps:
+      - name: Start instance
+        id: start-instance
+        uses: zama-ai/slab-github-runner@79939325c3c429837c10d6041e4fd8589d328bac
+        with:
+          mode: start
+          github-token: ${{ secrets.SLAB_ACTION_TOKEN }}
+          slab-url: ${{ secrets.SLAB_BASE_URL }}
+          job-secret: ${{ secrets.JOB_SECRET }}
+          backend: aws
+          profile: cpu-big
+
+  cpu-tests:
+    name: Long run CPU tests
+    needs: [ setup-instance ]
+    concurrency:
+      group: ${{ github.workflow_ref }}_${{github.event_name}}
+      cancel-in-progress: true
+    runs-on: ${{ needs.setup-instance.outputs.runner-name }}
+    timeout-minutes: 4320 # 72 hours
+    steps:
+      - name: Checkout tfhe-rs
+        uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
+        with:
+          persist-credentials: 'false'
+          token: ${{ secrets.REPO_CHECKOUT_TOKEN }}
+
+      - name: Install latest stable
+        uses: dtolnay/rust-toolchain@888c2e1ea69ab0d4330cbf0af1ecc7b68f368cc1
+        with:
+          toolchain: stable
+
+      - name: Run tests
+        run: |
+          make test_integer_long_run
+
+      - name: Slack Notification
+        if: ${{ failure() }}
+        continue-on-error: true
+        uses: rtCamp/action-slack-notify@c33737706dea87cd7784c687dadc9adf1be59990
+        env:
+          SLACK_COLOR: ${{ job.status }}
+          SLACK_MESSAGE: "CPU long run tests finished with status: ${{ job.status }}. (${{ env.ACTION_RUN_URL }})"
+
+  teardown-instance:
+    name: Teardown instance (cpu-tests)
+    if: ${{ always() && needs.setup-instance.result == 'success' }}
+    needs: [ setup-instance, cpu-tests ]
+    runs-on: ubuntu-latest
+    steps:
+      - name: Stop instance
+        id: stop-instance
+        uses: zama-ai/slab-github-runner@79939325c3c429837c10d6041e4fd8589d328bac
+        with:
+          mode: stop
+          github-token: ${{ secrets.SLAB_ACTION_TOKEN }}
+          slab-url: ${{ secrets.SLAB_BASE_URL }}
+          job-secret: ${{ secrets.JOB_SECRET }}
+          label: ${{ needs.setup-instance.outputs.runner-name }}
+
+      - name: Slack Notification
+        if: ${{ failure() }}
+        continue-on-error: true
+        uses: rtCamp/action-slack-notify@c33737706dea87cd7784c687dadc9adf1be59990
+        env:
+          SLACK_COLOR: ${{ job.status }}
+          SLACK_MESSAGE: "Instance teardown (cpu-long-run-tests) finished with status: ${{ job.status }}. (${{ env.ACTION_RUN_URL }})"

.github/workflows/m1_tests.yml

@@ -3,7 +3,7 @@ name: Tests on M1 CPU
 on:
   workflow_dispatch:
   pull_request:
-    types: [labeled]
+    types: [ labeled ]
   # Have a nightly build for M1 tests
   schedule:
     # * is a special character in YAML so you have to quote this string
@@ -21,14 +21,17 @@ env:
   # We clear the cache to reduce memory pressure because of the numerous processes of cargo
   # nextest
   TFHE_RS_CLEAR_IN_MEMORY_KEY_CACHE: "1"
+  CHECKOUT_TOKEN: ${{ secrets.REPO_CHECKOUT_TOKEN || secrets.GITHUB_TOKEN }}
 
 concurrency:
-  group: ${{ github.workflow }}-${{ github.head_ref }}
+  group: ${{ github.workflow_ref }}
   cancel-in-progress: true
 
 jobs:
   cargo-builds-m1:
-    if: ${{ (github.event_name == 'schedule' &&  github.repository == 'zama-ai/tfhe-rs') || github.event_name == 'workflow_dispatch' || contains(github.event.label.name, 'm1_test') }}
+    if: ${{ (github.event_name == 'schedule' &&  github.repository == 'zama-ai/tfhe-rs') ||
+      github.event_name == 'workflow_dispatch' ||
+      contains(github.event.label.name, 'm1_test') }}
     runs-on: ["self-hosted", "m1mac"]
     # 12 hours, default is 6 hours, hopefully this is more than enough
     timeout-minutes: 720
@@ -37,9 +40,10 @@ jobs:
       - uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
         with:
           persist-credentials: "false"
+          token: ${{ env.CHECKOUT_TOKEN }}
 
       - name: Install latest stable
-        uses: dtolnay/rust-toolchain@7b1c307e0dcbda6122208f10795a713336a9b35a
+        uses: dtolnay/rust-toolchain@888c2e1ea69ab0d4330cbf0af1ecc7b68f368cc1
         with:
           toolchain: stable
 
@@ -194,3 +198,5 @@ jobs:
           SLACK_MESSAGE: "M1 tests finished with status: ${{ needs.cargo-builds-m1.result }}. (${{ env.ACTION_RUN_URL }})"
           SLACK_USERNAME: ${{ secrets.BOT_USERNAME }}
           SLACK_WEBHOOK: ${{ secrets.SLACK_WEBHOOK }}
+          MSG_MINIMAL: event,action url,commit
+          BRANCH: ${{ github.ref }}

.github/workflows/make_release.yml

@@ -43,13 +43,15 @@ jobs:
       hash: ${{ steps.hash.outputs.hash }}
     steps:
       - name: Checkout
-        uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
+        uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683 # v4.2.2
         with:
           fetch-depth: 0
+          persist-credentials: 'false'
+          token: ${{ secrets.REPO_CHECKOUT_TOKEN }}
       - name: Prepare package
         run: |
           cargo package -p tfhe
-      - uses: actions/upload-artifact@b4b15b8c7c6ac21ea08fcf65892d2ee8f75cf882 # v4.4.3
+      - uses: actions/upload-artifact@ea165f8d65b6e75b540449e92b4886f43607fa02 # v4.6.2
         with:
           name: crate
           path: target/package/*.crate
@@ -60,7 +62,7 @@ jobs:
   provenance:
     if: ${{ !inputs.dry_run  }}
     needs: [package]
-    uses: slsa-framework/slsa-github-generator/.github/workflows/generator_generic_slsa3.yml@v2.0.0
+    uses: slsa-framework/slsa-github-generator/.github/workflows/generator_generic_slsa3.yml@v2.1.0
     permissions:
       # Needed to detect the GitHub Actions environment
       actions: read
@@ -76,20 +78,23 @@ jobs:
     name: Publish Release
     needs: [package] # for comparing hashes
     runs-on: ubuntu-latest
+    # For provenance of npmjs publish
     permissions:
       contents: read
       id-token: write
     steps:
       - name: Checkout
-        uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
+        uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683 # v4.2.2
         with:
           fetch-depth: 0
+          persist-credentials: 'false'
+          token: ${{ secrets.REPO_CHECKOUT_TOKEN }}
       - name: Create NPM version tag
         if: ${{ inputs.npm_latest_tag }}
         run: |
           echo "NPM_TAG=latest" >> "${GITHUB_ENV}"
       - name: Download artifact
-        uses: actions/download-artifact@fa0a91b85d4f404e444e00e005971372dc801d16 # v4.1.8
+        uses: actions/download-artifact@95815c38cf2ff2164869cbab79da8d1f422bc89e # v4.2.1
         with:
           name: crate
           path: target/package
@@ -108,7 +113,7 @@ jobs:
       - name: Slack notification (hashes comparison)
         if: ${{ needs.package.outputs.hash != steps.published_hash.outputs.pub_hash }}
         continue-on-error: true
-        uses: rtCamp/action-slack-notify@c33737706dea87cd7784c687dadc9adf1be59990
+        uses: rtCamp/action-slack-notify@c33737706dea87cd7784c687dadc9adf1be59990 # v2.3.2
         env:
           SLACK_COLOR: failure
           SLACK_CHANNEL: ${{ secrets.SLACK_CHANNEL }}
@@ -153,7 +158,7 @@ jobs:
       - name: Slack Notification
         if: ${{ failure() }}
         continue-on-error: true
-        uses: rtCamp/action-slack-notify@c33737706dea87cd7784c687dadc9adf1be59990
+        uses: rtCamp/action-slack-notify@c33737706dea87cd7784c687dadc9adf1be59990 # v2.3.2
         env:
           SLACK_COLOR: ${{ job.status }}
           SLACK_CHANNEL: ${{ secrets.SLACK_CHANNEL }}

.github/workflows/make_release_concrete_csprng.yml

@@ -1,48 +0,0 @@
-name: Publish tfhe-csprng release
-
-on:
-  workflow_dispatch:
-    inputs:
-      dry_run:
-        description: "Dry-run"
-        type: boolean
-        default: true
-
-env:
-  ACTION_RUN_URL: ${{ github.server_url }}/${{ github.repository }}/actions/runs/${{ github.run_id }}
-
-jobs:
-  verify_tag:
-    uses: ./.github/workflows/verify_tagged_commit.yml
-    secrets:
-      RELEASE_TEAM: ${{ secrets.RELEASE_TEAM }}
-      READ_ORG_TOKEN: ${{ secrets.READ_ORG_TOKEN }}
-
-  publish_release:
-    name: Publish tfhe-csprng Release
-    needs: verify_tag
-    runs-on: ubuntu-latest
-    steps:
-      - name: Checkout
-        uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
-        with:
-          fetch-depth: 0
-
-      - name: Publish crate.io package
-        env:
-          CRATES_TOKEN: ${{ secrets.CARGO_REGISTRY_TOKEN }}
-          DRY_RUN: ${{ inputs.dry_run && '--dry-run' || '' }}
-        run: |
-          cargo publish -p tfhe-csprng --token ${{ env.CRATES_TOKEN }} ${{ env.DRY_RUN }}
-
-      - name: Slack Notification
-        if: ${{ failure() }}
-        continue-on-error: true
-        uses: rtCamp/action-slack-notify@c33737706dea87cd7784c687dadc9adf1be59990
-        env:
-          SLACK_COLOR: ${{ job.status }}
-          SLACK_CHANNEL: ${{ secrets.SLACK_CHANNEL }}
-          SLACK_ICON: https://pbs.twimg.com/profile_images/1274014582265298945/OjBKP9kn_400x400.png
-          SLACK_MESSAGE: "tfhe-csprng release finished with status: ${{ job.status }}. (${{ env.ACTION_RUN_URL }})"
-          SLACK_USERNAME: ${{ secrets.BOT_USERNAME }}
-          SLACK_WEBHOOK: ${{ secrets.SLACK_WEBHOOK }}

.github/workflows/make_release_cuda.yml

@@ -1,4 +1,3 @@
-# Publish new release of tfhe-cuda-backend on crates.io.
 name: Publish CUDA release
 
 on:
@@ -8,10 +7,6 @@ on:
         description: "Dry-run"
         type: boolean
         default: true
-      push_to_crates:
-        description: "Push to crate"
-        type: boolean
-        default: true
 
 env:
   ACTION_RUN_URL: ${{ github.server_url }}/${{ github.repository }}/actions/runs/${{ github.run_id }}
@@ -36,7 +31,7 @@ jobs:
     steps:
       - name: Start instance
         id: start-instance
-        uses: zama-ai/slab-github-runner@801df0b8db5ea2b06128b7476c652f5ed5f193a8
+        uses: zama-ai/slab-github-runner@79939325c3c429837c10d6041e4fd8589d328bac
         with:
           mode: start
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}
@@ -45,10 +40,12 @@ jobs:
           backend: aws
           profile: gpu-build
 
-  publish-cuda-release:
-    name: Publish CUDA Release
+  package:
+    name: Package CUDA Release for provenance
     needs: setup-instance
     runs-on: ${{ needs.setup-instance.outputs.runner-name }}
+    outputs:
+      hash: ${{ steps.hash.outputs.hash }}
     strategy:
       fail-fast: false
       # explicit include-based build matrix, of known valid options
@@ -61,16 +58,76 @@ jobs:
       CUDA_PATH: /usr/local/cuda-${{ matrix.cuda }}
     steps:
       - name: Checkout
-        uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
+        uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683 # v4.2.2
         with:
           fetch-depth: 0
-
-      - name: Set up home
-        run: |
-          echo "HOME=/home/ubuntu" >> "${GITHUB_ENV}"
+          persist-credentials: "false"
+          token: ${{ secrets.REPO_CHECKOUT_TOKEN }}
 
       - name: Install latest stable
-        uses: dtolnay/rust-toolchain@7b1c307e0dcbda6122208f10795a713336a9b35a
+        uses: dtolnay/rust-toolchain@888c2e1ea69ab0d4330cbf0af1ecc7b68f368cc1
+        with:
+          toolchain: stable
+
+      - name: Export CUDA variables
+        if: ${{ !cancelled() }}
+        run: |
+          echo "$CUDA_PATH/bin" >> "${GITHUB_PATH}"
+          {
+            echo "CUDA_PATH=$CUDA_PATH";
+            echo "LD_LIBRARY_PATH=$CUDA_PATH/lib:$LD_LIBRARY_PATH";
+            echo "CUDACXX=/usr/local/cuda-${{ matrix.cuda }}/bin/nvcc";
+          } >> "${GITHUB_ENV}"
+
+      # Specify the correct host compilers
+      - name: Export gcc and g++ variables
+        if: ${{ !cancelled() }}
+        run: |
+          {
+            echo "CC=/usr/bin/gcc-${{ matrix.gcc }}";
+            echo "CXX=/usr/bin/g++-${{ matrix.gcc }}";
+            echo "CUDAHOSTCXX=/usr/bin/g++-${{ matrix.gcc }}";
+            echo "HOME=/home/ubuntu";
+          } >> "${GITHUB_ENV}"
+      - name: Prepare package
+        run: |
+          cargo package -p tfhe-cuda-backend
+      - name: generate hash
+        id: hash
+        run: cd target/package && echo "hash=$(sha256sum ./*.crate | base64 -w0)" >> "${GITHUB_OUTPUT}"
+
+  provenance:
+    if: ${{ !inputs.dry_run  }}
+    needs: [package]
+    uses: slsa-framework/slsa-github-generator/.github/workflows/generator_generic_slsa3.yml@v2.1.0
+    permissions:
+      # Needed to detect the GitHub Actions environment
+      actions: read
+      # Needed to create the provenance via GitHub OIDC
+      id-token: write
+      # Needed to upload assets/artifacts
+      contents: write
+    with:
+      # SHA-256 hashes of the Crate package.
+      base64-subjects: ${{ needs.package.outputs.hash }}
+
+  publish-cuda-release:
+    name: Publish CUDA Release
+    needs: [setup-instance, package] # for comparing hashes
+    runs-on: ${{ needs.setup-instance.outputs.runner-name }}
+    strategy:
+      fail-fast: false
+      # explicit include-based build matrix, of known valid options
+      matrix:
+        include:
+          - os: ubuntu-22.04
+            cuda: "12.2"
+            gcc: 9
+    env:
+      CUDA_PATH: /usr/local/cuda-${{ matrix.cuda }}
+    steps:
+      - name: Install latest stable
+        uses: dtolnay/rust-toolchain@888c2e1ea69ab0d4330cbf0af1ecc7b68f368cc1
         with:
           toolchain: stable
 
@@ -96,30 +153,45 @@ jobs:
           } >> "${GITHUB_ENV}"
 
       - name: Publish crate.io package
-        if: ${{ inputs.push_to_crates }}
         env:
           CRATES_TOKEN: ${{ secrets.CARGO_REGISTRY_TOKEN }}
           DRY_RUN: ${{ inputs.dry_run && '--dry-run' || '' }}
         run: |
           cargo publish -p tfhe-cuda-backend --token ${{ env.CRATES_TOKEN }} ${{ env.DRY_RUN }}
 
+      - name: Generate hash
+        id: published_hash
+        run: cd target/package && echo "pub_hash=$(sha256sum ./*.crate | base64 -w0)" >> "${GITHUB_OUTPUT}"
+
+      - name: Slack notification (hashes comparison)
+        if: ${{ needs.package.outputs.hash != steps.published_hash.outputs.pub_hash }}
+        continue-on-error: true
+        uses: rtCamp/action-slack-notify@c33737706dea87cd7784c687dadc9adf1be59990 # v2.3.2
+        env:
+          SLACK_COLOR: failure
+          SLACK_CHANNEL: ${{ secrets.SLACK_CHANNEL }}
+          SLACK_ICON: https://pbs.twimg.com/profile_images/1274014582265298945/OjBKP9kn_400x400.png
+          SLACK_MESSAGE: "SLSA tfhe-cuda-backend crate - hash comparison failure: (${{ env.ACTION_RUN_URL }})"
+          SLACK_USERNAME: ${{ secrets.BOT_USERNAME }}
+          SLACK_WEBHOOK: ${{ secrets.SLACK_WEBHOOK }}
+
       - name: Slack Notification
         if: ${{ failure() }}
         continue-on-error: true
-        uses: rtCamp/action-slack-notify@c33737706dea87cd7784c687dadc9adf1be59990
+        uses: rtCamp/action-slack-notify@c33737706dea87cd7784c687dadc9adf1be59990 # v2.3.2
         env:
           SLACK_COLOR: ${{ job.status }}
           SLACK_MESSAGE: "tfhe-cuda-backend release finished with status: ${{ job.status }}. (${{ env.ACTION_RUN_URL }})"
 
   teardown-instance:
     name: Teardown instance (publish-release)
-    if: ${{ always() && needs.setup-instance.result != 'skipped' }}
-    needs: [ setup-instance, publish-cuda-release ]
+    if: ${{ always() && needs.setup-instance.result == 'success' }}
+    needs: [setup-instance, publish-cuda-release]
     runs-on: ubuntu-latest
     steps:
       - name: Stop instance
         id: stop-instance
-        uses: zama-ai/slab-github-runner@801df0b8db5ea2b06128b7476c652f5ed5f193a8
+        uses: zama-ai/slab-github-runner@79939325c3c429837c10d6041e4fd8589d328bac
         with:
           mode: stop
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}

.github/workflows/make_release_tfhe_csprng.yml

@@ -0,0 +1,103 @@
+name: Publish tfhe-csprng release
+
+on:
+  workflow_dispatch:
+    inputs:
+      dry_run:
+        description: "Dry-run"
+        type: boolean
+        default: true
+
+env:
+  ACTION_RUN_URL: ${{ github.server_url }}/${{ github.repository }}/actions/runs/${{ github.run_id }}
+
+jobs:
+  verify_tag:
+    uses: ./.github/workflows/verify_tagged_commit.yml
+    secrets:
+      RELEASE_TEAM: ${{ secrets.RELEASE_TEAM }}
+      READ_ORG_TOKEN: ${{ secrets.READ_ORG_TOKEN }}
+
+  package:
+    runs-on: ubuntu-latest
+    outputs:
+      hash: ${{ steps.hash.outputs.hash }}
+    steps:
+      - name: Checkout
+        uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683 # v4.2.2
+        with:
+          fetch-depth: 0
+      - name: Prepare package
+        run: |
+          cargo package -p tfhe-csprng
+      - uses: actions/upload-artifact@ea165f8d65b6e75b540449e92b4886f43607fa02 # v4.6.2
+        with:
+          name: crate-tfhe-csprng
+          path: target/package/*.crate
+      - name: generate hash
+        id: hash
+        run: cd target/package && echo "hash=$(sha256sum ./*.crate | base64 -w0)" >> "${GITHUB_OUTPUT}"
+
+
+  provenance:
+    if: ${{ !inputs.dry_run  }}
+    needs: [package]
+    uses: slsa-framework/slsa-github-generator/.github/workflows/generator_generic_slsa3.yml@v2.1.0
+    permissions:
+      # Needed to detect the GitHub Actions environment
+      actions: read
+      # Needed to create the provenance via GitHub OIDC
+      id-token: write
+      # Needed to upload assets/artifacts
+      contents: write
+    with:
+      # SHA-256 hashes of the Crate package.
+      base64-subjects: ${{ needs.package.outputs.hash }}
+
+
+  publish_release:
+    name: Publish tfhe-csprng Release
+    needs: [verify_tag, package]
+    runs-on: ubuntu-latest
+    steps:
+      - name: Checkout
+        uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683 # v4.2.2
+        with:
+          fetch-depth: 0
+          token: ${{ secrets.FHE_ACTIONS_TOKEN }}
+      - name: Download artifact
+        uses: actions/download-artifact@95815c38cf2ff2164869cbab79da8d1f422bc89e # v4.2.1
+        with:
+          name: crate-tfhe-csprng
+          path: target/package
+      - name: Publish crate.io package
+        env:
+          CRATES_TOKEN: ${{ secrets.CARGO_REGISTRY_TOKEN }}
+          DRY_RUN: ${{ inputs.dry_run && '--dry-run' || '' }}
+        run: |
+          cargo publish -p tfhe-csprng --token ${{ env.CRATES_TOKEN }} ${{ env.DRY_RUN }}
+      - name: Generate hash
+        id: published_hash
+        run: cd target/package && echo "pub_hash=$(sha256sum ./*.crate | base64 -w0)" >> "${GITHUB_OUTPUT}"
+      - name: Slack notification (hashes comparison)
+        if: ${{ needs.package.outputs.hash != steps.published_hash.outputs.pub_hash }}
+        continue-on-error: true
+        uses: rtCamp/action-slack-notify@c33737706dea87cd7784c687dadc9adf1be59990 # v2.3.2
+        env:
+          SLACK_COLOR: failure
+          SLACK_CHANNEL: ${{ secrets.SLACK_CHANNEL }}
+          SLACK_ICON: https://pbs.twimg.com/profile_images/1274014582265298945/OjBKP9kn_400x400.png
+          SLACK_MESSAGE: "SLSA tfhe-csprng - hash comparison failure: (${{ env.ACTION_RUN_URL }})"
+          SLACK_USERNAME: ${{ secrets.BOT_USERNAME }}
+          SLACK_WEBHOOK: ${{ secrets.SLACK_WEBHOOK }}
+      - name: Slack Notification
+        if: ${{ failure() }}
+        continue-on-error: true
+        uses: rtCamp/action-slack-notify@c33737706dea87cd7784c687dadc9adf1be59990 # v2.3.2
+        env:
+          SLACK_COLOR: ${{ job.status }}
+          SLACK_CHANNEL: ${{ secrets.SLACK_CHANNEL }}
+          SLACK_ICON: https://pbs.twimg.com/profile_images/1274014582265298945/OjBKP9kn_400x400.png
+          SLACK_MESSAGE: "tfhe-csprng release finished with status: ${{ job.status }}. (${{ env.ACTION_RUN_URL }})"
+          SLACK_USERNAME: ${{ secrets.BOT_USERNAME }}
+          SLACK_WEBHOOK: ${{ secrets.SLACK_WEBHOOK }}

.github/workflows/make_release_tfhe_fft.yml

@@ -19,15 +19,53 @@ jobs:
       RELEASE_TEAM: ${{ secrets.RELEASE_TEAM }}
       READ_ORG_TOKEN: ${{ secrets.READ_ORG_TOKEN }}
 
+  package:
+    runs-on: ubuntu-latest
+    needs: verify_tag
+    outputs:
+      hash: ${{ steps.hash.outputs.hash }}
+    steps:
+      - name: Checkout
+        uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683 # v4.2.2
+        with:
+          fetch-depth: 0
+          token: ${{ secrets.FHE_ACTIONS_TOKEN }}
+      - name: Prepare package
+        run: |
+          cargo package -p tfhe-fft
+      - uses: actions/upload-artifact@ea165f8d65b6e75b540449e92b4886f43607fa02 # v4.6.2
+        with:
+          name: crate
+          path: target/package/*.crate
+      - name: generate hash
+        id: hash
+        run: cd target/package && echo "hash=$(sha256sum ./*.crate | base64 -w0)" >> "${GITHUB_OUTPUT}"
+
+  provenance:
+    if: ${{ !inputs.dry_run  }}
+    needs: [package]
+    uses: slsa-framework/slsa-github-generator/.github/workflows/generator_generic_slsa3.yml@v2.1.0
+    permissions:
+      # Needed to detect the GitHub Actions environment
+      actions: read
+      # Needed to create the provenance via GitHub OIDC
+      id-token: write
+      # Needed to upload assets/artifacts
+      contents: write
+    with:
+      # SHA-256 hashes of the Crate package.
+      base64-subjects: ${{ needs.package.outputs.hash }}
+
   publish_release:
     name: Publish tfhe-fft Release
     runs-on: ubuntu-latest
-    needs: verify_tag
+    needs: [verify_tag, package] # for comparing hashes
     steps:
       - name: Checkout
-        uses: actions/checkout@b4ffde65f46336ab88eb53be808477a3936bae11
+        uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683 # v4.2.2
         with:
           fetch-depth: 0
+          token: ${{ secrets.FHE_ACTIONS_TOKEN }}
 
       - name: Publish crate.io package
         env:
@@ -36,10 +74,26 @@ jobs:
         run: |
           cargo publish -p tfhe-fft --token ${{ env.CRATES_TOKEN }} ${{ env.DRY_RUN }}
 
+      - name: Generate hash
+        id: published_hash
+        run: cd target/package && echo "pub_hash=$(sha256sum ./*.crate | base64 -w0)" >> "${GITHUB_OUTPUT}"
+
+      - name: Slack notification (hashes comparison)
+        if: ${{ needs.package.outputs.hash != steps.published_hash.outputs.pub_hash }}
+        continue-on-error: true
+        uses: rtCamp/action-slack-notify@c33737706dea87cd7784c687dadc9adf1be59990 # v2.3.2
+        env:
+          SLACK_COLOR: failure
+          SLACK_CHANNEL: ${{ secrets.SLACK_CHANNEL }}
+          SLACK_ICON: https://pbs.twimg.com/profile_images/1274014582265298945/OjBKP9kn_400x400.png
+          SLACK_MESSAGE: "SLSA tfhe-fft crate - hash comparison failure: (${{ env.ACTION_RUN_URL }})"
+          SLACK_USERNAME: ${{ secrets.BOT_USERNAME }}
+          SLACK_WEBHOOK: ${{ secrets.SLACK_WEBHOOK }}
+
       - name: Slack Notification
         if: ${{ failure() }}
         continue-on-error: true
-        uses: rtCamp/action-slack-notify@b24d75fe0e728a4bf9fc42ee217caa686d141ee8
+        uses: rtCamp/action-slack-notify@c33737706dea87cd7784c687dadc9adf1be59990 # v2.3.2
         env:
           SLACK_COLOR: ${{ job.status }}
           SLACK_CHANNEL: ${{ secrets.SLACK_CHANNEL }}

.github/workflows/make_release_tfhe_ntt.yml

@@ -19,13 +19,50 @@ jobs:
       RELEASE_TEAM: ${{ secrets.RELEASE_TEAM }}
       READ_ORG_TOKEN: ${{ secrets.READ_ORG_TOKEN }}
 
+  package:
+    runs-on: ubuntu-latest
+    needs: verify_tag
+    outputs:
+      hash: ${{ steps.hash.outputs.hash }}
+    steps:
+      - name: Checkout
+        uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683 # v4.2.2
+        with:
+          fetch-depth: 0
+          token: ${{ secrets.FHE_ACTIONS_TOKEN }}
+      - name: Prepare package
+        run: |
+          cargo package -p tfhe-ntt
+      - uses: actions/upload-artifact@ea165f8d65b6e75b540449e92b4886f43607fa02 # v4.6.2
+        with:
+          name: crate
+          path: target/package/*.crate
+      - name: generate hash
+        id: hash
+        run: cd target/package && echo "hash=$(sha256sum ./*.crate | base64 -w0)" >> "${GITHUB_OUTPUT}"
+
+  provenance:
+    if: ${{ !inputs.dry_run  }}
+    needs: [package]
+    uses: slsa-framework/slsa-github-generator/.github/workflows/generator_generic_slsa3.yml@v2.1.0
+    permissions:
+      # Needed to detect the GitHub Actions environment
+      actions: read
+      # Needed to create the provenance via GitHub OIDC
+      id-token: write
+      # Needed to upload assets/artifacts
+      contents: write
+    with:
+      # SHA-256 hashes of the Crate package.
+      base64-subjects: ${{ needs.package.outputs.hash }}
+
   publish_release:
     name: Publish tfhe-ntt Release
     runs-on: ubuntu-latest
-    needs: verify_tag
+    needs: [verify_tag, package] # for comparing hashes
     steps:
       - name: Checkout
-        uses: actions/checkout@b4ffde65f46336ab88eb53be808477a3936bae11
+        uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683 # v4.2.2
         with:
           fetch-depth: 0
 
@@ -36,10 +73,26 @@ jobs:
         run: |
           cargo publish -p tfhe-ntt --token ${{ env.CRATES_TOKEN }} ${{ env.DRY_RUN }}
 
+      - name: Generate hash
+        id: published_hash
+        run: cd target/package && echo "pub_hash=$(sha256sum ./*.crate | base64 -w0)" >> "${GITHUB_OUTPUT}"
+
+      - name: Slack notification (hashes comparison)
+        if: ${{ needs.package.outputs.hash != steps.published_hash.outputs.pub_hash }}
+        continue-on-error: true
+        uses: rtCamp/action-slack-notify@c33737706dea87cd7784c687dadc9adf1be59990 # v2.3.2
+        env:
+          SLACK_COLOR: failure
+          SLACK_CHANNEL: ${{ secrets.SLACK_CHANNEL }}
+          SLACK_ICON: https://pbs.twimg.com/profile_images/1274014582265298945/OjBKP9kn_400x400.png
+          SLACK_MESSAGE: "SLSA tfhe-ntt crate - hash comparison failure: (${{ env.ACTION_RUN_URL }})"
+          SLACK_USERNAME: ${{ secrets.BOT_USERNAME }}
+          SLACK_WEBHOOK: ${{ secrets.SLACK_WEBHOOK }}
+
       - name: Slack Notification
         if: ${{ failure() }}
         continue-on-error: true
-        uses: rtCamp/action-slack-notify@b24d75fe0e728a4bf9fc42ee217caa686d141ee8
+        uses: rtCamp/action-slack-notify@c33737706dea87cd7784c687dadc9adf1be59990 # v2.3.2
         env:
           SLACK_COLOR: ${{ job.status }}
           SLACK_CHANNEL: ${{ secrets.SLACK_CHANNEL }}

.github/workflows/make_release_tfhe_versionable.yml

@@ -2,14 +2,13 @@ name: Publish tfhe-versionable release
 
 on:
   workflow_dispatch:
-    inputs:
-      dry_run:
-        description: "Dry-run"
-        type: boolean
-        default: true
 
 env:
   ACTION_RUN_URL: ${{ github.server_url }}/${{ github.repository }}/actions/runs/${{ github.run_id }}
+  SLACK_CHANNEL: ${{ secrets.SLACK_CHANNEL }}
+  SLACK_ICON: https://pbs.twimg.com/profile_images/1274014582265298945/OjBKP9kn_400x400.png
+  SLACK_USERNAME: ${{ secrets.BOT_USERNAME }}
+  SLACK_WEBHOOK: ${{ secrets.SLACK_WEBHOOK }}
 
 jobs:
   verify_tag:
@@ -18,38 +17,149 @@ jobs:
       RELEASE_TEAM: ${{ secrets.RELEASE_TEAM }}
       READ_ORG_TOKEN: ${{ secrets.READ_ORG_TOKEN }}
 
+  package-derive:
+    name: Package tfhe-versionable-derive Release
+    runs-on: ubuntu-latest
+    outputs:
+      hash: ${{ steps.hash.outputs.hash }}
+    steps:
+      - name: Checkout
+        uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683 # v4.2.2
+        with:
+          fetch-depth: 0
+      - name: Prepare package
+        run: |
+          cargo package -p tfhe-versionable-derive
+      - uses: actions/upload-artifact@ea165f8d65b6e75b540449e92b4886f43607fa02 # v4.6.2
+        with:
+          name: crate-tfhe-versionable-derive
+          path: target/package/*.crate
+      - name: generate hash
+        id: hash
+        run: cd target/package && echo "hash=$(sha256sum ./*.crate | base64 -w0)" >> "${GITHUB_OUTPUT}"
+
+  provenance-derive:
+    needs: [package-derive]
+    uses: slsa-framework/slsa-github-generator/.github/workflows/generator_generic_slsa3.yml@v2.1.0
+    permissions:
+      # Needed to detect the GitHub Actions environment
+      actions: read
+      # Needed to create the provenance via GitHub OIDC
+      id-token: write
+      # Needed to upload assets/artifacts
+      contents: write
+    with:
+      # SHA-256 hashes of the Crate package.
+      base64-subjects: ${{ needs.package-derive.outputs.hash }}
+
+  publish_release-derive:
+    name: Publish tfhe-versionable-derive Release
+    needs: [ verify_tag, package-derive ] # for comparing hashes
+    runs-on: ubuntu-latest
+    steps:
+      - name: Checkout
+        uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683 # v4.2.2
+        with:
+          fetch-depth: 0
+          persist-credentials: 'false'
+          token: ${{ secrets.REPO_CHECKOUT_TOKEN }}
+      - name: Download artifact
+        uses: actions/download-artifact@95815c38cf2ff2164869cbab79da8d1f422bc89e # v4.2.1
+        with:
+          name: crate-tfhe-versionable-derive
+          path: target/package
+      - name: Publish crate.io package
+        env:
+          CRATES_TOKEN: ${{ secrets.CARGO_REGISTRY_TOKEN }}
+        run: |
+          cargo publish -p tfhe-versionable-derive --token ${{ env.CRATES_TOKEN }}
+      - name: Generate hash
+        id: published_hash
+        run: cd target/package && echo "pub_hash=$(sha256sum ./*.crate | base64 -w0)" >> "${GITHUB_OUTPUT}"
+      - name: Slack notification (hashes comparison)
+        if: ${{ needs.package-derive.outputs.hash != steps.published_hash.outputs.pub_hash }}
+        continue-on-error: true
+        uses: rtCamp/action-slack-notify@c33737706dea87cd7784c687dadc9adf1be59990 # v2.3.2
+        env:
+          SLACK_COLOR: failure
+          SLACK_MESSAGE: "SLSA tfhe-versionable-derive - hash comparison failure: (${{ env.ACTION_RUN_URL }})"
+      - name: Slack Notification
+        if: ${{ failure() }}
+        continue-on-error: true
+        uses: rtCamp/action-slack-notify@c33737706dea87cd7784c687dadc9adf1be59990 # v2.3.2
+        env:
+          SLACK_COLOR: ${{ job.status }}
+          SLACK_MESSAGE: "tfhe-versionable-derive release finished with status: ${{ job.status }}. (${{ env.ACTION_RUN_URL }})"
+
+  package:
+    name: Package tfhe-versionable Release
+    needs: publish_release-derive
+    runs-on: ubuntu-latest
+    outputs:
+      hash: ${{ steps.hash.outputs.hash }}
+    steps:
+      - name: Checkout
+        uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
+        with:
+          fetch-depth: 0
+      - name: Prepare package
+        run: |
+          cargo package -p tfhe-versionable
+      - uses: actions/upload-artifact@ea165f8d65b6e75b540449e92b4886f43607fa02 # v4.6.2
+        with:
+          name: crate-tfhe-versionable
+          path: target/package/*.crate
+      - name: generate hash
+        id: hash
+        run: cd target/package && echo "hash=$(sha256sum ./*.crate | base64 -w0)" >> "${GITHUB_OUTPUT}"
+
+  provenance:
+    needs: package
+    uses: slsa-framework/slsa-github-generator/.github/workflows/generator_generic_slsa3.yml@v2.1.0
+    permissions:
+      # Needed to detect the GitHub Actions environment
+      actions: read
+      # Needed to create the provenance via GitHub OIDC
+      id-token: write
+      # Needed to upload assets/artifacts
+      contents: write
+    with:
+      # SHA-256 hashes of the Crate package.
+      base64-subjects: ${{ needs.package.outputs.hash }}
+
   publish_release:
     name: Publish tfhe-versionable Release
-    needs: verify_tag
+    needs: package # for comparing hashes
     runs-on: ubuntu-latest
     steps:
       - name: Checkout
         uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
         with:
           fetch-depth: 0
-
-      - name: Publish proc-macro crate
-        env:
-          CRATES_TOKEN: ${{ secrets.CARGO_REGISTRY_TOKEN }}
-          DRY_RUN: ${{ inputs.dry_run && '--dry-run' || '' }}
-        run: |
-          cargo publish -p tfhe-versionable-derive --token ${{ env.CRATES_TOKEN }} ${{ env.DRY_RUN }}
-
-      - name: Publish main crate
-        if: ${{ ! inputs.dry_run }}
+      - name: Download artifact
+        uses: actions/download-artifact@95815c38cf2ff2164869cbab79da8d1f422bc89e # v4.2.1
+        with:
+          name: crate-tfhe-versionable
+          path: target/package
+      - name: Publish crate.io package
         env:
           CRATES_TOKEN: ${{ secrets.CARGO_REGISTRY_TOKEN }}
         run: |
           cargo publish -p tfhe-versionable --token ${{ env.CRATES_TOKEN }}
-
+      - name: Generate hash
+        id: published_hash
+        run: cd target/package && echo "pub_hash=$(sha256sum ./*.crate | base64 -w0)" >> "${GITHUB_OUTPUT}"
+      - name: Slack notification (hashes comparison)
+        if: ${{ needs.package.outputs.hash != steps.published_hash.outputs.pub_hash }}
+        continue-on-error: true
+        uses: rtCamp/action-slack-notify@c33737706dea87cd7784c687dadc9adf1be59990 # v2.3.2
+        env:
+          SLACK_COLOR: failure
+          SLACK_MESSAGE: "SLSA tfhe-versionable - hash comparison failure: (${{ env.ACTION_RUN_URL }})"
       - name: Slack Notification
         if: ${{ failure() }}
         continue-on-error: true
-        uses: rtCamp/action-slack-notify@c33737706dea87cd7784c687dadc9adf1be59990
+        uses: rtCamp/action-slack-notify@c33737706dea87cd7784c687dadc9adf1be59990 # v2.3.2
         env:
           SLACK_COLOR: ${{ job.status }}
-          SLACK_CHANNEL: ${{ secrets.SLACK_CHANNEL }}
-          SLACK_ICON: https://pbs.twimg.com/profile_images/1274014582265298945/OjBKP9kn_400x400.png
           SLACK_MESSAGE: "tfhe-versionable release finished with status: ${{ job.status }}. (${{ env.ACTION_RUN_URL }})"
-          SLACK_USERNAME: ${{ secrets.BOT_USERNAME }}
-          SLACK_WEBHOOK: ${{ secrets.SLACK_WEBHOOK }}

.github/workflows/make_release_zk_pok.yml

@@ -1,4 +1,3 @@
-# Publish new release of tfhe-zk-pok on crates.io.
 name: Publish tfhe-zk-pok release
 
 on:
@@ -13,6 +12,40 @@ env:
   ACTION_RUN_URL: ${{ github.server_url }}/${{ github.repository }}/actions/runs/${{ github.run_id }}
 
 jobs:
+  package:
+      runs-on: ubuntu-latest
+      outputs:
+        hash: ${{ steps.hash.outputs.hash }}
+      steps:
+        - name: Checkout
+          uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683 # v4.2.2
+          with:
+            fetch-depth: 0
+        - name: Prepare package
+          run: |
+            cargo package -p tfhe-zk-pok
+        - uses: actions/upload-artifact@ea165f8d65b6e75b540449e92b4886f43607fa02 # v4.6.2
+          with:
+            name: crate-zk-pok
+            path: target/package/*.crate
+        - name: generate hash
+          id: hash
+          run: cd target/package && echo "hash=$(sha256sum ./*.crate | base64 -w0)" >> "${GITHUB_OUTPUT}"
+  provenance:
+    if: ${{ !inputs.dry_run  }}
+    needs: [package]
+    uses: slsa-framework/slsa-github-generator/.github/workflows/generator_generic_slsa3.yml@v2.1.0
+    permissions:
+      # Needed to detect the GitHub Actions environment
+      actions: read
+      # Needed to create the provenance via GitHub OIDC
+      id-token: write
+      # Needed to upload assets/artifacts
+      contents: write
+    with:
+      # SHA-256 hashes of the Crate package.
+      base64-subjects: ${{ needs.package.outputs.hash }}
+
   verify_tag:
     uses: ./.github/workflows/verify_tagged_commit.yml
     secrets:
@@ -21,25 +54,44 @@ jobs:
 
   publish_release:
     name: Publish tfhe-zk-pok Release
-    needs: verify_tag
+    needs: [verify_tag, package] # for comparing hashes
     runs-on: ubuntu-latest
     steps:
       - name: Checkout
-        uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
+        uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683 # v4.2.2
         with:
           fetch-depth: 0
-
+          persist-credentials: 'false'
+          token: ${{ secrets.REPO_CHECKOUT_TOKEN }}
+      - name: Download artifact
+        uses: actions/download-artifact@95815c38cf2ff2164869cbab79da8d1f422bc89e # v4.2.1
+        with:
+          name: crate-zk-pok
+          path: target/package
       - name: Publish crate.io package
         env:
           CRATES_TOKEN: ${{ secrets.CARGO_REGISTRY_TOKEN }}
           DRY_RUN: ${{ inputs.dry_run && '--dry-run' || '' }}
         run: |
           cargo publish -p tfhe-zk-pok --token ${{ env.CRATES_TOKEN }} ${{ env.DRY_RUN }}
-
+      - name: Verify hash
+        id: published_hash
+        run: cd target/package && echo "pub_hash=$(sha256sum ./*.crate | base64 -w0)" >> "${GITHUB_OUTPUT}"
+      - name: Slack notification (hashes comparison)
+        if: ${{ needs.package.outputs.hash != steps.published_hash.outputs.pub_hash }}
+        continue-on-error: true
+        uses: rtCamp/action-slack-notify@c33737706dea87cd7784c687dadc9adf1be59990 # v2.3.2
+        env:
+          SLACK_COLOR: failure
+          SLACK_CHANNEL: ${{ secrets.SLACK_CHANNEL }}
+          SLACK_ICON: https://pbs.twimg.com/profile_images/1274014582265298945/OjBKP9kn_400x400.png
+          SLACK_MESSAGE: "SLSA tfhe-zk-pok crate - hash comparison failure: (${{ env.ACTION_RUN_URL }})"
+          SLACK_USERNAME: ${{ secrets.BOT_USERNAME }}
+          SLACK_WEBHOOK: ${{ secrets.SLACK_WEBHOOK }}
       - name: Slack Notification
         if: ${{ failure() }}
         continue-on-error: true
-        uses: rtCamp/action-slack-notify@c33737706dea87cd7784c687dadc9adf1be59990
+        uses: rtCamp/action-slack-notify@c33737706dea87cd7784c687dadc9adf1be59990 # v2.3.2
         env:
           SLACK_COLOR: ${{ job.status }}
           SLACK_CHANNEL: ${{ secrets.SLACK_CHANNEL }}

.github/workflows/parameters_check.yml

@@ -14,7 +14,7 @@ on:
 
 jobs:
   params-curves-security-check:
-    runs-on: large_ubuntu_16
+    runs-on: large_ubuntu_16-22.04
     steps:
       - name: Checkout tfhe-rs
         uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683

.github/workflows/sync_on_push.yml

@@ -16,6 +16,8 @@ jobs:
         uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
         with:
           fetch-depth: 0
+          persist-credentials: 'false'
+          token: ${{ secrets.REPO_CHECKOUT_TOKEN }}
       - name: git-sync
         uses: wei/git-sync@55c6b63b4f21607da0e9877ca9b4d11a29fc6d83
         with:

.gitignore

@@ -32,5 +32,9 @@ web-test-runner/
 node_modules/
 package-lock.json
 
+# Python .env
+.env
+
 # Dir used for backward compatibility test data
-tfhe/tfhe-backward-compat-data/
+tests/tfhe-backward-compat-data/
+ci/

.linelint.yml

@@ -1,11 +1,15 @@
 ignore:
   - .git
   - target
+  - tfhe/build
+  - venv
+  - web-test-runner
   - tfhe/benchmarks_parameters
   - tfhe/web_wasm_parallel_tests/node_modules
   - tfhe/web_wasm_parallel_tests/dist
   - keys
   - coverage
+  - utils/tfhe-lints/ui/main.stderr
 
 rules:
   # checks if file ends in a newline character

CONTRIBUTING.md

@@ -0,0 +1,233 @@
+# Contributing to TFHE-rs
+
+This document provides guidance on how to contribute to **TFHE-rs**.
+
+There are two ways to contribute:
+
+- **Report issues:** Open issues on GitHub to report bugs, suggest improvements, or note typos.
+- **Submit codes**: To become an official contributor, you must sign our Contributor License Agreement (CLA). Our CLA-bot will guide you through this process when you open your first pull request.
+
+## 1. Setting up the project
+
+Start by [forking](https://docs.github.com/en/pull-requests/collaborating-with-pull-requests/working-with-forks/fork-a-repo) the **TFHE-rs** repository.
+
+{% hint style="info" %}
+- **Rust version**:  Ensure that you use a Rust version >= 1.81 to compile **TFHE-rs**.
+- **Incompatibility**: AArch64-based machines are not yet supported for Windows as it's currently missing an entropy source to be able to seed the [CSPRNGs](https://en.wikipedia.org/wiki/Cryptographically_secure_pseudorandom_number_generator) used in **TFHE-rs**.
+- **Performance**: For optimal performance, it is highly recommended to run **TFHE-rs** code in release mode with cargo's `--release` flag.
+{% endhint %}
+
+To get more details about the library, please refer to the [documentation](https://docs.zama.ai/tfhe-rs).
+
+## 2. Creating a new branch
+
+When creating your branch, make sure to use the following format :
+
+```
+git checkout -b {feat|fix|docs|chore…}/short_description
+```
+
+For example:
+
+```
+git checkout -b feat/new_feature_X
+```
+
+## 3. Before committing
+
+### 3.1 Linting
+
+Each commit to **TFHE-rs** should conform to the standards of the project. In particular, every source code, docker or workflows files should be linted to prevent programmatic and stylistic errors.
+
+- Rust source code linters: `clippy`
+- Typescript/Javascript source code linters: `eslint`, `prettier`
+
+To apply automatic code formatting, run:
+
+```
+make fmt
+```
+
+You can perform linting of all Cargo targets with:
+
+```
+make clippy_all_targets
+```
+
+### 3.2 Testing
+
+Your contributions must include comprehensive documentation and tests without breaking existing tests. To run pre-commit checks, execute:
+
+```
+make pcc
+```
+
+This command ensure that all the targets in the library are building correctly.
+For a faster check, use:
+
+```
+make fpcc
+```
+
+If you're contributing to GPU code, run also:
+
+```
+make pcc_gpu
+```
+
+Unit testing suites are heavy and can require a lot of computing power and RAM availability.
+Whilst tests are run automatically in continuous integration pipeline, you can run tests locally.
+
+All unit tests have a command formatted as:
+
+```
+make test_*
+```
+
+Run `make help` to display a list of all the commands available.
+
+To quickly test your changes locally, follow these steps:
+ 1. Locate where the code has changed.
+ 2. Add (or modify) a Cargo test filter to the corresponding `make` target in Makefile.
+ 3. Run the target.
+
+{% hint style="success" %}
+`make test_<something>` will print the underlying cargo command in STDOUT. You can quickly test your changes by copy/pasting the command and then modify it to suit your needs.
+{% endhint %}
+
+For example, if you made changes in `tfhe/src/integer/*`, you can test them with the following steps:
+ 1. In `test_integer` target, replace the filter `-- integer::` by `-- my_new_test`.
+ 2. Run `make test_integer`.
+
+## 4. Committing
+
+**TFHE-rs** follows the conventional commit specification to maintain a consistent commit history, essential for Semantic Versioning ([semver.org](https://semver.org/)).
+Commit messages are automatically checked in CI and will be rejected if they do not comply, so make sure that you follow the commit conventions detailed on [this page]
+(https://www.conventionalcommits.org/en/v1.0.0/).
+
+## 5. Rebasing
+
+Before creating a pull request, rebase your branch on the repository's `main` branch. Merge commits are not permitted, thus rebasing ensures fewer conflicts and a smoother PR review process.
+
+## 6. Opening a Pull Request
+
+Once your changes are ready, open a pull request.
+
+For instructions on creating a PR from a fork, refer to GitHub's [official documentation](https://docs.github.com/en/pull-requests/collaborating-with-pull-requests/proposing-changes-to-your-work-with-pull-requests/creating-a-pull-request-from-a-fork).
+
+## 7. Continuous integration
+
+Before a pull request can be merged, several test suites run automatically. Below is an overview of the CI process:
+
+```mermaid
+---
+title: Continuous Integration Process
+---
+sequenceDiagram
+    autonumber
+
+    participant Contributor
+    participant GitHub
+    participant Reviewer
+    participant CI-pipeline
+
+    Contributor ->> GitHub: Open pull-request
+    GitHub -->> Contributor: Ask for CLA signing (once)
+    loop
+        Reviewer ->> GitHub: Review code
+        Reviewer ->> CI-pipeline: Approve workflows (short-run)
+        CI-pipeline -->> GitHub: Send checks results
+        Contributor ->> GitHub: Make changes
+    end
+    Reviewer ->> GitHub: Pull-request approval
+    Reviewer ->> CI-pipeline: Approve workflows (long-run)
+    CI-pipeline -->> GitHub: Send checks results
+    Reviewer -->> GitHub: Merge if pipeline green
+```
+
+> [!Note]
+>Useful details:
+>* pipeline is triggered by humans
+>* review team is located in Paris timezone, pipeline launch will most likely happen during office hours
+>* direct changes to CI related files are not allowed for external contributors
+>* run `make pcc` to fix any build errors before pushing commits
+
+## 8. Data versioning
+
+Data serialized with TFHE-rs must remain backward compatible. This is done using the [tfhe-versionable](https://crates.io/crates/tfhe-versionable) crate.
+
+If you modify a type that derives `Versionize` in a backward-incompatible way, an upgrade implementation must be provided.
+
+For example, these changes are data breaking:
+ * Adding a field to a struct.
+ * Changing the order of the fields within a struct or the variants within an enum.
+ * Renaming a field of a struct or a variant of an enum.
+ * Changing the type of field in a struct or a variant in an enum.
+
+On the contrary, these changes are *not* data breaking:
+ * Renaming a type (unless it implements the `Named` trait).
+ * Adding a variant to the end of an enum.
+
+## Example: adding a field
+
+Suppose you want to add an i32 field to a type named `MyType`. The original type is defined as:
+```rust
+#[derive(Serialize, Deserialize, Versionize)]
+#[versionize(MyTypeVersions)]
+struct MyType {
+  val: u64,
+}
+```
+And you want to change it to:
+```rust
+#[derive(Serialize, Deserialize, Versionize)]
+#[versionize(MyTypeVersions)]
+struct MyType {
+  val: u64,
+  other_val: i32
+}
+```
+
+Follow these steps:
+
+ 1. Navigate to the definition of the dispatch enum of this type. This is the type inside the `#[versionize(MyTypeVersions)]` macro attribute. In general, this type has the same name as the base type with a `Versions` suffix. You should find something like
+
+```rust
+#[derive(VersionsDispatch)]
+enum MyTypeVersions {
+  V0(MyTypeV0),
+  V1(MyType)
+}
+```
+
+ 2. Add a new variant to the enum to preserve the previous version of the type. You can simply copy and paste the previous definition of the type and add a version suffix:
+
+```rust
+#[derive(Version)]
+struct MyTypeV1 {
+  val: u64,
+}
+
+#[derive(VersionsDispatch)]
+enum MyTypeVersions {
+  V0(MyTypeV0),
+  V1(MyTypeV1),
+  V2(MyType) // Here this points to your modified type
+}
+```
+
+ 3. Implement the `Upgrade` trait to define how we should go from the previous version to the current version:
+```rust
+impl Upgrade<MyType> for MyTypeV1 {
+  type Error = Infallible;
+
+   fn upgrade(self) -> Result<MyType, Self::Error> {
+       Ok(MyType {
+           val: self.val,
+           other_val: 0
+        })
+   }
+}
+```
+
+ 4. Fix the upgrade target of the previous version. In this example, `impl Upgrade<MyType> for MyTypeV0 {` should simply be changed to `impl Upgrade<MyTypeV1> for MyTypeV0 {`

Cargo.toml

@@ -6,26 +6,29 @@ members = [
     "tfhe-ntt",
     "tfhe-zk-pok",
     "tasks",
-    "apps/trivium",
     "tfhe-csprng",
     "backends/tfhe-cuda-backend",
     "utils/tfhe-versionable",
     "utils/tfhe-versionable-derive",
+    "tests",
 ]
 
 exclude = [
-    "tfhe/backward_compatibility_tests",
-    "utils/cargo-tfhe-lints-inner",
-    "utils/cargo-tfhe-lints"
+    "tests/backward_compatibility_tests",
+    "utils/tfhe-lints",
+    "apps/trivium",
 ]
 [workspace.dependencies]
-aligned-vec = { version = "0.5", default-features = false }
+aligned-vec = { version = "0.6", default-features = false }
 bytemuck = "1.14.3"
-dyn-stack = { version = "0.10", default-features = false }
+dyn-stack = { version = "0.11", default-features = false }
+itertools = "0.14"
 num-complex = "0.4"
-pulp = { version = "0.18.22", default-features = false }
+pulp = { version = "0.21", default-features = false }
+rand = "0.8"
+rayon = "1"
 serde = { version = "1.0", default-features = false }
-wasm-bindgen = ">=0.2.86,<0.2.94"
+wasm-bindgen = "0.2.100"
 
 [profile.bench]
 lto = "fat"
@@ -43,3 +46,6 @@ inherits = "dev"
 opt-level = 3
 lto = "off"
 debug-assertions = false
+
+[workspace.metadata.dylint]
+libraries = [{ path = "utils/tfhe-lints" }]

LICENSE

@@ -1,6 +1,6 @@
 BSD 3-Clause Clear License
 
-Copyright © 2024 ZAMA.
+Copyright © 2025 ZAMA.
 All rights reserved.
 
 Redistribution and use in source and binary forms, with or without modification,

Makefile

@@ -2,7 +2,6 @@ SHELL:=$(shell /usr/bin/env which bash)
 OS:=$(shell uname)
 RS_CHECK_TOOLCHAIN:=$(shell cat toolchain.txt | tr -d '\n')
 CARGO_RS_CHECK_TOOLCHAIN:=+$(RS_CHECK_TOOLCHAIN)
-TARGET_ARCH_FEATURE:=$(shell ./scripts/get_arch_feature.sh)
 CPU_COUNT=$(shell ./scripts/cpu_count.sh)
 RS_BUILD_TOOLCHAIN:=stable
 CARGO_RS_BUILD_TOOLCHAIN:=+$(RS_BUILD_TOOLCHAIN)
@@ -19,12 +18,15 @@ FAST_BENCH?=FALSE
 NIGHTLY_TESTS?=FALSE
 BENCH_OP_FLAVOR?=DEFAULT
 BENCH_TYPE?=latency
+BENCH_PARAM_TYPE?=classical
+BENCH_PARAMS_SET?=default
 NODE_VERSION=22.6
 BACKWARD_COMPAT_DATA_URL=https://github.com/zama-ai/tfhe-backward-compat-data.git
-BACKWARD_COMPAT_DATA_BRANCH?=v0.4
+BACKWARD_COMPAT_DATA_BRANCH?=$(shell ./scripts/backward_compat_data_version.py)
 BACKWARD_COMPAT_DATA_PROJECT=tfhe-backward-compat-data
 BACKWARD_COMPAT_DATA_DIR=$(BACKWARD_COMPAT_DATA_PROJECT)
 TFHE_SPEC:=tfhe
+WASM_PACK_VERSION="0.13.1"
 # We are kind of hacking the cut here, the version cannot contain a quote '"'
 WASM_BINDGEN_VERSION:=$(shell grep '^wasm-bindgen[[:space:]]*=' Cargo.toml | cut -d '"' -f 2 | xargs)
 WEB_RUNNER_DIR=web-test-runner
@@ -116,8 +118,8 @@ install_wasm_bindgen_cli: install_rs_build_toolchain
 
 .PHONY: install_wasm_pack # Install wasm-pack to build JS packages
 install_wasm_pack: install_rs_build_toolchain
-	@wasm-pack --version > /dev/null 2>&1 || \
-	cargo $(CARGO_RS_BUILD_TOOLCHAIN) install --locked wasm-pack@0.13.0 || \
+	@wasm-pack --version | grep "$(WASM_PACK_VERSION)" > /dev/null 2>&1 || \
+	cargo $(CARGO_RS_BUILD_TOOLCHAIN) install --locked wasm-pack@$(WASM_PACK_VERSION) || \
 	( echo "Unable to install cargo wasm-pack, unknown error." && exit 1 )
 
 .PHONY: install_node # Install last version of NodeJS via nvm
@@ -151,10 +153,9 @@ install_tarpaulin: install_rs_build_toolchain
 	cargo $(CARGO_RS_BUILD_TOOLCHAIN) install cargo-tarpaulin --locked || \
 	( echo "Unable to install cargo tarpaulin, unknown error." && exit 1 )
 
-.PHONY: install_tfhe_lints # Install custom tfhe-rs lints
-install_tfhe_lints:
-	(cd utils/cargo-tfhe-lints-inner && cargo install --path .) && \
-	cd utils/cargo-tfhe-lints && cargo install --path .
+.PHONY: install_cargo_dylint # Install custom tfhe-rs lints
+install_cargo_dylint:
+	cargo install cargo-dylint dylint-link
 
 .PHONY: install_typos_checker # Install typos checker
 install_typos_checker: install_rs_build_toolchain
@@ -243,7 +244,8 @@ fmt_js: check_nvm_installed
 	source ~/.nvm/nvm.sh && \
 	nvm install $(NODE_VERSION) && \
 	nvm use $(NODE_VERSION) && \
-	$(MAKE) -C tfhe/web_wasm_parallel_tests fmt
+	$(MAKE) -C tfhe/web_wasm_parallel_tests fmt && \
+	$(MAKE) -C tfhe/js_on_wasm_tests fmt
 
 .PHONY: fmt_gpu # Format rust and cuda code
 fmt_gpu: install_rs_check_toolchain
@@ -272,7 +274,8 @@ check_fmt_js: check_nvm_installed
 	source ~/.nvm/nvm.sh && \
 	nvm install $(NODE_VERSION) && \
 	nvm use $(NODE_VERSION) && \
-	$(MAKE) -C tfhe/web_wasm_parallel_tests check_fmt
+	$(MAKE) -C tfhe/web_wasm_parallel_tests check_fmt && \
+	$(MAKE) -C tfhe/js_on_wasm_tests check_fmt
 
 .PHONY: check_typos # Check for typos in codebase
 check_typos: install_typos_checker
@@ -281,14 +284,14 @@ check_typos: install_typos_checker
 .PHONY: clippy_gpu # Run clippy lints on tfhe with "gpu" enabled
 clippy_gpu: install_rs_check_toolchain
 	RUSTFLAGS="$(RUSTFLAGS)" cargo "$(CARGO_RS_CHECK_TOOLCHAIN)" clippy \
-		--features=$(TARGET_ARCH_FEATURE),boolean,shortint,integer,internal-keycache,gpu \
+		--features=boolean,shortint,integer,internal-keycache,gpu,pbs-stats \
 		--all-targets \
 		-p $(TFHE_SPEC) -- --no-deps -D warnings
 
 .PHONY: check_gpu # Run check on tfhe with "gpu" enabled
 check_gpu: install_rs_check_toolchain
 	RUSTFLAGS="$(RUSTFLAGS)" cargo "$(CARGO_RS_CHECK_TOOLCHAIN)" check \
-		--features=$(TARGET_ARCH_FEATURE),boolean,shortint,integer,internal-keycache,gpu \
+		--features=boolean,shortint,integer,internal-keycache,gpu,pbs-stats \
 		--all-targets \
 		-p $(TFHE_SPEC)
 
@@ -307,52 +310,54 @@ lint_workflow: check_actionlint_installed
 .PHONY: clippy_core # Run clippy lints on core_crypto with and without experimental features
 clippy_core: install_rs_check_toolchain
 	RUSTFLAGS="$(RUSTFLAGS)" cargo "$(CARGO_RS_CHECK_TOOLCHAIN)" clippy \
-		--features=$(TARGET_ARCH_FEATURE) \
 		-p $(TFHE_SPEC) -- --no-deps -D warnings
 	RUSTFLAGS="$(RUSTFLAGS)" cargo "$(CARGO_RS_CHECK_TOOLCHAIN)" clippy \
-		--features=$(TARGET_ARCH_FEATURE),experimental \
+		--features=experimental \
 		-p $(TFHE_SPEC) -- --no-deps -D warnings
 	RUSTFLAGS="$(RUSTFLAGS)" cargo "$(CARGO_RS_CHECK_TOOLCHAIN)" clippy \
-		--features=$(TARGET_ARCH_FEATURE),nightly-avx512 \
+		--features=nightly-avx512 \
 		-p $(TFHE_SPEC) -- --no-deps -D warnings
 	RUSTFLAGS="$(RUSTFLAGS)" cargo "$(CARGO_RS_CHECK_TOOLCHAIN)" clippy \
-		--features=$(TARGET_ARCH_FEATURE),experimental,nightly-avx512 \
+		--features=experimental,nightly-avx512 \
 		-p $(TFHE_SPEC) -- --no-deps -D warnings
 	RUSTFLAGS="$(RUSTFLAGS)" cargo "$(CARGO_RS_CHECK_TOOLCHAIN)" clippy \
-		--features=$(TARGET_ARCH_FEATURE),zk-pok \
+		--features=zk-pok \
 		-p $(TFHE_SPEC) -- --no-deps -D warnings
 
 .PHONY: clippy_boolean # Run clippy lints enabling the boolean features
 clippy_boolean: install_rs_check_toolchain
 	RUSTFLAGS="$(RUSTFLAGS)" cargo "$(CARGO_RS_CHECK_TOOLCHAIN)" clippy \
-		--features=$(TARGET_ARCH_FEATURE),boolean \
+		--features=boolean \
 		-p $(TFHE_SPEC) -- --no-deps -D warnings
 
 .PHONY: clippy_shortint # Run clippy lints enabling the shortint features
 clippy_shortint: install_rs_check_toolchain
 	RUSTFLAGS="$(RUSTFLAGS)" cargo "$(CARGO_RS_CHECK_TOOLCHAIN)" clippy \
-		--features=$(TARGET_ARCH_FEATURE),shortint \
+		--features=shortint \
 		-p $(TFHE_SPEC) -- --no-deps -D warnings
 	RUSTFLAGS="$(RUSTFLAGS)" cargo "$(CARGO_RS_CHECK_TOOLCHAIN)" clippy \
-		--features=$(TARGET_ARCH_FEATURE),shortint,experimental \
+		--features=shortint,experimental \
 		-p $(TFHE_SPEC) -- --no-deps -D warnings
 	RUSTFLAGS="$(RUSTFLAGS)" cargo "$(CARGO_RS_CHECK_TOOLCHAIN)" clippy \
-		--features=$(TARGET_ARCH_FEATURE),zk-pok,shortint \
+		--features=zk-pok,shortint \
 		-p $(TFHE_SPEC) -- --no-deps -D warnings
 
 .PHONY: clippy_integer # Run clippy lints enabling the integer features
 clippy_integer: install_rs_check_toolchain
 	RUSTFLAGS="$(RUSTFLAGS)" cargo "$(CARGO_RS_CHECK_TOOLCHAIN)" clippy \
-		--features=$(TARGET_ARCH_FEATURE),integer \
+		--features=integer \
 		-p $(TFHE_SPEC) -- --no-deps -D warnings
 	RUSTFLAGS="$(RUSTFLAGS)" cargo "$(CARGO_RS_CHECK_TOOLCHAIN)" clippy \
-		--features=$(TARGET_ARCH_FEATURE),integer,experimental \
+		--features=integer,experimental \
+		-p $(TFHE_SPEC) -- --no-deps -D warnings
+	RUSTFLAGS="$(RUSTFLAGS)" cargo "$(CARGO_RS_CHECK_TOOLCHAIN)" clippy \
+		--features=integer,experimental,extended-types \
 		-p $(TFHE_SPEC) -- --no-deps -D warnings
 
 .PHONY: clippy # Run clippy lints enabling the boolean, shortint, integer
 clippy: install_rs_check_toolchain
 	RUSTFLAGS="$(RUSTFLAGS)" cargo "$(CARGO_RS_CHECK_TOOLCHAIN)" clippy --all-targets \
-		--features=$(TARGET_ARCH_FEATURE),boolean,shortint,integer \
+		--features=boolean,shortint,integer \
 		-p $(TFHE_SPEC) -- --no-deps -D warnings
 
 .PHONY: clippy_rustdoc # Run clippy lints on doctests enabling the boolean, shortint, integer and zk-pok
@@ -363,22 +368,33 @@ clippy_rustdoc: install_rs_check_toolchain
 	fi && \
 	CLIPPYFLAGS="-D warnings" RUSTDOCFLAGS="--no-run --nocapture --test-builder ./scripts/clippy_driver.sh -Z unstable-options" \
 		cargo "$(CARGO_RS_CHECK_TOOLCHAIN)" test --doc \
-		--features=$(TARGET_ARCH_FEATURE),boolean,shortint,integer,zk-pok,pbs-stats,strings \
+		--features=boolean,shortint,integer,zk-pok,pbs-stats,strings,experimental \
+		-p $(TFHE_SPEC)
+
+.PHONY: clippy_rustdoc_gpu # Run clippy lints on doctests enabling the boolean, shortint, integer and zk-pok
+clippy_rustdoc_gpu: install_rs_check_toolchain
+	if [[ "$(OS)" != "Linux" ]]; then \
+		echo "WARNING: skipped clippy_rustdoc_gpu, unsupported OS $(OS)"; \
+		exit 0; \
+	fi && \
+	CLIPPYFLAGS="-D warnings" RUSTDOCFLAGS="--no-run --nocapture --test-builder ./scripts/clippy_driver.sh -Z unstable-options" \
+		cargo "$(CARGO_RS_CHECK_TOOLCHAIN)" test --doc \
+		--features=boolean,shortint,integer,zk-pok,pbs-stats,strings,experimental,gpu \
 		-p $(TFHE_SPEC)
 
 .PHONY: clippy_c_api # Run clippy lints enabling the boolean, shortint and the C API
 clippy_c_api: install_rs_check_toolchain
 	RUSTFLAGS="$(RUSTFLAGS)" cargo "$(CARGO_RS_CHECK_TOOLCHAIN)" clippy \
-		--features=$(TARGET_ARCH_FEATURE),boolean-c-api,shortint-c-api,high-level-c-api \
+		--features=boolean-c-api,shortint-c-api,high-level-c-api,extended-types \
 		-p $(TFHE_SPEC) -- --no-deps -D warnings
 
 .PHONY: clippy_js_wasm_api # Run clippy lints enabling the boolean, shortint, integer and the js wasm API
 clippy_js_wasm_api: install_rs_check_toolchain
 	RUSTFLAGS="$(RUSTFLAGS)" cargo "$(CARGO_RS_CHECK_TOOLCHAIN)" clippy \
-		--features=boolean-client-js-wasm-api,shortint-client-js-wasm-api,integer-client-js-wasm-api,high-level-client-js-wasm-api,zk-pok \
+		--features=boolean-client-js-wasm-api,shortint-client-js-wasm-api,integer-client-js-wasm-api,high-level-client-js-wasm-api,zk-pok,extended-types \
 		-p $(TFHE_SPEC) -- --no-deps -D warnings
 	RUSTFLAGS="$(RUSTFLAGS)" cargo "$(CARGO_RS_CHECK_TOOLCHAIN)" clippy \
-		--features=boolean-client-js-wasm-api,shortint-client-js-wasm-api,integer-client-js-wasm-api,high-level-client-js-wasm-api \
+		--features=boolean-client-js-wasm-api,shortint-client-js-wasm-api,integer-client-js-wasm-api,high-level-client-js-wasm-api,extended-types \
 		-p $(TFHE_SPEC) -- --no-deps -D warnings
 
 .PHONY: clippy_tasks # Run clippy lints on helper tasks crate.
@@ -388,23 +404,22 @@ clippy_tasks: install_rs_check_toolchain
 
 .PHONY: clippy_trivium # Run clippy lints on Trivium app
 clippy_trivium: install_rs_check_toolchain
-	RUSTFLAGS="$(RUSTFLAGS)" cargo "$(CARGO_RS_CHECK_TOOLCHAIN)" clippy --all-targets \
+	cd apps/trivium; RUSTFLAGS="$(RUSTFLAGS)" cargo "$(CARGO_RS_CHECK_TOOLCHAIN)" clippy --all-targets \
 		-p tfhe-trivium -- --no-deps -D warnings
 
 .PHONY: clippy_all_targets # Run clippy lints on all targets (benches, examples, etc.)
 clippy_all_targets: install_rs_check_toolchain
 	RUSTFLAGS="$(RUSTFLAGS)" cargo "$(CARGO_RS_CHECK_TOOLCHAIN)" clippy --all-targets \
-		--features=$(TARGET_ARCH_FEATURE),boolean,shortint,integer,internal-keycache,zk-pok,strings \
+		--features=boolean,shortint,integer,internal-keycache,zk-pok,strings,pbs-stats,extended-types \
 		-p $(TFHE_SPEC) -- --no-deps -D warnings
 	RUSTFLAGS="$(RUSTFLAGS)" cargo "$(CARGO_RS_CHECK_TOOLCHAIN)" clippy --all-targets \
-		--features=$(TARGET_ARCH_FEATURE),boolean,shortint,integer,internal-keycache,zk-pok,strings,experimental \
+		--features=boolean,shortint,integer,internal-keycache,zk-pok,strings,pbs-stats,extended-types,experimental \
 		-p $(TFHE_SPEC) -- --no-deps -D warnings
 
 .PHONY: clippy_tfhe_csprng # Run clippy lints on tfhe-csprng
 clippy_tfhe_csprng: install_rs_check_toolchain
 	RUSTFLAGS="$(RUSTFLAGS)" cargo "$(CARGO_RS_CHECK_TOOLCHAIN)" clippy --all-targets \
-		--features=$(TARGET_ARCH_FEATURE) \
-		-p tfhe-csprng -- --no-deps -D warnings
+		--features=parallel,software-prng -p tfhe-csprng -- --no-deps -D warnings
 
 .PHONY: clippy_zk_pok # Run clippy lints on tfhe-zk-pok
 clippy_zk_pok: install_rs_check_toolchain
@@ -418,10 +433,16 @@ clippy_versionable: install_rs_check_toolchain
 	RUSTFLAGS="$(RUSTFLAGS)" cargo "$(CARGO_RS_CHECK_TOOLCHAIN)" clippy --all-targets \
 		-p tfhe-versionable -- --no-deps -D warnings
 
+.PHONY: clippy_tfhe_lints # Run clippy lints on tfhe-lints
+clippy_tfhe_lints: install_cargo_dylint # the toolchain is selected with toolchain.toml
+	cd utils/tfhe-lints && \
+	rustup toolchain install && \
+	cargo clippy --all-targets -- --no-deps -D warnings
+
 .PHONY: clippy_all # Run all clippy targets
 clippy_all: clippy_rustdoc clippy clippy_boolean clippy_shortint clippy_integer clippy_all_targets \
 clippy_c_api clippy_js_wasm_api clippy_tasks clippy_core clippy_tfhe_csprng clippy_zk_pok clippy_trivium \
-clippy_versionable
+clippy_versionable clippy_tfhe_lints
 
 .PHONY: clippy_fast # Run main clippy targets
 clippy_fast: clippy_rustdoc clippy clippy_all_targets clippy_c_api clippy_js_wasm_api clippy_tasks \
@@ -437,73 +458,73 @@ check_rust_bindings_did_not_change:
 	cargo build -p tfhe-cuda-backend && "$(MAKE)" fmt_gpu && \
 	git diff --quiet HEAD -- backends/tfhe-cuda-backend/src/bindings.rs || \
 	( echo "Generated bindings have changed! Please run 'git add backends/tfhe-cuda-backend/src/bindings.rs' \
-	and commit the changes." && exit 1 ) 
+	and commit the changes." && exit 1 )
 
 
 .PHONY: tfhe_lints # Run custom tfhe-rs lints
-tfhe_lints: install_tfhe_lints
-	cd tfhe && RUSTFLAGS="$(RUSTFLAGS)" cargo tfhe-lints \
-		--features=$(TARGET_ARCH_FEATURE),boolean,shortint,integer,zk-pok -- -D warnings
+tfhe_lints: install_cargo_dylint
+	RUSTFLAGS="$(RUSTFLAGS)" cargo dylint --all -p tfhe --no-deps -- \
+		--features=boolean,shortint,integer,strings,zk-pok
 
 .PHONY: build_core # Build core_crypto without experimental features
 build_core: install_rs_build_toolchain install_rs_check_toolchain
 	RUSTFLAGS="$(RUSTFLAGS)" cargo $(CARGO_RS_BUILD_TOOLCHAIN) build --profile $(CARGO_PROFILE) \
-		--features=$(TARGET_ARCH_FEATURE) -p $(TFHE_SPEC)
+		-p $(TFHE_SPEC)
 	@if [[ "$(AVX512_SUPPORT)" == "ON" ]]; then \
 		RUSTFLAGS="$(RUSTFLAGS)" cargo $(CARGO_RS_CHECK_TOOLCHAIN) build --profile $(CARGO_PROFILE) \
-			--features=$(TARGET_ARCH_FEATURE),nightly-avx512 -p $(TFHE_SPEC); \
+			--features=nightly-avx512 -p $(TFHE_SPEC); \
 	fi
 
 .PHONY: build_core_experimental # Build core_crypto with experimental features
 build_core_experimental: install_rs_build_toolchain install_rs_check_toolchain
 	RUSTFLAGS="$(RUSTFLAGS)" cargo $(CARGO_RS_BUILD_TOOLCHAIN) build --profile $(CARGO_PROFILE) \
-		--features=$(TARGET_ARCH_FEATURE),experimental -p $(TFHE_SPEC)
+		--features=experimental -p $(TFHE_SPEC)
 	@if [[ "$(AVX512_SUPPORT)" == "ON" ]]; then \
 		RUSTFLAGS="$(RUSTFLAGS)" cargo $(CARGO_RS_CHECK_TOOLCHAIN) build --profile $(CARGO_PROFILE) \
-			--features=$(TARGET_ARCH_FEATURE),experimental,nightly-avx512 -p $(TFHE_SPEC); \
+			--features=experimental,nightly-avx512 -p $(TFHE_SPEC); \
 	fi
 
 .PHONY: build_boolean # Build with boolean enabled
 build_boolean: install_rs_build_toolchain
 	RUSTFLAGS="$(RUSTFLAGS)" cargo $(CARGO_RS_BUILD_TOOLCHAIN) build --profile $(CARGO_PROFILE) \
-		--features=$(TARGET_ARCH_FEATURE),boolean -p $(TFHE_SPEC) --all-targets
+		--features=boolean -p $(TFHE_SPEC) --all-targets
 
 .PHONY: build_shortint # Build with shortint enabled
 build_shortint: install_rs_build_toolchain
 	RUSTFLAGS="$(RUSTFLAGS)" cargo $(CARGO_RS_BUILD_TOOLCHAIN) build --profile $(CARGO_PROFILE) \
-		--features=$(TARGET_ARCH_FEATURE),shortint -p $(TFHE_SPEC) --all-targets
+		--features=shortint -p $(TFHE_SPEC) --all-targets
 
 .PHONY: build_integer # Build with integer enabled
 build_integer: install_rs_build_toolchain
 	RUSTFLAGS="$(RUSTFLAGS)" cargo $(CARGO_RS_BUILD_TOOLCHAIN) build --profile $(CARGO_PROFILE) \
-		--features=$(TARGET_ARCH_FEATURE),integer -p $(TFHE_SPEC) --all-targets
+		--features=integer -p $(TFHE_SPEC) --all-targets
 
 .PHONY: build_tfhe_full # Build with boolean, shortint and integer enabled
 build_tfhe_full: install_rs_build_toolchain
 	RUSTFLAGS="$(RUSTFLAGS)" cargo $(CARGO_RS_BUILD_TOOLCHAIN) build --profile $(CARGO_PROFILE) \
-		--features=$(TARGET_ARCH_FEATURE),boolean,shortint,integer -p $(TFHE_SPEC) --all-targets
+		--features=boolean,shortint,integer -p $(TFHE_SPEC) --all-targets
 
 .PHONY: build_tfhe_coverage # Build with test coverage enabled
 build_tfhe_coverage: install_rs_build_toolchain
 	RUSTFLAGS="$(RUSTFLAGS) --cfg tarpaulin" cargo $(CARGO_RS_BUILD_TOOLCHAIN) build --profile $(CARGO_PROFILE) \
-		--features=$(TARGET_ARCH_FEATURE),boolean,shortint,integer,internal-keycache -p $(TFHE_SPEC) --tests
+		--features=boolean,shortint,integer,internal-keycache -p $(TFHE_SPEC) --tests
 
 .PHONY: build_c_api # Build the C API for boolean, shortint and integer
 build_c_api: install_rs_check_toolchain
 	RUSTFLAGS="$(RUSTFLAGS)" cargo $(CARGO_RS_CHECK_TOOLCHAIN) build --profile $(CARGO_PROFILE) \
-		--features=$(TARGET_ARCH_FEATURE),boolean-c-api,shortint-c-api,high-level-c-api,zk-pok \
+		--features=boolean-c-api,shortint-c-api,high-level-c-api,zk-pok,extended-types \
 		-p $(TFHE_SPEC)
 
 .PHONY: build_c_api_gpu # Build the C API for boolean, shortint and integer
 build_c_api_gpu: install_rs_check_toolchain
 	RUSTFLAGS="$(RUSTFLAGS)" cargo $(CARGO_RS_CHECK_TOOLCHAIN) build --profile $(CARGO_PROFILE) \
-		--features=$(TARGET_ARCH_FEATURE),boolean-c-api,shortint-c-api,high-level-c-api,zk-pok,gpu \
+		--features=boolean-c-api,shortint-c-api,high-level-c-api,zk-pok,extended-types,gpu \
 		-p $(TFHE_SPEC)
 
 .PHONY: build_c_api_experimental_deterministic_fft # Build the C API for boolean, shortint and integer with experimental deterministic FFT
 build_c_api_experimental_deterministic_fft: install_rs_check_toolchain
 	RUSTFLAGS="$(RUSTFLAGS)" cargo $(CARGO_RS_CHECK_TOOLCHAIN) build --profile $(CARGO_PROFILE) \
-		--features=$(TARGET_ARCH_FEATURE),boolean-c-api,shortint-c-api,high-level-c-api,zk-pok,experimental-force_fft_algo_dif4 \
+		--features=boolean-c-api,shortint-c-api,high-level-c-api,zk-pok,experimental-force_fft_algo_dif4 \
 		-p $(TFHE_SPEC)
 
 .PHONY: build_web_js_api # Build the js API targeting the web browser
@@ -511,17 +532,17 @@ build_web_js_api: install_rs_build_toolchain install_wasm_pack
 	cd tfhe && \
 	RUSTFLAGS="$(WASM_RUSTFLAGS)" rustup run "$(RS_BUILD_TOOLCHAIN)" \
 		wasm-pack build --release --target=web \
-		-- --features=boolean-client-js-wasm-api,shortint-client-js-wasm-api,integer-client-js-wasm-api,zk-pok
+		-- --features=boolean-client-js-wasm-api,shortint-client-js-wasm-api,integer-client-js-wasm-api,zk-pok,extended-types
 
 .PHONY: build_web_js_api_parallel # Build the js API targeting the web browser with parallelism support
 build_web_js_api_parallel: install_rs_check_toolchain install_wasm_pack
 	cd tfhe && \
 	rustup component add rust-src --toolchain $(RS_CHECK_TOOLCHAIN) && \
-	RUSTFLAGS="$(WASM_RUSTFLAGS) -C target-feature=+atomics,+bulk-memory,+mutable-globals" rustup run $(RS_CHECK_TOOLCHAIN) \
+	RUSTFLAGS="$(WASM_RUSTFLAGS) -C target-feature=+atomics,+bulk-memory" rustup run $(RS_CHECK_TOOLCHAIN) \
 		wasm-pack build --release --target=web \
-		-- --features=boolean-client-js-wasm-api,shortint-client-js-wasm-api,integer-client-js-wasm-api,parallel-wasm-api,zk-pok \
+		-- --features=boolean-client-js-wasm-api,shortint-client-js-wasm-api,integer-client-js-wasm-api,parallel-wasm-api,zk-pok,extended-types \
 		-Z build-std=panic_abort,std && \
-	find pkg/snippets -type f -iname workerHelpers.worker.js -exec sed -i "s|from '..\/..\/..\/';|from '..\/..\/..\/tfhe.js';|" {} \;
+	find pkg/snippets -type f -iname workerHelpers.js -exec sed -i "s|const pkg = await import('..\/..\/..');|const pkg = await import('..\/..\/..\/tfhe.js');|" {} \;
 	jq '.files += ["snippets"]' tfhe/pkg/package.json > tmp_pkg.json && mv -f tmp_pkg.json tfhe/pkg/package.json
 
 .PHONY: build_node_js_api # Build the js API targeting nodejs
@@ -529,20 +550,20 @@ build_node_js_api: install_rs_build_toolchain install_wasm_pack
 	cd tfhe && \
 	RUSTFLAGS="$(WASM_RUSTFLAGS)" rustup run "$(RS_BUILD_TOOLCHAIN)" \
 		wasm-pack build --release --target=nodejs \
-		-- --features=boolean-client-js-wasm-api,shortint-client-js-wasm-api,integer-client-js-wasm-api,zk-pok
+		-- --features=boolean-client-js-wasm-api,shortint-client-js-wasm-api,integer-client-js-wasm-api,zk-pok,extended-types
 
 .PHONY: build_tfhe_csprng # Build tfhe_csprng
 build_tfhe_csprng: install_rs_build_toolchain
 	RUSTFLAGS="$(RUSTFLAGS)" cargo $(CARGO_RS_BUILD_TOOLCHAIN) build --profile $(CARGO_PROFILE) \
-		--features=$(TARGET_ARCH_FEATURE) -p tfhe-csprng --all-targets
+		-p tfhe-csprng --all-targets
 
 .PHONY: test_core_crypto # Run the tests of the core_crypto module including experimental ones
 test_core_crypto: install_rs_build_toolchain install_rs_check_toolchain
 	RUSTFLAGS="$(RUSTFLAGS)" cargo $(CARGO_RS_BUILD_TOOLCHAIN) test --profile $(CARGO_PROFILE) \
-		--features=$(TARGET_ARCH_FEATURE),experimental,zk-pok -p $(TFHE_SPEC) -- core_crypto::
+		--features=experimental,zk-pok -p $(TFHE_SPEC) -- core_crypto::
 	@if [[ "$(AVX512_SUPPORT)" == "ON" ]]; then \
 		RUSTFLAGS="$(RUSTFLAGS)" cargo $(CARGO_RS_CHECK_TOOLCHAIN) test --profile $(CARGO_PROFILE) \
-			--features=$(TARGET_ARCH_FEATURE),experimental,zk-pok,nightly-avx512 -p $(TFHE_SPEC) -- core_crypto::; \
+			--features=experimental,zk-pok,nightly-avx512 -p $(TFHE_SPEC) -- core_crypto::; \
 	fi
 
 .PHONY: test_core_crypto_cov # Run the tests of the core_crypto module with code coverage
@@ -550,13 +571,13 @@ test_core_crypto_cov: install_rs_build_toolchain install_rs_check_toolchain inst
 	RUSTFLAGS="$(RUSTFLAGS)" cargo $(CARGO_RS_BUILD_TOOLCHAIN) tarpaulin --profile $(CARGO_PROFILE) \
 		--out xml --output-dir coverage/core_crypto --line --engine llvm --timeout 500 \
 		--implicit-test-threads $(COVERAGE_EXCLUDED_FILES) \
-		--features=$(TARGET_ARCH_FEATURE),experimental,internal-keycache \
+		--features=experimental,internal-keycache \
 		-p $(TFHE_SPEC) -- core_crypto::
 	@if [[ "$(AVX512_SUPPORT)" == "ON" ]]; then \
 		RUSTFLAGS="$(RUSTFLAGS)" cargo $(CARGO_RS_CHECK_TOOLCHAIN) tarpaulin --profile $(CARGO_PROFILE) \
 			--out xml --output-dir coverage/core_crypto_avx512 --line --engine llvm --timeout 500 \
 			--implicit-test-threads $(COVERAGE_EXCLUDED_FILES) \
-			--features=$(TARGET_ARCH_FEATURE),experimental,internal-keycache,nightly-avx512 \
+			--features=experimental,internal-keycache,nightly-avx512 \
 			-p $(TFHE_SPEC) -- -Z unstable-options --report-time core_crypto::; \
 	fi
 
@@ -574,30 +595,38 @@ test_gpu: test_core_crypto_gpu test_integer_gpu test_cuda_backend
 .PHONY: test_core_crypto_gpu # Run the tests of the core_crypto module including experimental on the gpu backend
 test_core_crypto_gpu: install_rs_build_toolchain
 	RUSTFLAGS="$(RUSTFLAGS)" cargo $(CARGO_RS_BUILD_TOOLCHAIN) test --profile $(CARGO_PROFILE) \
-		--features=$(TARGET_ARCH_FEATURE),gpu -p $(TFHE_SPEC) -- core_crypto::gpu::
+		--features=gpu -p $(TFHE_SPEC) -- core_crypto::gpu::
 	RUSTFLAGS="$(RUSTFLAGS)" cargo $(CARGO_RS_BUILD_TOOLCHAIN) test --doc --profile $(CARGO_PROFILE) \
-		--features=$(TARGET_ARCH_FEATURE),gpu -p $(TFHE_SPEC) -- core_crypto::gpu::
+		--features=gpu -p $(TFHE_SPEC) -- core_crypto::gpu::
 
 .PHONY: test_integer_gpu # Run the tests of the integer module including experimental on the gpu backend
 test_integer_gpu: install_rs_build_toolchain
 	RUSTFLAGS="$(RUSTFLAGS)" cargo $(CARGO_RS_BUILD_TOOLCHAIN) test --profile $(CARGO_PROFILE) \
-		--features=$(TARGET_ARCH_FEATURE),integer,gpu -p $(TFHE_SPEC) -- integer::gpu::server_key:: --test-threads=6
+		--features=integer,gpu -p $(TFHE_SPEC) -- integer::gpu::server_key:: --test-threads=4
 	RUSTFLAGS="$(RUSTFLAGS)" cargo $(CARGO_RS_BUILD_TOOLCHAIN) test --doc --profile $(CARGO_PROFILE) \
-		--features=$(TARGET_ARCH_FEATURE),integer,gpu -p $(TFHE_SPEC) -- integer::gpu::server_key::
+		--features=integer,gpu -p $(TFHE_SPEC) -- integer::gpu::server_key::
+
+.PHONY: test_integer_long_run_gpu # Run the long run integer tests on the gpu backend
+test_integer_long_run_gpu: install_rs_check_toolchain install_cargo_nextest
+	BIG_TESTS_INSTANCE="$(BIG_TESTS_INSTANCE)" \
+	LONG_TESTS=TRUE \
+		./scripts/integer-tests.sh --rust-toolchain $(CARGO_RS_BUILD_TOOLCHAIN) \
+		--cargo-profile "$(CARGO_PROFILE)" --avx512-support "$(AVX512_SUPPORT)" \
+		--tfhe-package "$(TFHE_SPEC)" --backend "gpu"
 
 .PHONY: test_integer_compression
 test_integer_compression: install_rs_build_toolchain
 	RUSTFLAGS="$(RUSTFLAGS)" cargo $(CARGO_RS_BUILD_TOOLCHAIN) test --profile $(CARGO_PROFILE) \
-		--features=$(TARGET_ARCH_FEATURE),integer -p $(TFHE_SPEC) -- integer::ciphertext::compressed_ciphertext_list::tests::
+		--features=integer -p $(TFHE_SPEC) -- integer::ciphertext::compressed_ciphertext_list::tests::
 	RUSTFLAGS="$(RUSTFLAGS)" cargo $(CARGO_RS_BUILD_TOOLCHAIN) test --doc --profile $(CARGO_PROFILE) \
-		--features=$(TARGET_ARCH_FEATURE),integer -p $(TFHE_SPEC) -- integer::ciphertext::compress
+		--features=integer -p $(TFHE_SPEC) -- integer::ciphertext::compress
 
 .PHONY: test_integer_compression_gpu
 test_integer_compression_gpu: install_rs_build_toolchain
 	RUSTFLAGS="$(RUSTFLAGS)" cargo $(CARGO_RS_BUILD_TOOLCHAIN) test --profile $(CARGO_PROFILE) \
-		--features=$(TARGET_ARCH_FEATURE),integer,gpu -p $(TFHE_SPEC) -- integer::gpu::ciphertext::compressed_ciphertext_list::tests::
+		--features=integer,gpu -p $(TFHE_SPEC) -- integer::gpu::ciphertext::compressed_ciphertext_list::tests::
 	RUSTFLAGS="$(RUSTFLAGS)" cargo $(CARGO_RS_BUILD_TOOLCHAIN) test --doc --profile $(CARGO_PROFILE) \
-		--features=$(TARGET_ARCH_FEATURE),integer,gpu -p $(TFHE_SPEC) -- integer::gpu::ciphertext::compress
+		--features=integer,gpu -p $(TFHE_SPEC) -- integer::gpu::ciphertext::compress
 
 .PHONY: test_integer_gpu_ci # Run the tests for integer ci on gpu backend
 test_integer_gpu_ci: install_rs_check_toolchain install_cargo_nextest
@@ -656,20 +685,20 @@ test_signed_integer_multi_bit_gpu_ci: install_rs_check_toolchain install_cargo_n
 .PHONY: test_boolean # Run the tests of the boolean module
 test_boolean: install_rs_build_toolchain
 	RUSTFLAGS="$(RUSTFLAGS)" cargo $(CARGO_RS_BUILD_TOOLCHAIN) test --profile $(CARGO_PROFILE) \
-		--features=$(TARGET_ARCH_FEATURE),boolean -p $(TFHE_SPEC) -- boolean::
+		--features=boolean -p $(TFHE_SPEC) -- boolean::
 
 .PHONY: test_boolean_cov # Run the tests of the boolean module with code coverage
 test_boolean_cov: install_rs_check_toolchain install_tarpaulin
 	RUSTFLAGS="$(RUSTFLAGS)" cargo $(CARGO_RS_CHECK_TOOLCHAIN) tarpaulin --profile $(CARGO_PROFILE) \
 		--out xml --output-dir coverage/boolean --line --engine llvm --timeout 500 \
 		$(COVERAGE_EXCLUDED_FILES) \
-		--features=$(TARGET_ARCH_FEATURE),boolean,internal-keycache \
+		--features=boolean,internal-keycache \
 		-p $(TFHE_SPEC) -- -Z unstable-options --report-time boolean::
 
 .PHONY: test_c_api_rs # Run the rust tests for the C API
 test_c_api_rs: install_rs_check_toolchain
 	RUSTFLAGS="$(RUSTFLAGS)" cargo $(CARGO_RS_CHECK_TOOLCHAIN) test --profile $(CARGO_PROFILE) \
-		--features=$(TARGET_ARCH_FEATURE),boolean-c-api,shortint-c-api,high-level-c-api \
+		--features=boolean-c-api,shortint-c-api,high-level-c-api \
 		-p $(TFHE_SPEC) \
 		c_api
 
@@ -701,14 +730,14 @@ test_shortint_multi_bit_ci: install_rs_build_toolchain install_cargo_nextest
 .PHONY: test_shortint # Run all the tests for shortint
 test_shortint: install_rs_build_toolchain
 	RUSTFLAGS="$(RUSTFLAGS)" cargo $(CARGO_RS_BUILD_TOOLCHAIN) test --profile $(CARGO_PROFILE) \
-		--features=$(TARGET_ARCH_FEATURE),shortint,internal-keycache -p $(TFHE_SPEC) -- shortint::
+		--features=shortint,internal-keycache -p $(TFHE_SPEC) -- shortint::
 
 .PHONY: test_shortint_cov # Run the tests of the shortint module with code coverage
 test_shortint_cov: install_rs_check_toolchain install_tarpaulin
 	RUSTFLAGS="$(RUSTFLAGS)" cargo $(CARGO_RS_CHECK_TOOLCHAIN) tarpaulin --profile $(CARGO_PROFILE) \
 		--out xml --output-dir coverage/shortint --line --engine llvm --timeout 500 \
 		$(COVERAGE_EXCLUDED_FILES) \
-		--features=$(TARGET_ARCH_FEATURE),shortint,internal-keycache \
+		--features=shortint,internal-keycache \
 		-p $(TFHE_SPEC) -- -Z unstable-options --report-time shortint::
 
 .PHONY: test_integer_ci # Run the tests for integer ci
@@ -765,20 +794,28 @@ test_signed_integer_multi_bit_ci: install_rs_check_toolchain install_cargo_nexte
 		--cargo-profile "$(CARGO_PROFILE)" --multi-bit --avx512-support "$(AVX512_SUPPORT)" \
 		--signed-only --tfhe-package "$(TFHE_SPEC)"
 
+.PHONY: test_integer_long_run # Run the long run integer tests
+test_integer_long_run: install_rs_check_toolchain install_cargo_nextest
+	BIG_TESTS_INSTANCE="$(BIG_TESTS_INSTANCE)" \
+	LONG_TESTS=TRUE \
+		./scripts/integer-tests.sh --rust-toolchain $(CARGO_RS_BUILD_TOOLCHAIN) \
+		--cargo-profile "$(CARGO_PROFILE)" --avx512-support "$(AVX512_SUPPORT)" \
+		--tfhe-package "$(TFHE_SPEC)"
+
 .PHONY: test_safe_serialization # Run the tests for safe serialization
 test_safe_serialization: install_rs_build_toolchain install_cargo_nextest
 	RUSTFLAGS="$(RUSTFLAGS)" cargo $(CARGO_RS_BUILD_TOOLCHAIN) test --profile $(CARGO_PROFILE) \
-		--features=$(TARGET_ARCH_FEATURE),boolean,shortint,integer,internal-keycache -p $(TFHE_SPEC) -- safe_serialization::
+		--features=boolean,shortint,integer,internal-keycache -p $(TFHE_SPEC) -- safe_serialization::
 
 .PHONY: test_zk # Run the tests for the zk module of the TFHE-rs crate
 test_zk: install_rs_build_toolchain install_cargo_nextest
 	RUSTFLAGS="$(RUSTFLAGS)" cargo $(CARGO_RS_BUILD_TOOLCHAIN) test --profile $(CARGO_PROFILE) \
-		--features=$(TARGET_ARCH_FEATURE),shortint,zk-pok -p $(TFHE_SPEC) -- zk::
+		--features=shortint,zk-pok -p $(TFHE_SPEC) -- zk::
 
 .PHONY: test_integer # Run all the tests for integer
 test_integer: install_rs_build_toolchain
 	RUSTFLAGS="$(RUSTFLAGS)" cargo $(CARGO_RS_BUILD_TOOLCHAIN) test --profile $(CARGO_PROFILE) \
-		--features=$(TARGET_ARCH_FEATURE),integer,internal-keycache -p $(TFHE_SPEC) -- integer::
+		--features=integer,internal-keycache -p $(TFHE_SPEC) -- integer::
 
 .PHONY: test_integer_cov # Run the tests of the integer module with code coverage
 test_integer_cov: install_rs_check_toolchain install_tarpaulin
@@ -786,38 +823,38 @@ test_integer_cov: install_rs_check_toolchain install_tarpaulin
 		--out xml --output-dir coverage/integer --line --engine llvm --timeout 500 \
 		--implicit-test-threads \
 		--exclude-files $(COVERAGE_EXCLUDED_FILES) \
-		--features=$(TARGET_ARCH_FEATURE),integer,internal-keycache \
+		--features=integer,internal-keycache \
 		-p $(TFHE_SPEC) -- -Z unstable-options --report-time integer::
 
 .PHONY: test_high_level_api # Run all the tests for high_level_api
 test_high_level_api: install_rs_build_toolchain
 	RUSTFLAGS="$(RUSTFLAGS)" cargo $(CARGO_RS_BUILD_TOOLCHAIN) test --profile $(CARGO_PROFILE) \
-		--features=$(TARGET_ARCH_FEATURE),boolean,shortint,integer,internal-keycache,zk-pok -p $(TFHE_SPEC) \
+		--features=boolean,shortint,integer,internal-keycache,zk-pok,strings -p $(TFHE_SPEC) \
 		-- high_level_api::
 
 test_high_level_api_gpu: install_rs_build_toolchain install_cargo_nextest
 	RUSTFLAGS="$(RUSTFLAGS)" cargo $(CARGO_RS_BUILD_TOOLCHAIN) nextest run --cargo-profile $(CARGO_PROFILE) \
-		--features=$(TARGET_ARCH_FEATURE),integer,internal-keycache,gpu -p $(TFHE_SPEC) \
+		--features=integer,internal-keycache,gpu -p $(TFHE_SPEC) \
 		-E "test(/high_level_api::.*gpu.*/)"
 
 .PHONY: test_strings # Run the tests for strings ci
 test_strings: install_rs_build_toolchain
 	RUSTFLAGS="$(RUSTFLAGS)" cargo $(CARGO_RS_BUILD_TOOLCHAIN) test --profile $(CARGO_PROFILE) \
-		--features=$(TARGET_ARCH_FEATURE),shortint,integer,strings -p $(TFHE_SPEC) \
+		--features=shortint,integer,strings -p $(TFHE_SPEC) \
 		-- strings::
 
 
 .PHONY: test_user_doc # Run tests from the .md documentation
 test_user_doc: install_rs_build_toolchain
 	RUSTFLAGS="$(RUSTFLAGS)" cargo $(CARGO_RS_BUILD_TOOLCHAIN) test --profile $(CARGO_PROFILE) --doc \
-		--features=$(TARGET_ARCH_FEATURE),boolean,shortint,integer,internal-keycache,pbs-stats,zk-pok \
+		--features=boolean,shortint,integer,internal-keycache,pbs-stats,zk-pok,strings \
 		-p $(TFHE_SPEC) \
 		-- test_user_docs::
 
 .PHONY: test_user_doc_gpu # Run tests for GPU from the .md documentation
 test_user_doc_gpu: install_rs_build_toolchain
 	RUSTFLAGS="$(RUSTFLAGS)" cargo $(CARGO_RS_BUILD_TOOLCHAIN) test --profile $(CARGO_PROFILE) --doc \
-		--features=$(TARGET_ARCH_FEATURE),boolean,shortint,integer,internal-keycache,gpu,zk-pok -p $(TFHE_SPEC) \
+		--features=boolean,shortint,integer,internal-keycache,gpu,zk-pok -p $(TFHE_SPEC) \
 		-- test_user_docs::
 
 
@@ -825,32 +862,30 @@ test_user_doc_gpu: install_rs_build_toolchain
 .PHONY: test_regex_engine # Run tests for regex_engine example
 test_regex_engine: install_rs_build_toolchain
 	RUSTFLAGS="$(RUSTFLAGS)" cargo $(CARGO_RS_BUILD_TOOLCHAIN) test --profile $(CARGO_PROFILE) \
-		--example regex_engine \
-		--features=$(TARGET_ARCH_FEATURE),integer
+		--example regex_engine --features=integer
 
 .PHONY: test_sha256_bool # Run tests for sha256_bool example
 test_sha256_bool: install_rs_build_toolchain
 	RUSTFLAGS="$(RUSTFLAGS)" cargo $(CARGO_RS_BUILD_TOOLCHAIN) test --profile $(CARGO_PROFILE) \
-		--example sha256_bool \
-		--features=$(TARGET_ARCH_FEATURE),boolean
+		--example sha256_bool --features=boolean
 
 .PHONY: test_examples # Run tests for examples
 test_examples: test_sha256_bool test_regex_engine
 
 .PHONY: test_trivium # Run tests for trivium
 test_trivium: install_rs_build_toolchain
-	RUSTFLAGS="$(RUSTFLAGS)" cargo $(CARGO_RS_BUILD_TOOLCHAIN) test --profile $(CARGO_PROFILE) \
+	cd apps/trivium; RUSTFLAGS="$(RUSTFLAGS)" cargo $(CARGO_RS_BUILD_TOOLCHAIN) test --profile $(CARGO_PROFILE) \
 		-p tfhe-trivium -- --test-threads=1 trivium::
 
 .PHONY: test_kreyvium # Run tests for kreyvium
 test_kreyvium: install_rs_build_toolchain
-	RUSTFLAGS="$(RUSTFLAGS)" cargo $(CARGO_RS_BUILD_TOOLCHAIN) test --profile $(CARGO_PROFILE) \
+	cd apps/trivium; RUSTFLAGS="$(RUSTFLAGS)" cargo $(CARGO_RS_BUILD_TOOLCHAIN) test --profile $(CARGO_PROFILE) \
 		-p tfhe-trivium -- --test-threads=1 kreyvium::
 
 .PHONY: test_tfhe_csprng # Run tfhe-csprng tests
 test_tfhe_csprng: install_rs_build_toolchain
 	RUSTFLAGS="$(RUSTFLAGS)" cargo $(CARGO_RS_BUILD_TOOLCHAIN) test --profile $(CARGO_PROFILE) \
-		--features=$(TARGET_ARCH_FEATURE) -p tfhe-csprng
+		-p tfhe-csprng
 
 .PHONY: test_zk_pok # Run tfhe-zk-pok tests
 test_zk_pok: install_rs_build_toolchain
@@ -868,23 +903,29 @@ test_zk_wasm_x86_compat_ci: check_nvm_installed
 test_zk_wasm_x86_compat: install_rs_build_toolchain build_node_js_api
 	cd tfhe/tests/zk_wasm_x86_test && npm install
 	RUSTFLAGS="$(RUSTFLAGS)" cargo $(CARGO_RS_BUILD_TOOLCHAIN) test --profile $(CARGO_PROFILE) \
-		-p tfhe --test zk_wasm_x86_test --features=$(TARGET_ARCH_FEATURE),integer,zk-pok
+		-p tfhe --test zk_wasm_x86_test --features=integer,zk-pok
 
 .PHONY: test_versionable # Run tests for tfhe-versionable subcrate
 test_versionable: install_rs_build_toolchain
 	RUSTFLAGS="$(RUSTFLAGS)" cargo $(CARGO_RS_BUILD_TOOLCHAIN) test --profile $(CARGO_PROFILE) \
 		--all-targets -p tfhe-versionable
 
+.PHONY: test_tfhe_lints # Run test on tfhe-lints
+test_tfhe_lints: install_cargo_dylint
+	cd utils/tfhe-lints && \
+	rustup toolchain install && \
+	cargo test
+
 # The backward compat data repo holds historical binary data but also rust code to generate and load them.
 # Here we use the "patch" functionality of Cargo to make sure the repo used for the data is the same as the one used for the code.
 .PHONY: test_backward_compatibility_ci
 test_backward_compatibility_ci: install_rs_build_toolchain
 	TFHE_BACKWARD_COMPAT_DATA_DIR="$(BACKWARD_COMPAT_DATA_DIR)" RUSTFLAGS="$(RUSTFLAGS)" cargo $(CARGO_RS_BUILD_TOOLCHAIN) test --profile $(CARGO_PROFILE) \
-		--config "patch.'$(BACKWARD_COMPAT_DATA_URL)'.$(BACKWARD_COMPAT_DATA_PROJECT).path=\"tfhe/$(BACKWARD_COMPAT_DATA_DIR)\"" \
-		--features=$(TARGET_ARCH_FEATURE),shortint,integer,zk-pok -p $(TFHE_SPEC) test_backward_compatibility -- --nocapture
+		--config "patch.'$(BACKWARD_COMPAT_DATA_URL)'.$(BACKWARD_COMPAT_DATA_PROJECT).path=\"tests/$(BACKWARD_COMPAT_DATA_DIR)\"" \
+		--features=shortint,integer,zk-pok -p tests test_backward_compatibility -- --nocapture
 
 .PHONY: test_backward_compatibility # Same as test_backward_compatibility_ci but tries to clone the data repo first if needed
-test_backward_compatibility: tfhe/$(BACKWARD_COMPAT_DATA_DIR) test_backward_compatibility_ci
+test_backward_compatibility: tests/$(BACKWARD_COMPAT_DATA_DIR) test_backward_compatibility_ci
 
 .PHONY: backward_compat_branch # Prints the required backward compatibility branch
 backward_compat_branch:
@@ -896,7 +937,7 @@ doc: install_rs_check_toolchain
 	DOCS_RS=1 \
 	RUSTDOCFLAGS="--html-in-header katex-header.html" \
 	cargo "$(CARGO_RS_CHECK_TOOLCHAIN)" doc \
-		--features=$(TARGET_ARCH_FEATURE),boolean,shortint,integer,gpu,internal-keycache,experimental,zk-pok --no-deps -p $(TFHE_SPEC)
+		--features=boolean,shortint,integer,strings,gpu,internal-keycache,experimental,zk-pok --no-deps -p $(TFHE_SPEC)
 
 .PHONY: docs # Build rust doc alias for doc
 docs: doc
@@ -907,7 +948,7 @@ lint_doc: install_rs_check_toolchain
 	DOCS_RS=1 \
 	RUSTDOCFLAGS="--html-in-header katex-header.html -Dwarnings" \
 	cargo "$(CARGO_RS_CHECK_TOOLCHAIN)" doc \
-		--features=$(TARGET_ARCH_FEATURE),boolean,shortint,integer,gpu,internal-keycache,experimental,zk-pok -p $(TFHE_SPEC) --no-deps
+		--features=boolean,shortint,integer,strings,gpu,internal-keycache,experimental,zk-pok -p $(TFHE_SPEC) --no-deps
 
 .PHONY: lint_docs # Build rust doc with linting enabled alias for lint_doc
 lint_docs: lint_doc
@@ -933,10 +974,14 @@ check_intra_md_links: install_mlc
 check_md_links: install_mlc
 	mlc --match-file-extension tfhe/docs
 
+.PHONY: check_parameter_export_ok # Checks exported "current" shortint parameter module is correct
+check_parameter_export_ok:
+	python3 ./scripts/check_current_param_export.py
+
 .PHONY: check_compile_tests # Build tests in debug without running them
 check_compile_tests: install_rs_build_toolchain
 	RUSTFLAGS="$(RUSTFLAGS)" cargo $(CARGO_RS_BUILD_TOOLCHAIN) test --no-run \
-		--features=$(TARGET_ARCH_FEATURE),experimental,boolean,shortint,integer,internal-keycache \
+		--features=experimental,boolean,shortint,integer,internal-keycache \
 		-p $(TFHE_SPEC)
 
 	@if [[ "$(OS)" == "Linux" || "$(OS)" == "Darwin" ]]; then \
@@ -947,7 +992,7 @@ check_compile_tests: install_rs_build_toolchain
 .PHONY: check_compile_tests_benches_gpu # Build tests in debug without running them
 check_compile_tests_benches_gpu: install_rs_build_toolchain
 	RUSTFLAGS="$(RUSTFLAGS)" cargo $(CARGO_RS_BUILD_TOOLCHAIN) test --no-run \
-		--features=$(TARGET_ARCH_FEATURE),experimental,boolean,shortint,integer,internal-keycache,gpu \
+		--features=experimental,boolean,shortint,integer,internal-keycache,gpu \
 		-p $(TFHE_SPEC)
 	mkdir -p "$(TFHECUDA_BUILD)" && \
 		cd "$(TFHECUDA_BUILD)" && \
@@ -1026,42 +1071,49 @@ dieharder_csprng: install_dieharder build_tfhe_csprng
 .PHONY: print_doc_bench_parameters # Print parameters used in doc benchmarks
 print_doc_bench_parameters:
 	RUSTFLAGS="" cargo run --example print_doc_bench_parameters \
-	--features=$(TARGET_ARCH_FEATURE),shortint,internal-keycache -p tfhe
+	--features=shortint,internal-keycache -p tfhe
 
 .PHONY: bench_integer # Run benchmarks for unsigned integer
 bench_integer: install_rs_check_toolchain
 	RUSTFLAGS="$(RUSTFLAGS)" __TFHE_RS_BENCH_OP_FLAVOR=$(BENCH_OP_FLAVOR) __TFHE_RS_FAST_BENCH=$(FAST_BENCH) __TFHE_RS_BENCH_TYPE=$(BENCH_TYPE) \
 	cargo $(CARGO_RS_CHECK_TOOLCHAIN) bench \
 	--bench integer-bench \
-	--features=$(TARGET_ARCH_FEATURE),integer,internal-keycache,nightly-avx512 -p $(TFHE_SPEC) --
+	--features=integer,internal-keycache,nightly-avx512,pbs-stats -p $(TFHE_SPEC) --
 
 .PHONY: bench_signed_integer # Run benchmarks for signed integer
 bench_signed_integer: install_rs_check_toolchain
 	RUSTFLAGS="$(RUSTFLAGS)" __TFHE_RS_BENCH_OP_FLAVOR=$(BENCH_OP_FLAVOR) __TFHE_RS_FAST_BENCH=$(FAST_BENCH) __TFHE_RS_BENCH_TYPE=$(BENCH_TYPE) \
 	cargo $(CARGO_RS_CHECK_TOOLCHAIN) bench \
 	--bench integer-signed-bench \
-	--features=$(TARGET_ARCH_FEATURE),integer,internal-keycache,nightly-avx512 -p $(TFHE_SPEC) --
+	--features=integer,internal-keycache,nightly-avx512,pbs-stats -p $(TFHE_SPEC) --
 
 .PHONY: bench_integer_gpu # Run benchmarks for integer on GPU backend
 bench_integer_gpu: install_rs_check_toolchain
 	RUSTFLAGS="$(RUSTFLAGS)" __TFHE_RS_BENCH_OP_FLAVOR=$(BENCH_OP_FLAVOR) __TFHE_RS_FAST_BENCH=$(FAST_BENCH) __TFHE_RS_BENCH_TYPE=$(BENCH_TYPE) \
 	cargo $(CARGO_RS_CHECK_TOOLCHAIN) bench \
 	--bench integer-bench \
-	--features=$(TARGET_ARCH_FEATURE),integer,gpu,internal-keycache,nightly-avx512 -p $(TFHE_SPEC) --
+	--features=integer,gpu,internal-keycache,nightly-avx512,pbs-stats -p $(TFHE_SPEC) --
+
+.PHONY: bench_signed_integer_gpu # Run benchmarks for signed integer on GPU backend
+bench_signed_integer_gpu: install_rs_check_toolchain
+	RUSTFLAGS="$(RUSTFLAGS)" __TFHE_RS_BENCH_OP_FLAVOR=$(BENCH_OP_FLAVOR) __TFHE_RS_FAST_BENCH=$(FAST_BENCH) __TFHE_RS_BENCH_TYPE=$(BENCH_TYPE) \
+	cargo $(CARGO_RS_CHECK_TOOLCHAIN) bench \
+	--bench integer-signed-bench \
+	--features=integer,gpu,internal-keycache,nightly-avx512,pbs-stats -p $(TFHE_SPEC) --
 
 .PHONY: bench_integer_compression # Run benchmarks for unsigned integer compression
 bench_integer_compression: install_rs_check_toolchain
 	RUSTFLAGS="$(RUSTFLAGS)" __TFHE_RS_BENCH_TYPE=$(BENCH_TYPE) \
 	cargo $(CARGO_RS_CHECK_TOOLCHAIN) bench \
 	--bench	glwe_packing_compression-integer-bench \
-	--features=$(TARGET_ARCH_FEATURE),integer,internal-keycache,nightly-avx512 -p $(TFHE_SPEC) --
+	--features=integer,internal-keycache,nightly-avx512,pbs-stats -p $(TFHE_SPEC) --
 
 .PHONY: bench_integer_compression_gpu
 bench_integer_compression_gpu: install_rs_check_toolchain
 	RUSTFLAGS="$(RUSTFLAGS)" __TFHE_RS_BENCH_TYPE=$(BENCH_TYPE) \
 	cargo $(CARGO_RS_CHECK_TOOLCHAIN) bench \
 	--bench	glwe_packing_compression-integer-bench \
-	--features=$(TARGET_ARCH_FEATURE),integer,internal-keycache,gpu -p $(TFHE_SPEC) --
+	--features=integer,internal-keycache,gpu,pbs-stats -p $(TFHE_SPEC) --
 
 .PHONY: bench_integer_multi_bit # Run benchmarks for unsigned integer using multi-bit parameters
 bench_integer_multi_bit: install_rs_check_toolchain
@@ -1069,7 +1121,7 @@ bench_integer_multi_bit: install_rs_check_toolchain
 	__TFHE_RS_BENCH_OP_FLAVOR=$(BENCH_OP_FLAVOR) __TFHE_RS_FAST_BENCH=$(FAST_BENCH) \
 	cargo $(CARGO_RS_CHECK_TOOLCHAIN) bench \
 	--bench integer-bench \
-	--features=$(TARGET_ARCH_FEATURE),integer,internal-keycache,nightly-avx512 -p $(TFHE_SPEC) --
+	--features=integer,internal-keycache,nightly-avx512,pbs-stats -p $(TFHE_SPEC) --
 
 .PHONY: bench_signed_integer_multi_bit # Run benchmarks for signed integer using multi-bit parameters
 bench_signed_integer_multi_bit: install_rs_check_toolchain
@@ -1077,7 +1129,7 @@ bench_signed_integer_multi_bit: install_rs_check_toolchain
 	__TFHE_RS_BENCH_OP_FLAVOR=$(BENCH_OP_FLAVOR) __TFHE_RS_FAST_BENCH=$(FAST_BENCH) \
 	cargo $(CARGO_RS_CHECK_TOOLCHAIN) bench \
 	--bench integer-signed-bench \
-	--features=$(TARGET_ARCH_FEATURE),integer,internal-keycache,nightly-avx512 -p $(TFHE_SPEC) --
+	--features=integer,internal-keycache,nightly-avx512,pbs-stats -p $(TFHE_SPEC) --
 
 .PHONY: bench_integer_multi_bit_gpu # Run benchmarks for integer on GPU backend using multi-bit parameters
 bench_integer_multi_bit_gpu: install_rs_check_toolchain
@@ -1085,22 +1137,22 @@ bench_integer_multi_bit_gpu: install_rs_check_toolchain
 	__TFHE_RS_BENCH_OP_FLAVOR=$(BENCH_OP_FLAVOR) __TFHE_RS_FAST_BENCH=$(FAST_BENCH) __TFHE_RS_BENCH_TYPE=$(BENCH_TYPE) \
 	cargo $(CARGO_RS_CHECK_TOOLCHAIN) bench \
 	--bench integer-bench \
-	--features=$(TARGET_ARCH_FEATURE),integer,gpu,internal-keycache,nightly-avx512 -p $(TFHE_SPEC) --
+	--features=integer,gpu,internal-keycache,nightly-avx512,pbs-stats -p $(TFHE_SPEC) --
 
-.PHONY: bench_unsigned_integer_multi_bit_gpu # Run benchmarks for unsigned integer on GPU backend using multi-bit parameters
-bench_unsigned_integer_multi_bit_gpu: install_rs_check_toolchain
+.PHONY: bench_signed_integer_multi_bit_gpu # Run benchmarks for signed integer on GPU backend using multi-bit parameters
+bench_signed_integer_multi_bit_gpu: install_rs_check_toolchain
 	RUSTFLAGS="$(RUSTFLAGS)" __TFHE_RS_PARAM_TYPE=MULTI_BIT \
 	__TFHE_RS_BENCH_OP_FLAVOR=$(BENCH_OP_FLAVOR) __TFHE_RS_FAST_BENCH=$(FAST_BENCH) __TFHE_RS_BENCH_TYPE=$(BENCH_TYPE) \
 	cargo $(CARGO_RS_CHECK_TOOLCHAIN) bench \
-	--bench integer-bench \
-	--features=$(TARGET_ARCH_FEATURE),integer,gpu,internal-keycache,nightly-avx512 -p $(TFHE_SPEC) -- ::unsigned
+	--bench integer-signed-bench \
+	--features=integer,gpu,internal-keycache,nightly-avx512,pbs-stats -p $(TFHE_SPEC) --
 
 .PHONY: bench_integer_zk # Run benchmarks for integer encryption with ZK proofs
 bench_integer_zk: install_rs_check_toolchain
 	RUSTFLAGS="$(RUSTFLAGS)" __TFHE_RS_BENCH_TYPE=$(BENCH_TYPE) \
 	cargo $(CARGO_RS_CHECK_TOOLCHAIN) bench \
 	--bench zk-pke-bench \
-	--features=$(TARGET_ARCH_FEATURE),integer,internal-keycache,zk-pok,nightly-avx512 \
+	--features=integer,internal-keycache,zk-pok,nightly-avx512,pbs-stats \
 	-p $(TFHE_SPEC) --
 
 .PHONY: bench_shortint # Run benchmarks for shortint
@@ -1108,14 +1160,14 @@ bench_shortint: install_rs_check_toolchain
 	RUSTFLAGS="$(RUSTFLAGS)" __TFHE_RS_BENCH_OP_FLAVOR=$(BENCH_OP_FLAVOR) \
 	cargo $(CARGO_RS_CHECK_TOOLCHAIN) bench \
 	--bench shortint-bench \
-	--features=$(TARGET_ARCH_FEATURE),shortint,internal-keycache,nightly-avx512 -p $(TFHE_SPEC)
+	--features=shortint,internal-keycache,nightly-avx512 -p $(TFHE_SPEC)
 
 .PHONY: bench_shortint_oprf # Run benchmarks for shortint
 bench_shortint_oprf: install_rs_check_toolchain
 	RUSTFLAGS="$(RUSTFLAGS)" \
 	cargo $(CARGO_RS_CHECK_TOOLCHAIN) bench \
 	--bench oprf-shortint-bench \
-	--features=$(TARGET_ARCH_FEATURE),shortint,internal-keycache,nightly-avx512 -p $(TFHE_SPEC)
+	--features=shortint,internal-keycache,nightly-avx512 -p $(TFHE_SPEC)
 
 .PHONY: bench_shortint_multi_bit # Run benchmarks for shortint using multi-bit parameters
 bench_shortint_multi_bit: install_rs_check_toolchain
@@ -1123,43 +1175,64 @@ bench_shortint_multi_bit: install_rs_check_toolchain
 	__TFHE_RS_BENCH_OP_FLAVOR=$(BENCH_OP_FLAVOR) \
 	cargo $(CARGO_RS_CHECK_TOOLCHAIN) bench \
 	--bench shortint-bench \
-	--features=$(TARGET_ARCH_FEATURE),shortint,internal-keycache,nightly-avx512 -p $(TFHE_SPEC) --
+	--features=shortint,internal-keycache,nightly-avx512 -p $(TFHE_SPEC) --
 
 .PHONY: bench_boolean # Run benchmarks for boolean
 bench_boolean: install_rs_check_toolchain
 	RUSTFLAGS="$(RUSTFLAGS)" cargo $(CARGO_RS_CHECK_TOOLCHAIN) bench \
 	--bench boolean-bench \
-	--features=$(TARGET_ARCH_FEATURE),boolean,internal-keycache,nightly-avx512 -p $(TFHE_SPEC)
+	--features=boolean,internal-keycache,nightly-avx512 -p $(TFHE_SPEC)
 
 .PHONY: bench_pbs # Run benchmarks for PBS
 bench_pbs: install_rs_check_toolchain
-	RUSTFLAGS="$(RUSTFLAGS)" cargo $(CARGO_RS_CHECK_TOOLCHAIN) bench \
+	RUSTFLAGS="$(RUSTFLAGS)" __TFHE_RS_PARAMS_SET=$(BENCH_PARAMS_SET) cargo $(CARGO_RS_CHECK_TOOLCHAIN) bench \
 	--bench pbs-bench \
-	--features=$(TARGET_ARCH_FEATURE),boolean,shortint,internal-keycache,nightly-avx512 -p $(TFHE_SPEC)
+	--features=boolean,shortint,internal-keycache,nightly-avx512 -p $(TFHE_SPEC)
+
+.PHONY: bench_ks_pbs # Run benchmarks for KS-PBS
+bench_ks_pbs: install_rs_check_toolchain
+	RUSTFLAGS="$(RUSTFLAGS)" __TFHE_RS_PARAM_TYPE=$(BENCH_PARAM_TYPE) __TFHE_RS_PARAMS_SET=$(BENCH_PARAMS_SET) \
+	cargo $(CARGO_RS_CHECK_TOOLCHAIN) bench \
+	--bench ks-pbs-bench \
+	--features=boolean,shortint,internal-keycache,nightly-avx512 -p $(TFHE_SPEC)
+
+.PHONY: bench_ks_pbs_gpu # Run benchmarks for KS-PBS on GPU backend
+bench_ks_pbs_gpu: install_rs_check_toolchain
+	RUSTFLAGS="$(RUSTFLAGS)" __TFHE_RS_PARAM_TYPE=$(BENCH_PARAM_TYPE) __TFHE_RS_PARAMS_SET=$(BENCH_PARAMS_SET) \
+	cargo $(CARGO_RS_CHECK_TOOLCHAIN) bench \
+	--bench ks-pbs-bench \
+	--features=boolean,shortint,gpu,internal-keycache,nightly-avx512 -p $(TFHE_SPEC)
 
 .PHONY: bench_pbs128 # Run benchmarks for PBS using FFT 128 bits
 bench_pbs128: install_rs_check_toolchain
 	RUSTFLAGS="$(RUSTFLAGS)" cargo $(CARGO_RS_CHECK_TOOLCHAIN) bench \
 	--bench pbs128-bench \
-	--features=$(TARGET_ARCH_FEATURE),boolean,shortint,internal-keycache,nightly-avx512 -p $(TFHE_SPEC)
+	--features=boolean,shortint,internal-keycache,nightly-avx512 -p $(TFHE_SPEC)
+
+.PHONY: bench_pbs128_gpu # Run benchmarks for PBS using FFT 128 bits on GPU
+bench_pbs128_gpu: install_rs_check_toolchain
+	RUSTFLAGS="$(RUSTFLAGS)" cargo $(CARGO_RS_CHECK_TOOLCHAIN) bench \
+	--bench pbs128-bench \
+	--features=boolean,shortint,gpu,internal-keycache,nightly-avx512 -p $(TFHE_SPEC)
 
 .PHONY: bench_pbs_gpu # Run benchmarks for PBS on GPU backend
 bench_pbs_gpu: install_rs_check_toolchain
-	RUSTFLAGS="$(RUSTFLAGS)" __TFHE_RS_FAST_BENCH=$(FAST_BENCH) cargo $(CARGO_RS_CHECK_TOOLCHAIN) bench \
+	RUSTFLAGS="$(RUSTFLAGS)" __TFHE_RS_FAST_BENCH=$(FAST_BENCH) __TFHE_RS_PARAMS_SET=$(BENCH_PARAMS_SET) \
+	cargo $(CARGO_RS_CHECK_TOOLCHAIN) bench \
 	--bench pbs-bench \
-	--features=$(TARGET_ARCH_FEATURE),boolean,shortint,gpu,internal-keycache,nightly-avx512 -p $(TFHE_SPEC)
+	--features=boolean,shortint,gpu,internal-keycache,nightly-avx512 -p $(TFHE_SPEC)
 
 .PHONY: bench_ks # Run benchmarks for keyswitch
 bench_ks: install_rs_check_toolchain
-	RUSTFLAGS="$(RUSTFLAGS)" cargo $(CARGO_RS_CHECK_TOOLCHAIN) bench \
+	RUSTFLAGS="$(RUSTFLAGS)" __TFHE_RS_PARAMS_SET=$(BENCH_PARAMS_SET) cargo $(CARGO_RS_CHECK_TOOLCHAIN) bench \
 	--bench ks-bench \
-	--features=$(TARGET_ARCH_FEATURE),boolean,shortint,internal-keycache,nightly-avx512 -p $(TFHE_SPEC)
+	--features=boolean,shortint,internal-keycache,nightly-avx512 -p $(TFHE_SPEC)
 
 .PHONY: bench_ks_gpu # Run benchmarks for PBS on GPU backend
 bench_ks_gpu: install_rs_check_toolchain
-	RUSTFLAGS="$(RUSTFLAGS)" cargo $(CARGO_RS_CHECK_TOOLCHAIN) bench \
+	RUSTFLAGS="$(RUSTFLAGS)" __TFHE_RS_PARAMS_SET=$(BENCH_PARAMS_SET) cargo $(CARGO_RS_CHECK_TOOLCHAIN) bench \
 	--bench ks-bench \
-	--features=$(TARGET_ARCH_FEATURE),boolean,shortint,gpu,internal-keycache,nightly-avx512 -p $(TFHE_SPEC)
+	--features=boolean,shortint,gpu,internal-keycache,nightly-avx512 -p $(TFHE_SPEC)
 
 bench_web_js_api_parallel_chrome: browser_path = "$(WEB_RUNNER_DIR)/chrome/chrome-linux64/chrome"
 bench_web_js_api_parallel_chrome: driver_path = "$(WEB_RUNNER_DIR)/chrome/chromedriver-linux64/chromedriver"
@@ -1195,13 +1268,13 @@ bench_web_js_api_parallel_firefox_ci: setup_venv
 bench_hlapi_erc20: install_rs_check_toolchain
 	RUSTFLAGS="$(RUSTFLAGS)" cargo $(CARGO_RS_CHECK_TOOLCHAIN) bench \
 	--bench hlapi-erc20 \
-	--features=$(TARGET_ARCH_FEATURE),integer,internal-keycache,pbs-stats,nightly-avx512 -p $(TFHE_SPEC) --
+	--features=integer,internal-keycache,pbs-stats,nightly-avx512 -p $(TFHE_SPEC) --
 
 .PHONY: bench_hlapi_erc20_gpu # Run benchmarks for ECR20 operations on GPU
 bench_hlapi_erc20_gpu: install_rs_check_toolchain
 	RUSTFLAGS="$(RUSTFLAGS)" cargo $(CARGO_RS_CHECK_TOOLCHAIN) bench \
 	--bench hlapi-erc20 \
-	--features=$(TARGET_ARCH_FEATURE),integer,gpu,internal-keycache,pbs-stats,nightly-avx512 -p $(TFHE_SPEC) --
+	--features=integer,gpu,internal-keycache,pbs-stats,nightly-avx512 -p $(TFHE_SPEC) --
 
 .PHONY: bench_tfhe_zk_pok # Run benchmarks for the tfhe_zk_pok crate
 bench_tfhe_zk_pok: install_rs_check_toolchain
@@ -1216,32 +1289,32 @@ bench_tfhe_zk_pok: install_rs_check_toolchain
 gen_key_cache: install_rs_build_toolchain
 	RUSTFLAGS="$(RUSTFLAGS) --cfg tarpaulin" cargo $(CARGO_RS_BUILD_TOOLCHAIN) run --profile $(CARGO_PROFILE) \
 		--example generates_test_keys \
-		--features=$(TARGET_ARCH_FEATURE),boolean,shortint,experimental,internal-keycache -p $(TFHE_SPEC) \
+		--features=boolean,shortint,experimental,internal-keycache -p $(TFHE_SPEC) \
 		-- $(MULTI_BIT_ONLY) $(COVERAGE_ONLY)
 
 .PHONY: gen_key_cache_core_crypto # Run function to generate keys and cache them for core_crypto tests
 gen_key_cache_core_crypto: install_rs_build_toolchain
 	RUSTFLAGS="$(RUSTFLAGS)" cargo $(CARGO_RS_BUILD_TOOLCHAIN) test --tests --profile $(CARGO_PROFILE) \
-		--features=$(TARGET_ARCH_FEATURE),experimental,internal-keycache -p $(TFHE_SPEC) -- --nocapture \
+		--features=experimental,internal-keycache -p $(TFHE_SPEC) -- --nocapture \
 		core_crypto::keycache::generate_keys
 
 .PHONY: measure_hlapi_compact_pk_ct_sizes # Measure sizes of public keys and ciphertext for high-level API
 measure_hlapi_compact_pk_ct_sizes: install_rs_check_toolchain
 	RUSTFLAGS="$(RUSTFLAGS)" cargo $(CARGO_RS_CHECK_TOOLCHAIN) run --profile $(CARGO_PROFILE) \
 	--example hlapi_compact_pk_ct_sizes \
-	--features=$(TARGET_ARCH_FEATURE),integer,internal-keycache
+	--features=integer,internal-keycache
 
 .PHONY: measure_shortint_key_sizes # Measure sizes of bootstrapping and key switching keys for shortint
 measure_shortint_key_sizes: install_rs_check_toolchain
 	RUSTFLAGS="$(RUSTFLAGS)" cargo $(CARGO_RS_CHECK_TOOLCHAIN) run --profile $(CARGO_PROFILE) \
 	--example shortint_key_sizes \
-	--features=$(TARGET_ARCH_FEATURE),shortint,internal-keycache
+	--features=shortint,internal-keycache
 
 .PHONY: measure_boolean_key_sizes # Measure sizes of bootstrapping and key switching keys for boolean
 measure_boolean_key_sizes: install_rs_check_toolchain
 	RUSTFLAGS="$(RUSTFLAGS)" cargo $(CARGO_RS_CHECK_TOOLCHAIN) run --profile $(CARGO_PROFILE) \
 	--example boolean_key_sizes \
-	--features=$(TARGET_ARCH_FEATURE),boolean,internal-keycache
+	--features=boolean,internal-keycache
 
 .PHONY: parse_integer_benches # Run python parser to output a csv containing integer benches data
 parse_integer_benches:
@@ -1253,20 +1326,19 @@ parse_integer_benches:
 parse_wasm_benchmarks: install_rs_check_toolchain
 	RUSTFLAGS="$(RUSTFLAGS)" cargo $(CARGO_RS_CHECK_TOOLCHAIN) run --profile $(CARGO_PROFILE) \
 	--example wasm_benchmarks_parser \
-	--features=$(TARGET_ARCH_FEATURE),shortint,internal-keycache \
+	--features=shortint,internal-keycache \
 	-- wasm_benchmark_results.json
 
 .PHONY: write_params_to_file # Gather all crypto parameters into a file with a Sage readable format.
 write_params_to_file: install_rs_check_toolchain
-	RUSTFLAGS="$(RUSTFLAGS)" cargo $(CARGO_RS_CHECK_TOOLCHAIN) run --profile $(CARGO_PROFILE) \
-	--example write_params_to_file \
-	--features=$(TARGET_ARCH_FEATURE),boolean,shortint,internal-keycache
+	RUSTFLAGS="$(RUSTFLAGS)" cargo $(CARGO_RS_CHECK_TOOLCHAIN) run \
+	--example write_params_to_file --features=boolean,shortint,internal-keycache
 
 .PHONY: clone_backward_compat_data # Clone the data repo needed for backward compatibility tests
 clone_backward_compat_data:
-	./scripts/clone_backward_compat_data.sh $(BACKWARD_COMPAT_DATA_URL) $(BACKWARD_COMPAT_DATA_BRANCH) tfhe/$(BACKWARD_COMPAT_DATA_DIR)
+	./scripts/clone_backward_compat_data.sh $(BACKWARD_COMPAT_DATA_URL) $(BACKWARD_COMPAT_DATA_BRANCH) tests/$(BACKWARD_COMPAT_DATA_DIR)
 
-tfhe/$(BACKWARD_COMPAT_DATA_DIR): clone_backward_compat_data
+tests/$(BACKWARD_COMPAT_DATA_DIR): clone_backward_compat_data
 
 #
 # Real use case examples
@@ -1275,36 +1347,36 @@ tfhe/$(BACKWARD_COMPAT_DATA_DIR): clone_backward_compat_data
 .PHONY: regex_engine # Run regex_engine example
 regex_engine: install_rs_check_toolchain
 	RUSTFLAGS="$(RUSTFLAGS)" cargo $(CARGO_RS_CHECK_TOOLCHAIN) run --profile $(CARGO_PROFILE) \
-	--example regex_engine \
-	--features=$(TARGET_ARCH_FEATURE),integer \
+	--example regex_engine --features=integer \
 	-- $(REGEX_STRING) $(REGEX_PATTERN)
 
 .PHONY: dark_market # Run dark market example
 dark_market: install_rs_check_toolchain
 	RUSTFLAGS="$(RUSTFLAGS)" cargo $(CARGO_RS_CHECK_TOOLCHAIN) run --profile $(CARGO_PROFILE) \
 	--example dark_market \
-	--features=$(TARGET_ARCH_FEATURE),integer,internal-keycache \
+	--features=integer,internal-keycache \
 	-- fhe-modified fhe-parallel plain fhe
 
 .PHONY: sha256_bool # Run sha256_bool example
 sha256_bool: install_rs_check_toolchain
 	RUSTFLAGS="$(RUSTFLAGS)" cargo $(CARGO_RS_CHECK_TOOLCHAIN) run --profile $(CARGO_PROFILE) \
-	--example sha256_bool \
-	--features=$(TARGET_ARCH_FEATURE),boolean
+	--example sha256_bool --features=boolean
 
 .PHONY: pcc # pcc stands for pre commit checks (except GPU)
-pcc: no_tfhe_typo no_dbg_log check_fmt check_typos lint_doc check_md_docs_are_tested check_intra_md_links \
-clippy_all tfhe_lints check_compile_tests
+pcc: no_tfhe_typo no_dbg_log check_parameter_export_ok check_fmt check_typos lint_doc \
+check_md_docs_are_tested check_intra_md_links clippy_all check_compile_tests test_tfhe_lints \
+tfhe_lints
 
 .PHONY: pcc_gpu # pcc stands for pre commit checks for GPU compilation
-pcc_gpu: clippy_gpu clippy_cuda_backend check_compile_tests_benches_gpu check_rust_bindings_did_not_change
+pcc_gpu: check_rust_bindings_did_not_change clippy_rustdoc_gpu \
+clippy_gpu clippy_cuda_backend check_compile_tests_benches_gpu
 
 .PHONY: fpcc # pcc stands for pre commit checks, the f stands for fast
-fpcc: no_tfhe_typo no_dbg_log check_fmt check_typos lint_doc check_md_docs_are_tested clippy_fast \
-check_compile_tests
+fpcc: no_tfhe_typo no_dbg_log check_parameter_export_ok check_fmt check_typos lint_doc \
+check_md_docs_are_tested clippy_fast check_compile_tests
 
 .PHONY: conformance # Automatically fix problems that can be fixed
-conformance: fix_newline fmt
+conformance: fix_newline fmt fmt_js
 
 #=============================== FFT Section ==================================
 .PHONY: doc_fft # Build rust doc for tfhe-fft
@@ -1376,7 +1448,7 @@ test_fft_nightly: install_rs_check_toolchain
 .PHONY: test_fft_no_std
 test_fft_no_std: install_rs_build_toolchain
 	RUSTFLAGS="$(RUSTFLAGS)" cargo $(CARGO_RS_BUILD_TOOLCHAIN) test --release -p tfhe-fft \
-		--no-default-features 
+		--no-default-features
 	RUSTFLAGS="$(RUSTFLAGS)" cargo $(CARGO_RS_BUILD_TOOLCHAIN) test --release -p tfhe-fft \
 		--no-default-features \
 		--features=fft128
@@ -1470,7 +1542,7 @@ test_ntt_nightly: install_rs_check_toolchain
 .PHONY: test_ntt_no_std
 test_ntt_no_std: install_rs_build_toolchain
 	RUSTFLAGS="$(RUSTFLAGS)" cargo $(CARGO_RS_BUILD_TOOLCHAIN) test --release -p tfhe-ntt \
-		--no-default-features 
+		--no-default-features
 
 .PHONY: test_ntt_no_std_nightly
 test_ntt_no_std_nightly: install_rs_check_toolchain

README.md

@@ -10,7 +10,7 @@
 <hr/>
 
 <p align="center">
-  <a href="https://docs.zama.ai/tfhe-rs"> 📒 Documentation</a> | <a href="https://zama.ai/community"> 💛 Community support</a> | <a href="https://github.com/zama-ai/awesome-zama"> 📚 FHE resources by Zama</a>
+  <a href="https://github.com/zama-ai/tfhe-rs-handbook/blob/main/tfhe-rs-handbook.pdf"> 📃 Read Handbook</a> |<a href="https://docs.zama.ai/tfhe-rs"> 📒 Documentation</a> | <a href="https://zama.ai/community"> 💛 Community support</a> | <a href="https://github.com/zama-ai/awesome-zama"> 📚 FHE resources by Zama</a>
 </p>
 
 
@@ -67,35 +67,24 @@ production-ready library for all the advanced features of TFHE.
 
 ## Getting started
 
+> [!Important]
+> **TFHE-rs** released its first stable version v1.0.0 in February 2025, stabilizing the high-level API for the x86 CPU backend.
+
 ### Cargo.toml configuration
 To use the latest version of `TFHE-rs` in your project, you first need to add it as a dependency in your `Cargo.toml`:
 
-+ For x86_64-based machines running Unix-like OSes:
-
 ```toml
-tfhe = { version = "*", features = ["boolean", "shortint", "integer", "x86_64-unix"] }
-```
-
-+ For Apple Silicon or aarch64-based machines running Unix-like OSes:
-
-```toml
-tfhe = { version = "*", features = ["boolean", "shortint", "integer", "aarch64-unix"] }
-```
-
-+ For x86_64-based machines with the [`rdseed instruction`](https://en.wikipedia.org/wiki/RDRAND) running Windows:
-
-```toml
-tfhe = { version = "*", features = ["boolean", "shortint", "integer", "x86_64"] }
+tfhe = { version = "*", features = ["boolean", "shortint", "integer"] }
 ```
 
 > [!Note]
-> Note: You need to use a Rust version >= 1.81 to compile TFHE-rs.
+> Note: You need to use Rust version >= 1.84 to compile TFHE-rs.
 
 > [!Note]
-> Note: aarch64-based machines are not yet supported for Windows as it's currently missing an entropy source to be able to seed the [CSPRNGs](https://en.wikipedia.org/wiki/Cryptographically_secure_pseudorandom_number_generator) used in TFHE-rs.
+> Note: AArch64-based machines are not supported for Windows as it's currently missing an entropy source to be able to seed the [CSPRNGs](https://en.wikipedia.org/wiki/Cryptographically_secure_pseudorandom_number_generator) used in TFHE-rs.
 
 <p align="right">
-  <a href="#about" > ↑ Back to top </a> 
+  <a href="#about" > ↑ Back to top </a>
 </p>
 
 ### A simple example
@@ -152,7 +141,7 @@ fn main() -> Result<(), Box<dyn std::error::Error>> {
 }
 ```
 
-To run this code, use the following command: 
+To run this code, use the following command:
 <p align="center"> <code> cargo run --release </code> </p>
 
 > [!Note]
@@ -162,12 +151,15 @@ to run in release mode with cargo's `--release` flag to have the best performanc
 *Find an example with more explanations in [this part of the documentation](https://docs.zama.ai/tfhe-rs/get-started/quick_start)*
 
 <p align="right">
-  <a href="#about" > ↑ Back to top </a> 
+  <a href="#about" > ↑ Back to top </a>
 </p>
 
 
 
-## Resources 
+## Resources
+
+### TFHE-rs Handbook
+A document containing scientific and technical details about algorithms implemented into the library is available here: [TFHE-rs: A (Practical) Handbook](https://github.com/zama-ai/tfhe-rs-handbook/blob/main/tfhe-rs-handbook.pdf).
 
 ### TFHE deep dive
 - [TFHE Deep Dive - Part I - Ciphertext types](https://www.zama.ai/post/tfhe-deep-dive-part-1)
@@ -190,7 +182,7 @@ to run in release mode with cargo's `--release` flag to have the best performanc
 
 Full, comprehensive documentation is available here: [https://docs.zama.ai/tfhe-rs](https://docs.zama.ai/tfhe-rs).
 <p align="right">
-  <a href="#about" > ↑ Back to top </a> 
+  <a href="#about" > ↑ Back to top </a>
 </p>
 
 
@@ -208,9 +200,13 @@ When a new update is published in the Lattice Estimator, we update parameters ac
 
 ### Security model
 
-The default parameters for the TFHE-rs library are chosen considering the IND-CPA security model, and are selected with a bootstrapping failure probability fixed at p_error = $2^{-64}$. In particular, it is assumed that the results of decrypted computations are not shared by the secret key owner with any third parties, as such an action can lead to leakage of the secret encryption key. If you are designing an application where decryptions must be shared, you will need to craft custom encryption parameters which are chosen in consideration of the IND-CPA^D security model [1]. 
+By default, the parameter sets used in the High-Level API with the x86 CPU backend have a failure probability $\le 2^{128}$ to securely work in the IND-CPA^D model using the algorithmic techniques provided in our code base [1].
+If you want to work within the IND-CPA security model, which is less strict than the IND-CPA-D model, the parameter sets can easily be changed and would have slightly better performance. More details can be found in the [TFHE-rs documentation](https://docs.zama.ai/tfhe-rs).
 
-[1] Li, Baiyu, et al. "Securing approximate homomorphic encryption using differential privacy." Annual International Cryptology Conference. Cham: Springer Nature Switzerland, 2022. https://eprint.iacr.org/2022/816.pdf
+The default parameters used in the High-Level API with the GPU backend are chosen considering the IND-CPA security model, and are selected with a bootstrapping failure probability fixed at $p_{error} \le 2^{-64}$. In particular, it is assumed that the results of decrypted computations are not shared by the secret key owner with any third parties, as such an action can lead to leakage of the secret encryption key. If you are designing an application where decryptions must be shared, you will need to craft custom encryption parameters which are chosen in consideration of the IND-CPA^D security model [2].
+
+[1] Bernard, Olivier, et al. "Drifting Towards Better Error Probabilities in Fully Homomorphic Encryption Schemes". https://eprint.iacr.org/2024/1718.pdf
+[2] Li, Baiyu, et al. "Securing approximate homomorphic encryption using differential privacy." Annual International Cryptology Conference. Cham: Springer Nature Switzerland, 2022. https://eprint.iacr.org/2022/816.pdf
 
 #### Side-channel attacks
 
@@ -259,7 +255,7 @@ This software is distributed under the **BSD-3-Clause-Clear** license. Read [thi
 >We are open to collaborating and advancing the FHE space with our partners. If you have specific needs, please email us at hello@zama.ai.
 
 <p align="right">
-  <a href="#about" > ↑ Back to top </a> 
+  <a href="#about" > ↑ Back to top </a>
 </p>
 
 
@@ -273,8 +269,8 @@ This software is distributed under the **BSD-3-Clause-Clear** license. Read [thi
 </picture>
 </a>
 
-🌟 If you find this project helpful or interesting, please consider giving it a star on GitHub! Your support helps to grow the community and motivates further development. 
+🌟 If you find this project helpful or interesting, please consider giving it a star on GitHub! Your support helps to grow the community and motivates further development.
 
 <p align="right">
-  <a href="#about" > ↑ Back to top </a> 
+  <a href="#about" > ↑ Back to top </a>
 </p>

_typos.toml

@@ -13,3 +13,9 @@ extend-ignore-identifiers-re = [
     # Example in trivium
     "C9217BA0D762ACA1"
 ]
+
+[files]
+extend-exclude = [
+    "backends/tfhe-cuda-backend/cuda/src/fft128/twiddles.cu",
+    "backends/tfhe-cuda-backend/cuda/src/fft/twiddles.cu",
+]

apps/trivium/Cargo.toml

@@ -6,18 +6,17 @@ edition = "2021"
 # See more keys and their definitions at https://doc.rust-lang.org/cargo/reference/manifest.html
 
 [dependencies]
-rayon = { version = "1.7.0"}
-
-[target.'cfg(target_arch = "x86_64")'.dependencies.tfhe]
-path = "../../tfhe"
-features = [ "boolean", "shortint", "integer", "x86_64" ]
-
-[target.'cfg(target_arch = "aarch64")'.dependencies.tfhe]
-path = "../../tfhe"
-features = [ "boolean", "shortint", "integer", "aarch64-unix" ]
+rayon = "1"
+tfhe = { path = "../../tfhe", features = ["boolean", "shortint", "integer"] }
 
 [dev-dependencies]
-criterion = { version = "0.5.1", features = [ "html_reports" ]}
+criterion = { version = "0.5.1", features = ["html_reports"] }
+
+[profile.devo]
+inherits = "dev"
+opt-level = 3
+lto = "off"
+debug-assertions = false
 
 [[bench]]
 name = "trivium"

apps/trivium/README.md

@@ -18,102 +18,102 @@ use tfhe::prelude::*;
 use tfhe_trivium::TriviumStream;
 
 fn get_hexadecimal_string_from_lsb_first_stream(a: Vec<bool>) -> String {
-	assert!(a.len() % 8 == 0);
-	let mut hexadecimal: String = "".to_string();
-	for test in a.chunks(8) {
-		// Encoding is bytes in LSB order
-		match test[4..8] {
-			[false, false, false, false] => hexadecimal.push('0'),
-			[true, false, false, false] => hexadecimal.push('1'),
-			[false, true, false, false] => hexadecimal.push('2'),
-			[true, true, false, false] => hexadecimal.push('3'),
+    assert!(a.len() % 8 == 0);
+    let mut hexadecimal: String = "".to_string();
+    for test in a.chunks(8) {
+        // Encoding is bytes in LSB order
+        match test[4..8] {
+            [false, false, false, false] => hexadecimal.push('0'),
+            [true, false, false, false] => hexadecimal.push('1'),
+            [false, true, false, false] => hexadecimal.push('2'),
+            [true, true, false, false] => hexadecimal.push('3'),
 
-			[false, false, true, false] => hexadecimal.push('4'),
-			[true, false, true, false] => hexadecimal.push('5'),
-			[false, true, true, false] => hexadecimal.push('6'),
-			[true, true, true, false] => hexadecimal.push('7'),
+            [false, false, true, false] => hexadecimal.push('4'),
+            [true, false, true, false] => hexadecimal.push('5'),
+            [false, true, true, false] => hexadecimal.push('6'),
+            [true, true, true, false] => hexadecimal.push('7'),
 
-			[false, false, false, true] => hexadecimal.push('8'),
-			[true, false, false, true] => hexadecimal.push('9'),
-			[false, true, false, true] => hexadecimal.push('A'),
-			[true, true, false, true] => hexadecimal.push('B'),
+            [false, false, false, true] => hexadecimal.push('8'),
+            [true, false, false, true] => hexadecimal.push('9'),
+            [false, true, false, true] => hexadecimal.push('A'),
+            [true, true, false, true] => hexadecimal.push('B'),
 
-			[false, false, true, true] => hexadecimal.push('C'),
-			[true, false, true, true] => hexadecimal.push('D'),
-			[false, true, true, true] => hexadecimal.push('E'),
-			[true, true, true, true] => hexadecimal.push('F'),
-			_ => ()
-		};
-		match test[0..4] {
-			[false, false, false, false] => hexadecimal.push('0'),
-			[true, false, false, false] => hexadecimal.push('1'),
-			[false, true, false, false] => hexadecimal.push('2'),
-			[true, true, false, false] => hexadecimal.push('3'),
+            [false, false, true, true] => hexadecimal.push('C'),
+            [true, false, true, true] => hexadecimal.push('D'),
+            [false, true, true, true] => hexadecimal.push('E'),
+            [true, true, true, true] => hexadecimal.push('F'),
+            _ => ()
+        };
+        match test[0..4] {
+            [false, false, false, false] => hexadecimal.push('0'),
+            [true, false, false, false] => hexadecimal.push('1'),
+            [false, true, false, false] => hexadecimal.push('2'),
+            [true, true, false, false] => hexadecimal.push('3'),
 
-			[false, false, true, false] => hexadecimal.push('4'),
-			[true, false, true, false] => hexadecimal.push('5'),
-			[false, true, true, false] => hexadecimal.push('6'),
-			[true, true, true, false] => hexadecimal.push('7'),
+            [false, false, true, false] => hexadecimal.push('4'),
+            [true, false, true, false] => hexadecimal.push('5'),
+            [false, true, true, false] => hexadecimal.push('6'),
+            [true, true, true, false] => hexadecimal.push('7'),
 
-			[false, false, false, true] => hexadecimal.push('8'),
-			[true, false, false, true] => hexadecimal.push('9'),
-			[false, true, false, true] => hexadecimal.push('A'),
-			[true, true, false, true] => hexadecimal.push('B'),
+            [false, false, false, true] => hexadecimal.push('8'),
+            [true, false, false, true] => hexadecimal.push('9'),
+            [false, true, false, true] => hexadecimal.push('A'),
+            [true, true, false, true] => hexadecimal.push('B'),
 
-			[false, false, true, true] => hexadecimal.push('C'),
-			[true, false, true, true] => hexadecimal.push('D'),
-			[false, true, true, true] => hexadecimal.push('E'),
-			[true, true, true, true] => hexadecimal.push('F'),
-			_ => ()
-		};
-	}
-	return hexadecimal;
+            [false, false, true, true] => hexadecimal.push('C'),
+            [true, false, true, true] => hexadecimal.push('D'),
+            [false, true, true, true] => hexadecimal.push('E'),
+            [true, true, true, true] => hexadecimal.push('F'),
+            _ => ()
+        };
+    }
+    return hexadecimal;
 }
 
 fn main() {
-	let config = ConfigBuilder::default().build();
-	let (client_key, server_key) = generate_keys(config);
+    let config = ConfigBuilder::default().build();
+    let (client_key, server_key) = generate_keys(config);
 
-	let key_string = "0053A6F94C9FF24598EB".to_string();
-	let mut key = [false; 80];
+    let key_string = "0053A6F94C9FF24598EB".to_string();
+    let mut key = [false; 80];
 
-	for i in (0..key_string.len()).step_by(2) {
-		let mut val: u8 = u8::from_str_radix(&key_string[i..i+2], 16).unwrap();
-		for j in 0..8 {
-			key[8*(i>>1) + j] = val % 2 == 1;
-			val >>= 1;
-		}
-	}
+    for i in (0..key_string.len()).step_by(2) {
+        let mut val: u8 = u8::from_str_radix(&key_string[i..i+2], 16).unwrap();
+        for j in 0..8 {
+            key[8*(i>>1) + j] = val % 2 == 1;
+            val >>= 1;
+        }
+    }
 
-	let iv_string = "0D74DB42A91077DE45AC".to_string();
-	let mut iv = [false; 80];
+    let iv_string = "0D74DB42A91077DE45AC".to_string();
+    let mut iv = [false; 80];
 
-	for i in (0..iv_string.len()).step_by(2) {
-		let mut val: u8 = u8::from_str_radix(&iv_string[i..i+2], 16).unwrap();
-		for j in 0..8 {
-			iv[8*(i>>1) + j] = val % 2 == 1;
-			val >>= 1;
-		}
-	}
-	
-	let output_0_63    = "F4CD954A717F26A7D6930830C4E7CF0819F80E03F25F342C64ADC66ABA7F8A8E6EAA49F23632AE3CD41A7BD290A0132F81C6D4043B6E397D7388F3A03B5FE358".to_string();
+    for i in (0..iv_string.len()).step_by(2) {
+        let mut val: u8 = u8::from_str_radix(&iv_string[i..i+2], 16).unwrap();
+        for j in 0..8 {
+            iv[8*(i>>1) + j] = val % 2 == 1;
+            val >>= 1;
+        }
+    }
 
-	let cipher_key = key.map(|x| FheBool::encrypt(x, &client_key));
-	let cipher_iv = iv.map(|x| FheBool::encrypt(x, &client_key));
+    let output_0_63    = "F4CD954A717F26A7D6930830C4E7CF0819F80E03F25F342C64ADC66ABA7F8A8E6EAA49F23632AE3CD41A7BD290A0132F81C6D4043B6E397D7388F3A03B5FE358".to_string();
+
+    let cipher_key = key.map(|x| FheBool::encrypt(x, &client_key));
+    let cipher_iv = iv.map(|x| FheBool::encrypt(x, &client_key));
 
 
-	let mut trivium = TriviumStream::<FheBool>::new(cipher_key, cipher_iv, &server_key);
+    let mut trivium = TriviumStream::<FheBool>::new(cipher_key, cipher_iv, &server_key);
 
-	let mut vec = Vec::<bool>::with_capacity(64*8);
-	while vec.len() < 64*8 {
-		let cipher_outputs = trivium.next_64();
-		for c in cipher_outputs {
-			vec.push(c.decrypt(&client_key))
-		}
-	}
+    let mut vec = Vec::<bool>::with_capacity(64*8);
+    while vec.len() < 64*8 {
+        let cipher_outputs = trivium.next_64();
+        for c in cipher_outputs {
+            vec.push(c.decrypt(&client_key))
+        }
+    }
 
-	let hexadecimal = get_hexadecimal_string_from_lsb_first_stream(vec);
-	assert_eq!(output_0_63, hexadecimal[0..64*2]);
+    let hexadecimal = get_hexadecimal_string_from_lsb_first_stream(vec);
+    assert_eq!(output_0_63, hexadecimal[0..64*2]);
 }
 ```
 
@@ -129,63 +129,76 @@ Other sizes than 64 bit are expected to be available in the future.
 
 # FHE shortint Trivium implementation
 
-The same implementation is also available for generic Ciphertexts representing bits (meant to be used with parameters `PARAM_MESSAGE_1_CARRY_1_KS_PBS`). It uses a lower level API 
-of tfhe-rs, so the syntax is a little bit different. It also implements the `TransCiphering` trait. For optimization purposes, it does not internally run on the same 
-cryptographic parameters as the high level API of tfhe-rs. As such, it requires the usage of a casting key, to switch from one parameter space to another, which makes 
+The same implementation is also available for generic Ciphertexts representing bits (meant to be used with parameters `V1_1_PARAM_MESSAGE_1_CARRY_1_KS_PBS_GAUSSIAN_2M128`).
+It uses a lower level API of tfhe-rs, so the syntax is a little bit different. It also implements the `TransCiphering` trait. For optimization purposes, it does not internally run
+on the same cryptographic parameters as the high level API of tfhe-rs. As such, it requires the usage of a casting key, to switch from one parameter space to another, which makes
 its setup a little more intricate.
 
 Example code:
 ```rust
 use tfhe::shortint::prelude::*;
-use tfhe::shortint::CastingKey;
+use tfhe::shortint::parameters::v1_1::{
+    V1_1_PARAM_MESSAGE_1_CARRY_1_KS_PBS_GAUSSIAN_2M128,
+    V1_1_PARAM_MESSAGE_2_CARRY_2_KS_PBS_GAUSSIAN_2M128,
+    V1_1_PARAM_KEYSWITCH_1_1_KS_PBS_TO_2_2_KS_PBS_GAUSSIAN_2M128,
+};
 use tfhe::{ConfigBuilder, generate_keys, FheUint64};
 use tfhe::prelude::*;
 use tfhe_trivium::TriviumStreamShortint;
 
 fn test_shortint() {
-	let config = ConfigBuilder::default().build();
-	let (hl_client_key, hl_server_key) = generate_keys(config);
-	let (client_key, server_key): (ClientKey, ServerKey) = gen_keys(PARAM_MESSAGE_1_CARRY_1_KS_PBS);
-	let ksk = CastingKey::new((&client_key, &server_key), (&hl_client_key, &hl_server_key));
+    let config = ConfigBuilder::default()
+        .use_custom_parameters(V1_1_PARAM_MESSAGE_2_CARRY_2_KS_PBS_GAUSSIAN_2M128)
+        .build();
+    let (hl_client_key, hl_server_key) = generate_keys(config);
+    let underlying_ck: tfhe::shortint::ClientKey = (*hl_client_key.as_ref()).clone().into();
+    let underlying_sk: tfhe::shortint::ServerKey = (*hl_server_key.as_ref()).clone().into();
 
-	let key_string = "0053A6F94C9FF24598EB".to_string();
-	let mut key = [0; 80];
+    let (client_key, server_key): (ClientKey, ServerKey) = gen_keys(V1_1_PARAM_MESSAGE_1_CARRY_1_KS_PBS_GAUSSIAN_2M128);
+    let ksk = KeySwitchingKey::new(
+        (&client_key, Some(&server_key)),
+        (&underlying_ck, &underlying_sk),
+        V1_1_PARAM_KEYSWITCH_1_1_KS_PBS_TO_2_2_KS_PBS_GAUSSIAN_2M128_2M128,
+    );
 
-	for i in (0..key_string.len()).step_by(2) {
-		let mut val = u64::from_str_radix(&key_string[i..i+2], 16).unwrap();
-		for j in 0..8 {
-			key[8*(i>>1) + j] = val % 2;
-			val >>= 1;
-		}
-	}
+    let key_string = "0053A6F94C9FF24598EB".to_string();
+    let mut key = [0; 80];
 
-	let iv_string = "0D74DB42A91077DE45AC".to_string();
-	let mut iv = [0; 80];
+    for i in (0..key_string.len()).step_by(2) {
+        let mut val = u64::from_str_radix(&key_string[i..i+2], 16).unwrap();
+        for j in 0..8 {
+            key[8*(i>>1) + j] = val % 2;
+            val >>= 1;
+        }
+    }
 
-	for i in (0..iv_string.len()).step_by(2) {
-		let mut val = u64::from_str_radix(&iv_string[i..i+2], 16).unwrap();
-		for j in 0..8 {
-			iv[8*(i>>1) + j] = val % 2;
-			val >>= 1;
-		}
-	}
-	let output_0_63    = "F4CD954A717F26A7D6930830C4E7CF0819F80E03F25F342C64ADC66ABA7F8A8E6EAA49F23632AE3CD41A7BD290A0132F81C6D4043B6E397D7388F3A03B5FE358".to_string();
+    let iv_string = "0D74DB42A91077DE45AC".to_string();
+    let mut iv = [0; 80];
 
-	let cipher_key = key.map(|x| client_key.encrypt(x));
-	let cipher_iv = iv.map(|x| client_key.encrypt(x));
+    for i in (0..iv_string.len()).step_by(2) {
+        let mut val = u64::from_str_radix(&iv_string[i..i+2], 16).unwrap();
+        for j in 0..8 {
+            iv[8*(i>>1) + j] = val % 2;
+            val >>= 1;
+        }
+    }
+    let output_0_63    = "F4CD954A717F26A7D6930830C4E7CF0819F80E03F25F342C64ADC66ABA7F8A8E6EAA49F23632AE3CD41A7BD290A0132F81C6D4043B6E397D7388F3A03B5FE358".to_string();
 
-	let mut ciphered_message = vec![FheUint64::try_encrypt(0u64, &hl_client_key).unwrap(); 9];
+    let cipher_key = key.map(|x| client_key.encrypt(x));
+    let cipher_iv = iv.map(|x| client_key.encrypt(x));
 
-	let mut trivium = TriviumStreamShortint::new(cipher_key, cipher_iv, &server_key, &ksk);
+    let mut ciphered_message = vec![FheUint64::try_encrypt(0u64, &hl_client_key).unwrap(); 9];
 
-	let mut vec = Vec::<u64>::with_capacity(8);
-	while vec.len() < 8 {
-		let trans_ciphered_message = trivium.trans_encrypt_64(ciphered_message.pop().unwrap(), &hl_server_key);
-		vec.push(trans_ciphered_message.decrypt(&hl_client_key));
-	}
+    let mut trivium = TriviumStreamShortint::new(cipher_key, cipher_iv, &server_key, &ksk);
 
-	let hexadecimal = get_hexagonal_string_from_u64(vec);
-	assert_eq!(output_0_63, hexadecimal[0..64*2]);
+    let mut vec = Vec::<u64>::with_capacity(8);
+    while vec.len() < 8 {
+        let trans_ciphered_message = trivium.trans_encrypt_64(ciphered_message.pop().unwrap(), &hl_server_key);
+        vec.push(trans_ciphered_message.decrypt(&hl_client_key));
+    }
+
+    let hexadecimal = get_hexagonal_string_from_u64(vec);
+    assert_eq!(output_0_63, hexadecimal[0..64*2]);
 }
 ```
 

apps/trivium/benches/kreyvium_shortint.rs

@@ -1,23 +1,29 @@
 use criterion::Criterion;
 use tfhe::prelude::*;
-use tfhe::shortint::parameters::PARAM_MESSAGE_1_CARRY_1_KS_PBS_GAUSSIAN_2M64;
+use tfhe::shortint::parameters::v1_1::{
+    V1_1_PARAM_KEYSWITCH_1_1_KS_PBS_TO_2_2_KS_PBS_GAUSSIAN_2M128,
+    V1_1_PARAM_MESSAGE_1_CARRY_1_KS_PBS_GAUSSIAN_2M128,
+    V1_1_PARAM_MESSAGE_2_CARRY_2_KS_PBS_GAUSSIAN_2M128,
+};
 use tfhe::shortint::prelude::*;
 use tfhe::{generate_keys, ConfigBuilder, FheUint64};
 use tfhe_trivium::{KreyviumStreamShortint, TransCiphering};
 
 pub fn kreyvium_shortint_warmup(c: &mut Criterion) {
-    let config = ConfigBuilder::default().build();
+    let config = ConfigBuilder::default()
+        .use_custom_parameters(V1_1_PARAM_MESSAGE_2_CARRY_2_KS_PBS_GAUSSIAN_2M128)
+        .build();
     let (hl_client_key, hl_server_key) = generate_keys(config);
     let underlying_ck: tfhe::shortint::ClientKey = (*hl_client_key.as_ref()).clone().into();
     let underlying_sk: tfhe::shortint::ServerKey = (*hl_server_key.as_ref()).clone().into();
 
     let (client_key, server_key): (ClientKey, ServerKey) =
-        gen_keys(PARAM_MESSAGE_1_CARRY_1_KS_PBS_GAUSSIAN_2M64);
+        gen_keys(V1_1_PARAM_MESSAGE_1_CARRY_1_KS_PBS_GAUSSIAN_2M128);
 
     let ksk = KeySwitchingKey::new(
         (&client_key, Some(&server_key)),
         (&underlying_ck, &underlying_sk),
-        PARAM_KEYSWITCH_1_1_KS_PBS_TO_2_2_KS_PBS,
+        V1_1_PARAM_KEYSWITCH_1_1_KS_PBS_TO_2_2_KS_PBS_GAUSSIAN_2M128,
     );
 
     let key_string = "0053A6F94C9FF24598EB000000000000".to_string();
@@ -57,18 +63,20 @@ pub fn kreyvium_shortint_warmup(c: &mut Criterion) {
 }
 
 pub fn kreyvium_shortint_gen(c: &mut Criterion) {
-    let config = ConfigBuilder::default().build();
+    let config = ConfigBuilder::default()
+        .use_custom_parameters(V1_1_PARAM_MESSAGE_2_CARRY_2_KS_PBS_GAUSSIAN_2M128)
+        .build();
     let (hl_client_key, hl_server_key) = generate_keys(config);
     let underlying_ck: tfhe::shortint::ClientKey = (*hl_client_key.as_ref()).clone().into();
     let underlying_sk: tfhe::shortint::ServerKey = (*hl_server_key.as_ref()).clone().into();
 
     let (client_key, server_key): (ClientKey, ServerKey) =
-        gen_keys(PARAM_MESSAGE_1_CARRY_1_KS_PBS_GAUSSIAN_2M64);
+        gen_keys(V1_1_PARAM_MESSAGE_1_CARRY_1_KS_PBS_GAUSSIAN_2M128);
 
     let ksk = KeySwitchingKey::new(
         (&client_key, Some(&server_key)),
         (&underlying_ck, &underlying_sk),
-        PARAM_KEYSWITCH_1_1_KS_PBS_TO_2_2_KS_PBS,
+        V1_1_PARAM_KEYSWITCH_1_1_KS_PBS_TO_2_2_KS_PBS_GAUSSIAN_2M128,
     );
 
     let key_string = "0053A6F94C9FF24598EB000000000000".to_string();
@@ -103,18 +111,20 @@ pub fn kreyvium_shortint_gen(c: &mut Criterion) {
 }
 
 pub fn kreyvium_shortint_trans(c: &mut Criterion) {
-    let config = ConfigBuilder::default().build();
+    let config = ConfigBuilder::default()
+        .use_custom_parameters(V1_1_PARAM_MESSAGE_2_CARRY_2_KS_PBS_GAUSSIAN_2M128)
+        .build();
     let (hl_client_key, hl_server_key) = generate_keys(config);
     let underlying_ck: tfhe::shortint::ClientKey = (*hl_client_key.as_ref()).clone().into();
     let underlying_sk: tfhe::shortint::ServerKey = (*hl_server_key.as_ref()).clone().into();
 
     let (client_key, server_key): (ClientKey, ServerKey) =
-        gen_keys(PARAM_MESSAGE_1_CARRY_1_KS_PBS_GAUSSIAN_2M64);
+        gen_keys(V1_1_PARAM_MESSAGE_1_CARRY_1_KS_PBS_GAUSSIAN_2M128);
 
     let ksk = KeySwitchingKey::new(
         (&client_key, Some(&server_key)),
         (&underlying_ck, &underlying_sk),
-        PARAM_KEYSWITCH_1_1_KS_PBS_TO_2_2_KS_PBS,
+        V1_1_PARAM_KEYSWITCH_1_1_KS_PBS_TO_2_2_KS_PBS_GAUSSIAN_2M128,
     );
 
     let key_string = "0053A6F94C9FF24598EB000000000000".to_string();

apps/trivium/benches/trivium_shortint.rs

@@ -1,23 +1,29 @@
 use criterion::Criterion;
 use tfhe::prelude::*;
-use tfhe::shortint::parameters::PARAM_MESSAGE_1_CARRY_1_KS_PBS_GAUSSIAN_2M64;
+use tfhe::shortint::parameters::v1_1::{
+    V1_1_PARAM_KEYSWITCH_1_1_KS_PBS_TO_2_2_KS_PBS_GAUSSIAN_2M128,
+    V1_1_PARAM_MESSAGE_1_CARRY_1_KS_PBS_GAUSSIAN_2M128,
+    V1_1_PARAM_MESSAGE_2_CARRY_2_KS_PBS_GAUSSIAN_2M128,
+};
 use tfhe::shortint::prelude::*;
 use tfhe::{generate_keys, ConfigBuilder, FheUint64};
 use tfhe_trivium::{TransCiphering, TriviumStreamShortint};
 
 pub fn trivium_shortint_warmup(c: &mut Criterion) {
-    let config = ConfigBuilder::default().build();
+    let config = ConfigBuilder::default()
+        .use_custom_parameters(V1_1_PARAM_MESSAGE_2_CARRY_2_KS_PBS_GAUSSIAN_2M128)
+        .build();
     let (hl_client_key, hl_server_key) = generate_keys(config);
     let underlying_ck: tfhe::shortint::ClientKey = (*hl_client_key.as_ref()).clone().into();
     let underlying_sk: tfhe::shortint::ServerKey = (*hl_server_key.as_ref()).clone().into();
 
     let (client_key, server_key): (ClientKey, ServerKey) =
-        gen_keys(PARAM_MESSAGE_1_CARRY_1_KS_PBS_GAUSSIAN_2M64);
+        gen_keys(V1_1_PARAM_MESSAGE_1_CARRY_1_KS_PBS_GAUSSIAN_2M128);
 
     let ksk = KeySwitchingKey::new(
         (&client_key, Some(&server_key)),
         (&underlying_ck, &underlying_sk),
-        PARAM_KEYSWITCH_1_1_KS_PBS_TO_2_2_KS_PBS,
+        V1_1_PARAM_KEYSWITCH_1_1_KS_PBS_TO_2_2_KS_PBS_GAUSSIAN_2M128,
     );
 
     let key_string = "0053A6F94C9FF24598EB".to_string();
@@ -57,18 +63,20 @@ pub fn trivium_shortint_warmup(c: &mut Criterion) {
 }
 
 pub fn trivium_shortint_gen(c: &mut Criterion) {
-    let config = ConfigBuilder::default().build();
+    let config = ConfigBuilder::default()
+        .use_custom_parameters(V1_1_PARAM_MESSAGE_2_CARRY_2_KS_PBS_GAUSSIAN_2M128)
+        .build();
     let (hl_client_key, hl_server_key) = generate_keys(config);
     let underlying_ck: tfhe::shortint::ClientKey = (*hl_client_key.as_ref()).clone().into();
     let underlying_sk: tfhe::shortint::ServerKey = (*hl_server_key.as_ref()).clone().into();
 
     let (client_key, server_key): (ClientKey, ServerKey) =
-        gen_keys(PARAM_MESSAGE_1_CARRY_1_KS_PBS_GAUSSIAN_2M64);
+        gen_keys(V1_1_PARAM_MESSAGE_1_CARRY_1_KS_PBS_GAUSSIAN_2M128);
 
     let ksk = KeySwitchingKey::new(
         (&client_key, Some(&server_key)),
         (&underlying_ck, &underlying_sk),
-        PARAM_KEYSWITCH_1_1_KS_PBS_TO_2_2_KS_PBS,
+        V1_1_PARAM_KEYSWITCH_1_1_KS_PBS_TO_2_2_KS_PBS_GAUSSIAN_2M128,
     );
 
     let key_string = "0053A6F94C9FF24598EB".to_string();
@@ -103,18 +111,20 @@ pub fn trivium_shortint_gen(c: &mut Criterion) {
 }
 
 pub fn trivium_shortint_trans(c: &mut Criterion) {
-    let config = ConfigBuilder::default().build();
+    let config = ConfigBuilder::default()
+        .use_custom_parameters(V1_1_PARAM_MESSAGE_2_CARRY_2_KS_PBS_GAUSSIAN_2M128)
+        .build();
     let (hl_client_key, hl_server_key) = generate_keys(config);
     let underlying_ck: tfhe::shortint::ClientKey = (*hl_client_key.as_ref()).clone().into();
     let underlying_sk: tfhe::shortint::ServerKey = (*hl_server_key.as_ref()).clone().into();
 
     let (client_key, server_key): (ClientKey, ServerKey) =
-        gen_keys(PARAM_MESSAGE_1_CARRY_1_KS_PBS_GAUSSIAN_2M64);
+        gen_keys(V1_1_PARAM_MESSAGE_1_CARRY_1_KS_PBS_GAUSSIAN_2M128);
 
     let ksk = KeySwitchingKey::new(
         (&client_key, Some(&server_key)),
         (&underlying_ck, &underlying_sk),
-        PARAM_KEYSWITCH_1_1_KS_PBS_TO_2_2_KS_PBS,
+        V1_1_PARAM_KEYSWITCH_1_1_KS_PBS_TO_2_2_KS_PBS_GAUSSIAN_2M128,
     );
 
     let key_string = "0053A6F94C9FF24598EB".to_string();

apps/trivium/src/kreyvium/test.rs

@@ -1,6 +1,10 @@
 use crate::{KreyviumStream, KreyviumStreamByte, KreyviumStreamShortint, TransCiphering};
 use tfhe::prelude::*;
-use tfhe::shortint::parameters::PARAM_MESSAGE_1_CARRY_1_KS_PBS_GAUSSIAN_2M64;
+use tfhe::shortint::parameters::v1_1::{
+    V1_1_PARAM_KEYSWITCH_1_1_KS_PBS_TO_2_2_KS_PBS_GAUSSIAN_2M128,
+    V1_1_PARAM_MESSAGE_1_CARRY_1_KS_PBS_GAUSSIAN_2M128,
+    V1_1_PARAM_MESSAGE_2_CARRY_2_KS_PBS_GAUSSIAN_2M128,
+};
 use tfhe::{generate_keys, ConfigBuilder, FheBool, FheUint64, FheUint8};
 // Values for these tests come from the github repo renaud1239/Kreyvium,
 // commit fd6828f68711276c25f55e605935028f5e843f43
@@ -216,18 +220,20 @@ use tfhe::shortint::prelude::*;
 
 #[test]
 fn kreyvium_test_shortint_long() {
-    let config = ConfigBuilder::default().build();
+    let config = ConfigBuilder::default()
+        .use_custom_parameters(V1_1_PARAM_MESSAGE_2_CARRY_2_KS_PBS_GAUSSIAN_2M128)
+        .build();
     let (hl_client_key, hl_server_key) = generate_keys(config);
     let underlying_ck: tfhe::shortint::ClientKey = (*hl_client_key.as_ref()).clone().into();
     let underlying_sk: tfhe::shortint::ServerKey = (*hl_server_key.as_ref()).clone().into();
 
     let (client_key, server_key): (ClientKey, ServerKey) =
-        gen_keys(PARAM_MESSAGE_1_CARRY_1_KS_PBS_GAUSSIAN_2M64);
+        gen_keys(V1_1_PARAM_MESSAGE_1_CARRY_1_KS_PBS_GAUSSIAN_2M128);
 
     let ksk = KeySwitchingKey::new(
         (&client_key, Some(&server_key)),
         (&underlying_ck, &underlying_sk),
-        PARAM_KEYSWITCH_1_1_KS_PBS_TO_2_2_KS_PBS,
+        V1_1_PARAM_KEYSWITCH_1_1_KS_PBS_TO_2_2_KS_PBS_GAUSSIAN_2M128,
     );
 
     let key_string = "0053A6F94C9FF24598EB000000000000".to_string();

apps/trivium/src/trans_ciphering/mod.rs

@@ -48,6 +48,8 @@ fn transcipher_from_1_1_stream(
 ) -> FheUint64 {
     assert_eq!(stream.len(), 64);
 
+    let id_lut = internal_server_key.generate_lookup_table(|x| x);
+
     let pairs = (0..32)
         .into_par_iter()
         .map(|i| {
@@ -57,10 +59,11 @@ fn transcipher_from_1_1_stream(
             let b0 = &stream[8 * byte_idx + 2 * pair_idx];
             let b1 = &stream[8 * byte_idx + 2 * pair_idx + 1];
 
-            casting_key.cast(
-                &internal_server_key
-                    .unchecked_add(b0, &internal_server_key.unchecked_scalar_mul(b1, 2)),
-            )
+            let mut combined = internal_server_key
+                .unchecked_add(b0, &internal_server_key.unchecked_scalar_mul(b1, 2));
+            internal_server_key.apply_lookup_table_assign(&mut combined, &id_lut);
+
+            casting_key.cast(&combined)
         })
         .collect::<Vec<_>>();
 

apps/trivium/src/trivium/test.rs

@@ -1,6 +1,10 @@
 use crate::{TransCiphering, TriviumStream, TriviumStreamByte, TriviumStreamShortint};
 use tfhe::prelude::*;
-use tfhe::shortint::parameters::PARAM_MESSAGE_1_CARRY_1_KS_PBS_GAUSSIAN_2M64;
+use tfhe::shortint::parameters::v1_1::{
+    V1_1_PARAM_KEYSWITCH_1_1_KS_PBS_TO_2_2_KS_PBS_GAUSSIAN_2M128,
+    V1_1_PARAM_MESSAGE_1_CARRY_1_KS_PBS_GAUSSIAN_2M128,
+    V1_1_PARAM_MESSAGE_2_CARRY_2_KS_PBS_GAUSSIAN_2M128,
+};
 use tfhe::{generate_keys, ConfigBuilder, FheBool, FheUint64, FheUint8};
 // Values for these tests come from the github repo cantora/avr-crypto-lib, commit 2a5b018,
 // file testvectors/trivium-80.80.test-vectors
@@ -352,18 +356,20 @@ use tfhe::shortint::prelude::*;
 
 #[test]
 fn trivium_test_shortint_long() {
-    let config = ConfigBuilder::default().build();
+    let config = ConfigBuilder::default()
+        .use_custom_parameters(V1_1_PARAM_MESSAGE_2_CARRY_2_KS_PBS_GAUSSIAN_2M128)
+        .build();
     let (hl_client_key, hl_server_key) = generate_keys(config);
     let underlying_ck: tfhe::shortint::ClientKey = (*hl_client_key.as_ref()).clone().into();
     let underlying_sk: tfhe::shortint::ServerKey = (*hl_server_key.as_ref()).clone().into();
 
     let (client_key, server_key): (ClientKey, ServerKey) =
-        gen_keys(PARAM_MESSAGE_1_CARRY_1_KS_PBS_GAUSSIAN_2M64);
+        gen_keys(V1_1_PARAM_MESSAGE_1_CARRY_1_KS_PBS_GAUSSIAN_2M128);
 
     let ksk = KeySwitchingKey::new(
         (&client_key, Some(&server_key)),
         (&underlying_ck, &underlying_sk),
-        PARAM_KEYSWITCH_1_1_KS_PBS_TO_2_2_KS_PBS,
+        V1_1_PARAM_KEYSWITCH_1_1_KS_PBS_TO_2_2_KS_PBS_GAUSSIAN_2M128,
     );
 
     let key_string = "0053A6F94C9FF24598EB".to_string();

backends/tfhe-cuda-backend/Cargo.toml

@@ -1,6 +1,6 @@
 [package]
 name = "tfhe-cuda-backend"
-version = "0.6.0"
+version = "0.9.0-alpha.0"
 edition = "2021"
 authors = ["Zama team"]
 license = "BSD-3-Clause-Clear"
@@ -14,4 +14,4 @@ keywords = ["fully", "homomorphic", "encryption", "fhe", "cryptography"]
 [build-dependencies]
 cmake = { version = "0.1" }
 pkg-config = { version = "0.3" }
-bindgen = "0.70.1"
+bindgen = "0.71"

backends/tfhe-cuda-backend/LICENSE

@@ -1,6 +1,6 @@
 BSD 3-Clause Clear License
 
-Copyright © 2024 ZAMA.
+Copyright © 2025 ZAMA.
 All rights reserved.
 
 Redistribution and use in source and binary forms, with or without modification,

backends/tfhe-cuda-backend/README.md

@@ -13,24 +13,26 @@ and forth between the CPU and the GPU, to create and destroy Cuda streams, etc.:
 - `cuda_get_number_of_gpus`
 - `cuda_synchronize_device`
 The cryptographic operations it provides are:
-- an amortized implementation of the TFHE programmable bootstrap: `cuda_bootstrap_amortized_lwe_ciphertext_vector_32` and `cuda_bootstrap_amortized_lwe_ciphertext_vector_64`
-- a low latency implementation of the TFHE programmable bootstrap: `cuda_bootstrap_low latency_lwe_ciphertext_vector_32` and `cuda_bootstrap_low_latency_lwe_ciphertext_vector_64`
-- the keyswitch: `cuda_keyswitch_lwe_ciphertext_vector_32` and `cuda_keyswitch_lwe_ciphertext_vector_64`
-- the larger precision programmable bootstrap (wop PBS, which supports up to 16 bits of message while the classical PBS only supports up to 8 bits of message) and its sub-components: `cuda_wop_pbs_64`, `cuda_extract_bits_64`, `cuda_circuit_bootstrap_64`, `cuda_cmux_tree_64`, `cuda_blind_rotation_sample_extraction_64`
-- acceleration for leveled operations: `cuda_negate_lwe_ciphertext_vector_64`, `cuda_add_lwe_ciphertext_vector_64`, `cuda_add_lwe_ciphertext_vector_plaintext_vector_64`, `cuda_mult_lwe_ciphertext_vector_cleartext_vector`.
+- an implementation of the classical TFHE programmable bootstrap,
+- an implementation of the multi-bit TFHE programmable bootstrap,
+- the keyswitch,
+- acceleration for leveled operations,
+- acceleration for arithmetics over encrypted integers of arbitrary size, 
+- acceleration for integer compression/decompression.
 
 ## Dependencies
 
 **Disclaimer**: Compilation on Windows/Mac is not supported yet. Only Nvidia GPUs are supported. 
 
-- nvidia driver - for example, if you're running Ubuntu 20.04 check this [page](https://linuxconfig.org/how-to-install-the-nvidia-drivers-on-ubuntu-20-04-focal-fossa-linux) for installation
+- nvidia driver - for example, if you're running Ubuntu 20.04 check this [page](https://linuxconfig.org/how-to-install-the-nvidia-drivers-on-ubuntu-20-04-focal-fossa-linux) for installation. You need an Nvidia GPU with Compute Capability >= 3.0
 - [nvcc](https://docs.nvidia.com/cuda/cuda-installation-guide-linux/index.html) >= 10.0
 - [gcc](https://gcc.gnu.org/) >= 8.0 - check this [page](https://gist.github.com/ax3l/9489132) for more details about nvcc/gcc compatible versions
 - [cmake](https://cmake.org/) >= 3.24
+- libclang, to match Rust bingen [requirements](https://rust-lang.github.io/rust-bindgen/requirements.html) >= 9.0
 
 ## Build
 
-The Cuda project held in `tfhe-cuda-backend` can be compiled independently from TFHE-rs in the following way:
+The Cuda project held in `tfhe-cuda-backend` can be compiled independently of TFHE-rs in the following way:
 ```
 git clone git@github.com:zama-ai/tfhe-rs
 cd backends/tfhe-cuda-backend/cuda

backends/tfhe-cuda-backend/build.rs

@@ -62,6 +62,7 @@ fn main() {
             "cuda/include/integer/integer.h",
             "cuda/include/keyswitch.h",
             "cuda/include/linear_algebra.h",
+            "cuda/include/fft/fft128.h",
             "cuda/include/pbs/programmable_bootstrap.h",
             "cuda/include/pbs/programmable_bootstrap_multibit.h",
         ];

backends/tfhe-cuda-backend/cuda/include/ciphertext.h

@@ -18,7 +18,7 @@ void cuda_convert_lwe_ciphertext_vector_to_cpu_64(void *stream,
 void cuda_glwe_sample_extract_64(void *stream, uint32_t gpu_index,
                                  void *lwe_array_out, void const *glwe_array_in,
                                  uint32_t const *nth_array, uint32_t num_nths,
-                                 uint32_t glwe_dimension,
+                                 uint32_t lwe_per_glwe, uint32_t glwe_dimension,
                                  uint32_t polynomial_size);
 }
 #endif

backends/tfhe-cuda-backend/cuda/include/device.h

@@ -27,19 +27,32 @@ inline void cuda_error(cudaError_t code, const char *file, int line) {
     std::abort();                                                              \
   }
 
+void cuda_set_device(uint32_t gpu_index);
+
+cudaEvent_t cuda_create_event(uint32_t gpu_index);
+
+void cuda_event_record(cudaEvent_t event, cudaStream_t stream,
+                       uint32_t gpu_index);
+void cuda_stream_wait_event(cudaStream_t stream, cudaEvent_t event,
+                            uint32_t gpu_index);
+
+void cuda_event_destroy(cudaEvent_t event, uint32_t gpu_index);
+
 cudaStream_t cuda_create_stream(uint32_t gpu_index);
 
 void cuda_destroy_stream(cudaStream_t stream, uint32_t gpu_index);
 
 void cuda_synchronize_stream(cudaStream_t stream, uint32_t gpu_index);
 
+uint32_t cuda_is_available();
+
 void *cuda_malloc(uint64_t size, uint32_t gpu_index);
 
 void *cuda_malloc_async(uint64_t size, cudaStream_t stream, uint32_t gpu_index);
 
 void cuda_check_valid_malloc(uint64_t size, uint32_t gpu_index);
 
-void cuda_memcpy_async_to_gpu(void *dest, void *src, uint64_t size,
+void cuda_memcpy_async_to_gpu(void *dest, const void *src, uint64_t size,
                               cudaStream_t stream, uint32_t gpu_index);
 
 void cuda_memcpy_async_gpu_to_gpu(void *dest, void const *src, uint64_t size,
@@ -63,7 +76,7 @@ void cuda_drop(void *ptr, uint32_t gpu_index);
 void cuda_drop_async(void *ptr, cudaStream_t stream, uint32_t gpu_index);
 }
 
-int cuda_get_max_shared_memory(uint32_t gpu_index);
+uint32_t cuda_get_max_shared_memory(uint32_t gpu_index);
 
 bool cuda_check_support_cooperative_groups();
 

backends/tfhe-cuda-backend/cuda/include/fft/fft128.h

@@ -0,0 +1,17 @@
+#include <stdint.h>
+extern "C" {
+void cuda_fourier_transform_forward_as_torus_f128_async(
+    void *stream, uint32_t gpu_index, void *re0, void *re1, void *im0,
+    void *im1, void const *standard, uint32_t const N,
+    const uint32_t number_of_samples);
+
+void cuda_fourier_transform_forward_as_integer_f128_async(
+    void *stream, uint32_t gpu_index, void *re0, void *re1, void *im0,
+    void *im1, void const *standard, uint32_t const N,
+    const uint32_t number_of_samples);
+
+void cuda_fourier_transform_backward_as_torus_f128_async(
+    void *stream, uint32_t gpu_index, void *standard, void const *re0,
+    void const *re1, void const *im0, void const *im1, uint32_t const N,
+    const uint32_t number_of_samples);
+}

backends/tfhe-cuda-backend/cuda/include/integer/compression/compression_utilities.h

@@ -38,6 +38,7 @@ template <typename Torus> struct int_compression {
 
       scratch_packing_keyswitch_lwe_list_to_glwe_64(
           streams[0], gpu_indexes[0], &fp_ks_buffer,
+          compression_params.small_lwe_dimension,
           compression_params.glwe_dimension, compression_params.polynomial_size,
           num_radix_blocks, true);
     }
@@ -64,7 +65,7 @@ template <typename Torus> struct int_decompression {
   Torus *tmp_extracted_lwe;
   uint32_t *tmp_indexes_array;
 
-  int_radix_lut<Torus> *carry_extract_lut;
+  int_radix_lut<Torus> *decompression_rescale_lut;
 
   int_decompression(cudaStream_t const *streams, uint32_t const *gpu_indexes,
                     uint32_t gpu_count, int_radix_params encryption_params,
@@ -83,7 +84,7 @@ template <typename Torus> struct int_decompression {
       Torus lwe_accumulator_size = (compression_params.glwe_dimension *
                                         compression_params.polynomial_size +
                                     1);
-      carry_extract_lut = new int_radix_lut<Torus>(
+      decompression_rescale_lut = new int_radix_lut<Torus>(
           streams, gpu_indexes, gpu_count, encryption_params, 1,
           num_radix_blocks, allocate_gpu_memory);
 
@@ -96,19 +97,28 @@ template <typename Torus> struct int_decompression {
           num_radix_blocks * lwe_accumulator_size * sizeof(Torus), streams[0],
           gpu_indexes[0]);
 
-      // Carry extract LUT
-      auto carry_extract_f = [encryption_params](Torus x) -> Torus {
-        return x / encryption_params.message_modulus;
-      };
+      // Rescale is done using an identity LUT
+      // Here we do not divide by message_modulus
+      // Example: in the 2_2 case we are mapping a 2 bits message onto a 4 bits
+      // space, we want to keep the original 2 bits value in the 4 bits space,
+      // so we apply the identity and the encoding will rescale it for us.
+      auto decompression_rescale_f = [](Torus x) -> Torus { return x; };
 
-      generate_device_accumulator<Torus>(
-          streams[0], gpu_indexes[0],
-          carry_extract_lut->get_lut(gpu_indexes[0], 0),
+      auto effective_compression_message_modulus =
+          encryption_params.carry_modulus;
+      auto effective_compression_carry_modulus = 1;
+
+      generate_device_accumulator_with_encoding<Torus>(
+          streams[0], gpu_indexes[0], decompression_rescale_lut->get_lut(0, 0),
+          decompression_rescale_lut->get_degree(0),
+          decompression_rescale_lut->get_max_degree(0),
           encryption_params.glwe_dimension, encryption_params.polynomial_size,
+          effective_compression_message_modulus,
+          effective_compression_carry_modulus,
           encryption_params.message_modulus, encryption_params.carry_modulus,
-          carry_extract_f);
+          decompression_rescale_f);
 
-      carry_extract_lut->broadcast_lut(streams, gpu_indexes, gpu_indexes[0]);
+      decompression_rescale_lut->broadcast_lut(streams, gpu_indexes, 0);
     }
   }
   void release(cudaStream_t const *streams, uint32_t const *gpu_indexes,
@@ -117,8 +127,8 @@ template <typename Torus> struct int_decompression {
     cuda_drop_async(tmp_extracted_lwe, streams[0], gpu_indexes[0]);
     cuda_drop_async(tmp_indexes_array, streams[0], gpu_indexes[0]);
 
-    carry_extract_lut->release(streams, gpu_indexes, gpu_count);
-    delete carry_extract_lut;
+    decompression_rescale_lut->release(streams, gpu_indexes, gpu_count);
+    delete decompression_rescale_lut;
   }
 };
 #endif

backends/tfhe-cuda-backend/cuda/include/integer/integer.h

@@ -35,7 +35,19 @@ enum CMP_ORDERING { IS_INFERIOR = 0, IS_EQUAL = 1, IS_SUPERIOR = 2 };
 
 enum SIGNED_OPERATION { ADDITION = 1, SUBTRACTION = -1 };
 
+enum outputFlag { FLAG_NONE = 0, FLAG_OVERFLOW = 1, FLAG_CARRY = 2 };
+
 extern "C" {
+
+typedef struct {
+  void *ptr;
+  uint64_t *degrees;
+  uint64_t *noise_levels;
+  uint32_t num_radix_blocks;
+  uint32_t max_num_radix_blocks;
+  uint32_t lwe_dimension;
+} CudaRadixCiphertextFFI;
+
 void scratch_cuda_apply_univariate_lut_kb_64(
     void *const *streams, uint32_t const *gpu_indexes, uint32_t gpu_count,
     int8_t **mem_ptr, void const *input_lut, uint32_t lwe_dimension,
@@ -43,14 +55,20 @@ void scratch_cuda_apply_univariate_lut_kb_64(
     uint32_t ks_base_log, uint32_t pbs_level, uint32_t pbs_base_log,
     uint32_t grouping_factor, uint32_t input_lwe_ciphertext_count,
     uint32_t message_modulus, uint32_t carry_modulus, PBS_TYPE pbs_type,
-    bool allocate_gpu_memory);
-
-void cuda_apply_univariate_lut_kb_64(void *const *streams,
-                                     uint32_t const *gpu_indexes,
-                                     uint32_t gpu_count, void *output_radix_lwe,
-                                     void const *input_radix_lwe,
-                                     int8_t *mem_ptr, void *const *ksks,
-                                     void *const *bsks, uint32_t num_blocks);
+    uint64_t lut_degree, bool allocate_gpu_memory);
+void scratch_cuda_apply_many_univariate_lut_kb_64(
+    void *const *streams, uint32_t const *gpu_indexes, uint32_t gpu_count,
+    int8_t **mem_ptr, void const *input_lut, uint32_t lwe_dimension,
+    uint32_t glwe_dimension, uint32_t polynomial_size, uint32_t ks_level,
+    uint32_t ks_base_log, uint32_t pbs_level, uint32_t pbs_base_log,
+    uint32_t grouping_factor, uint32_t num_radix_blocks,
+    uint32_t message_modulus, uint32_t carry_modulus, PBS_TYPE pbs_type,
+    uint32_t num_many_lut, uint64_t lut_degree, bool allocate_gpu_memory);
+void cuda_apply_univariate_lut_kb_64(
+    void *const *streams, uint32_t const *gpu_indexes, uint32_t gpu_count,
+    CudaRadixCiphertextFFI *output_radix_lwe,
+    CudaRadixCiphertextFFI const *input_radix_lwe, int8_t *mem_ptr,
+    void *const *ksks, void *const *bsks);
 
 void cleanup_cuda_apply_univariate_lut_kb_64(void *const *streams,
                                              uint32_t const *gpu_indexes,
@@ -64,13 +82,15 @@ void scratch_cuda_apply_bivariate_lut_kb_64(
     uint32_t ks_base_log, uint32_t pbs_level, uint32_t pbs_base_log,
     uint32_t grouping_factor, uint32_t input_lwe_ciphertext_count,
     uint32_t message_modulus, uint32_t carry_modulus, PBS_TYPE pbs_type,
-    bool allocate_gpu_memory);
+    uint64_t lut_degree, bool allocate_gpu_memory);
 
 void cuda_apply_bivariate_lut_kb_64(
     void *const *streams, uint32_t const *gpu_indexes, uint32_t gpu_count,
-    void *output_radix_lwe, void const *input_radix_lwe_1,
-    void const *input_radix_lwe_2, int8_t *mem_ptr, void *const *ksks,
-    void *const *bsks, uint32_t num_blocks, uint32_t shift);
+    CudaRadixCiphertextFFI *output_radix_lwe,
+    CudaRadixCiphertextFFI const *input_radix_lwe_1,
+    CudaRadixCiphertextFFI const *input_radix_lwe_2, int8_t *mem_ptr,
+    void *const *ksks, void *const *bsks, uint32_t num_radix_blocks,
+    uint32_t shift);
 
 void cleanup_cuda_apply_bivariate_lut_kb_64(void *const *streams,
                                             uint32_t const *gpu_indexes,
@@ -79,9 +99,10 @@ void cleanup_cuda_apply_bivariate_lut_kb_64(void *const *streams,
 
 void cuda_apply_many_univariate_lut_kb_64(
     void *const *streams, uint32_t const *gpu_indexes, uint32_t gpu_count,
-    void *output_radix_lwe, void const *input_radix_lwe, int8_t *mem_ptr,
-    void *const *ksks, void *const *bsks, uint32_t num_blocks,
-    uint32_t num_luts, uint32_t lut_stride);
+    CudaRadixCiphertextFFI *output_radix_lwe,
+    CudaRadixCiphertextFFI const *input_radix_lwe, int8_t *mem_ptr,
+    void *const *ksks, void *const *bsks, uint32_t num_luts,
+    uint32_t lut_stride);
 
 void scratch_cuda_full_propagation_64(
     void *const *streams, uint32_t const *gpu_indexes, uint32_t gpu_count,
@@ -93,7 +114,8 @@ void scratch_cuda_full_propagation_64(
 
 void cuda_full_propagation_64_inplace(void *const *streams,
                                       uint32_t const *gpu_indexes,
-                                      uint32_t gpu_count, void *input_blocks,
+                                      uint32_t gpu_count,
+                                      CudaRadixCiphertextFFI *input_blocks,
                                       int8_t *mem_ptr, void *const *ksks,
                                       void *const *bsks, uint32_t num_blocks);
 
@@ -112,10 +134,11 @@ void scratch_cuda_integer_mult_radix_ciphertext_kb_64(
 
 void cuda_integer_mult_radix_ciphertext_kb_64(
     void *const *streams, uint32_t const *gpu_indexes, uint32_t gpu_count,
-    void *radix_lwe_out, void const *radix_lwe_left, bool const is_bool_left,
-    void const *radix_lwe_right, bool const is_bool_right, void *const *bsks,
-    void *const *ksks, int8_t *mem_ptr, uint32_t polynomial_size,
-    uint32_t num_blocks);
+    CudaRadixCiphertextFFI *radix_lwe_out,
+    CudaRadixCiphertextFFI const *radix_lwe_left, bool const is_bool_left,
+    CudaRadixCiphertextFFI const *radix_lwe_right, bool const is_bool_right,
+    void *const *bsks, void *const *ksks, int8_t *mem_ptr,
+    uint32_t polynomial_size, uint32_t num_blocks);
 
 void cleanup_cuda_integer_mult(void *const *streams,
                                uint32_t const *gpu_indexes, uint32_t gpu_count,
@@ -123,14 +146,14 @@ void cleanup_cuda_integer_mult(void *const *streams,
 
 void cuda_negate_integer_radix_ciphertext_64(
     void *const *streams, uint32_t const *gpu_indexes, uint32_t gpu_count,
-    void *lwe_array_out, void const *lwe_array_in, uint32_t lwe_dimension,
-    uint32_t lwe_ciphertext_count, uint32_t message_modulus,
-    uint32_t carry_modulus);
+    CudaRadixCiphertextFFI *lwe_array_out,
+    CudaRadixCiphertextFFI const *lwe_array_in, uint32_t message_modulus,
+    uint32_t carry_modulus, uint32_t num_radix_blocks);
 
 void cuda_scalar_addition_integer_radix_ciphertext_64_inplace(
     void *const *streams, uint32_t const *gpu_indexes, uint32_t gpu_count,
-    void *lwe_array, void const *scalar_input, uint32_t lwe_dimension,
-    uint32_t lwe_ciphertext_count, uint32_t message_modulus,
+    CudaRadixCiphertextFFI *lwe_array, void const *scalar_input,
+    void const *h_scalar_input, uint32_t num_scalars, uint32_t message_modulus,
     uint32_t carry_modulus);
 
 void scratch_cuda_integer_radix_logical_scalar_shift_kb_64(
@@ -144,8 +167,8 @@ void scratch_cuda_integer_radix_logical_scalar_shift_kb_64(
 
 void cuda_integer_radix_logical_scalar_shift_kb_64_inplace(
     void *const *streams, uint32_t const *gpu_indexes, uint32_t gpu_count,
-    void *lwe_array, uint32_t shift, int8_t *mem_ptr, void *const *bsks,
-    void *const *ksks, uint32_t num_blocks);
+    CudaRadixCiphertextFFI *lwe_array, uint32_t shift, int8_t *mem_ptr,
+    void *const *bsks, void *const *ksks);
 
 void scratch_cuda_integer_radix_arithmetic_scalar_shift_kb_64(
     void *const *streams, uint32_t const *gpu_indexes, uint32_t gpu_count,
@@ -158,8 +181,8 @@ void scratch_cuda_integer_radix_arithmetic_scalar_shift_kb_64(
 
 void cuda_integer_radix_arithmetic_scalar_shift_kb_64_inplace(
     void *const *streams, uint32_t const *gpu_indexes, uint32_t gpu_count,
-    void *lwe_array, uint32_t shift, int8_t *mem_ptr, void *const *bsks,
-    void *const *ksks, uint32_t num_blocks);
+    CudaRadixCiphertextFFI *lwe_array, uint32_t shift, int8_t *mem_ptr,
+    void *const *bsks, void *const *ksks);
 
 void cleanup_cuda_integer_radix_logical_scalar_shift(
     void *const *streams, uint32_t const *gpu_indexes, uint32_t gpu_count,
@@ -180,8 +203,8 @@ void scratch_cuda_integer_radix_shift_and_rotate_kb_64(
 
 void cuda_integer_radix_shift_and_rotate_kb_64_inplace(
     void *const *streams, uint32_t const *gpu_indexes, uint32_t gpu_count,
-    void *lwe_array, void const *lwe_shift, int8_t *mem_ptr, void *const *bsks,
-    void *const *ksks, uint32_t num_blocks);
+    CudaRadixCiphertextFFI *lwe_array, CudaRadixCiphertextFFI const *lwe_shift,
+    int8_t *mem_ptr, void *const *bsks, void *const *ksks);
 
 void cleanup_cuda_integer_radix_shift_and_rotate(void *const *streams,
                                                  uint32_t const *gpu_indexes,
@@ -199,15 +222,17 @@ void scratch_cuda_integer_radix_comparison_kb_64(
 
 void cuda_comparison_integer_radix_ciphertext_kb_64(
     void *const *streams, uint32_t const *gpu_indexes, uint32_t gpu_count,
-    void *lwe_array_out, void const *lwe_array_1, void const *lwe_array_2,
-    int8_t *mem_ptr, void *const *bsks, void *const *ksks,
-    uint32_t lwe_ciphertext_count);
+    CudaRadixCiphertextFFI *lwe_array_out,
+    CudaRadixCiphertextFFI const *lwe_array_1,
+    CudaRadixCiphertextFFI const *lwe_array_2, int8_t *mem_ptr,
+    void *const *bsks, void *const *ksks);
 
 void cuda_scalar_comparison_integer_radix_ciphertext_kb_64(
     void *const *streams, uint32_t const *gpu_indexes, uint32_t gpu_count,
-    void *lwe_array_out, void const *lwe_array_in, void const *scalar_blocks,
-    int8_t *mem_ptr, void *const *bsks, void *const *ksks,
-    uint32_t lwe_ciphertext_count, uint32_t num_scalar_blocks);
+    CudaRadixCiphertextFFI *lwe_array_out,
+    CudaRadixCiphertextFFI const *lwe_array_in, void const *scalar_blocks,
+    void const *h_scalar_blocks, int8_t *mem_ptr, void *const *bsks,
+    void *const *ksks, uint32_t num_scalar_blocks);
 
 void cleanup_cuda_integer_comparison(void *const *streams,
                                      uint32_t const *gpu_indexes,
@@ -224,15 +249,17 @@ void scratch_cuda_integer_radix_bitop_kb_64(
 
 void cuda_bitop_integer_radix_ciphertext_kb_64(
     void *const *streams, uint32_t const *gpu_indexes, uint32_t gpu_count,
-    void *lwe_array_out, void const *lwe_array_1, void const *lwe_array_2,
-    int8_t *mem_ptr, void *const *bsks, void *const *ksks,
-    uint32_t lwe_ciphertext_count);
+    CudaRadixCiphertextFFI *lwe_array_out,
+    CudaRadixCiphertextFFI const *lwe_array_1,
+    CudaRadixCiphertextFFI const *lwe_array_2, int8_t *mem_ptr,
+    void *const *bsks, void *const *ksks);
 
 void cuda_scalar_bitop_integer_radix_ciphertext_kb_64(
     void *const *streams, uint32_t const *gpu_indexes, uint32_t gpu_count,
-    void *lwe_array_out, void const *lwe_array_input, void const *clear_blocks,
-    uint32_t num_clear_blocks, int8_t *mem_ptr, void *const *bsks,
-    void *const *ksks, uint32_t lwe_ciphertext_count, BITOP_TYPE op);
+    CudaRadixCiphertextFFI *lwe_array_out,
+    CudaRadixCiphertextFFI const *lwe_array_input, void const *clear_blocks,
+    void const *h_clear_blocks, uint32_t num_clear_blocks, int8_t *mem_ptr,
+    void *const *bsks, void *const *ksks);
 
 void cleanup_cuda_integer_bitop(void *const *streams,
                                 uint32_t const *gpu_indexes, uint32_t gpu_count,
@@ -249,9 +276,11 @@ void scratch_cuda_integer_radix_cmux_kb_64(
 
 void cuda_cmux_integer_radix_ciphertext_kb_64(
     void *const *streams, uint32_t const *gpu_indexes, uint32_t gpu_count,
-    void *lwe_array_out, void const *lwe_condition, void const *lwe_array_true,
-    void const *lwe_array_false, int8_t *mem_ptr, void *const *bsks,
-    void *const *ksks, uint32_t lwe_ciphertext_count);
+    CudaRadixCiphertextFFI *lwe_array_out,
+    CudaRadixCiphertextFFI const *lwe_condition,
+    CudaRadixCiphertextFFI const *lwe_array_true,
+    CudaRadixCiphertextFFI const *lwe_array_false, int8_t *mem_ptr,
+    void *const *bsks, void *const *ksks);
 
 void cleanup_cuda_integer_radix_cmux(void *const *streams,
                                      uint32_t const *gpu_indexes,
@@ -268,8 +297,8 @@ void scratch_cuda_integer_radix_scalar_rotate_kb_64(
 
 void cuda_integer_radix_scalar_rotate_kb_64_inplace(
     void *const *streams, uint32_t const *gpu_indexes, uint32_t gpu_count,
-    void *lwe_array, uint32_t n, int8_t *mem_ptr, void *const *bsks,
-    void *const *ksks, uint32_t num_blocks);
+    CudaRadixCiphertextFFI *lwe_array, uint32_t n, int8_t *mem_ptr,
+    void *const *bsks, void *const *ksks);
 
 void cleanup_cuda_integer_radix_scalar_rotate(void *const *streams,
                                               uint32_t const *gpu_indexes,
@@ -282,23 +311,63 @@ void scratch_cuda_propagate_single_carry_kb_64_inplace(
     uint32_t big_lwe_dimension, uint32_t small_lwe_dimension, uint32_t ks_level,
     uint32_t ks_base_log, uint32_t pbs_level, uint32_t pbs_base_log,
     uint32_t grouping_factor, uint32_t num_blocks, uint32_t message_modulus,
-    uint32_t carry_modulus, PBS_TYPE pbs_type, bool allocate_gpu_memory);
+    uint32_t carry_modulus, PBS_TYPE pbs_type, uint32_t requested_flag,
+    uint32_t uses_carry, bool allocate_gpu_memory);
+
+void scratch_cuda_add_and_propagate_single_carry_kb_64_inplace(
+    void *const *streams, uint32_t const *gpu_indexes, uint32_t gpu_count,
+    int8_t **mem_ptr, uint32_t glwe_dimension, uint32_t polynomial_size,
+    uint32_t big_lwe_dimension, uint32_t small_lwe_dimension, uint32_t ks_level,
+    uint32_t ks_base_log, uint32_t pbs_level, uint32_t pbs_base_log,
+    uint32_t grouping_factor, uint32_t num_blocks, uint32_t message_modulus,
+    uint32_t carry_modulus, PBS_TYPE pbs_type, uint32_t requested_flag,
+    uint32_t uses_carry, bool allocate_gpu_memory);
 
 void cuda_propagate_single_carry_kb_64_inplace(
     void *const *streams, uint32_t const *gpu_indexes, uint32_t gpu_count,
-    void *lwe_array, void *carry_out, int8_t *mem_ptr, void *const *bsks,
-    void *const *ksks, uint32_t num_blocks);
+    CudaRadixCiphertextFFI *lwe_array, CudaRadixCiphertextFFI *carry_out,
+    const CudaRadixCiphertextFFI *carry_in, int8_t *mem_ptr, void *const *bsks,
+    void *const *ksks, uint32_t requested_flag, uint32_t uses_carry);
 
-void cuda_propagate_single_carry_get_input_carries_kb_64_inplace(
+void cuda_add_and_propagate_single_carry_kb_64_inplace(
     void *const *streams, uint32_t const *gpu_indexes, uint32_t gpu_count,
-    void *lwe_array, void *carry_out, void *input_carries, int8_t *mem_ptr,
-    void *const *bsks, void *const *ksks, uint32_t num_blocks);
+    CudaRadixCiphertextFFI *lhs_array, const CudaRadixCiphertextFFI *rhs_array,
+    CudaRadixCiphertextFFI *carry_out, const CudaRadixCiphertextFFI *carry_in,
+    int8_t *mem_ptr, void *const *bsks, void *const *ksks,
+    uint32_t requested_flag, uint32_t uses_carry);
 
 void cleanup_cuda_propagate_single_carry(void *const *streams,
                                          uint32_t const *gpu_indexes,
                                          uint32_t gpu_count,
                                          int8_t **mem_ptr_void);
 
+void cleanup_cuda_add_and_propagate_single_carry(void *const *streams,
+                                                 uint32_t const *gpu_indexes,
+                                                 uint32_t gpu_count,
+                                                 int8_t **mem_ptr_void);
+
+void scratch_cuda_integer_overflowing_sub_kb_64_inplace(
+    void *const *streams, uint32_t const *gpu_indexes, uint32_t gpu_count,
+    int8_t **mem_ptr, uint32_t glwe_dimension, uint32_t polynomial_size,
+    uint32_t big_lwe_dimension, uint32_t small_lwe_dimension, uint32_t ks_level,
+    uint32_t ks_base_log, uint32_t pbs_level, uint32_t pbs_base_log,
+    uint32_t grouping_factor, uint32_t num_blocks, uint32_t message_modulus,
+    uint32_t carry_modulus, PBS_TYPE pbs_type, uint32_t compute_overflow,
+    bool allocate_gpu_memory);
+
+void cuda_integer_overflowing_sub_kb_64_inplace(
+    void *const *streams, uint32_t const *gpu_indexes, uint32_t gpu_count,
+    CudaRadixCiphertextFFI *lhs_array, const CudaRadixCiphertextFFI *rhs_array,
+    CudaRadixCiphertextFFI *overflow_block,
+    const CudaRadixCiphertextFFI *input_borrow, int8_t *mem_ptr,
+    void *const *bsks, void *const *ksks, uint32_t compute_overflow,
+    uint32_t uses_input_borrow);
+
+void cleanup_cuda_integer_overflowing_sub(void *const *streams,
+                                          uint32_t const *gpu_indexes,
+                                          uint32_t gpu_count,
+                                          int8_t **mem_ptr_void);
+
 void scratch_cuda_integer_radix_partial_sum_ciphertexts_vec_kb_64(
     void *const *streams, uint32_t const *gpu_indexes, uint32_t gpu_count,
     int8_t **mem_ptr, uint32_t glwe_dimension, uint32_t polynomial_size,
@@ -310,33 +379,14 @@ void scratch_cuda_integer_radix_partial_sum_ciphertexts_vec_kb_64(
 
 void cuda_integer_radix_partial_sum_ciphertexts_vec_kb_64(
     void *const *streams, uint32_t const *gpu_indexes, uint32_t gpu_count,
-    void *radix_lwe_out, void *radix_lwe_vec, uint32_t num_radix_in_vec,
-    int8_t *mem_ptr, void *const *bsks, void *const *ksks,
-    uint32_t num_blocks_in_radix);
+    CudaRadixCiphertextFFI *radix_lwe_out,
+    CudaRadixCiphertextFFI *radix_lwe_vec, int8_t *mem_ptr, void *const *bsks,
+    void *const *ksks);
 
 void cleanup_cuda_integer_radix_partial_sum_ciphertexts_vec(
     void *const *streams, uint32_t const *gpu_indexes, uint32_t gpu_count,
     int8_t **mem_ptr_void);
 
-void scratch_cuda_integer_radix_overflowing_sub_kb_64(
-    void *const *streams, uint32_t const *gpu_indexes, uint32_t gpu_count,
-    int8_t **mem_ptr, uint32_t glwe_dimension, uint32_t polynomial_size,
-    uint32_t big_lwe_dimension, uint32_t small_lwe_dimension, uint32_t ks_level,
-    uint32_t ks_base_log, uint32_t pbs_level, uint32_t pbs_base_log,
-    uint32_t grouping_factor, uint32_t num_blocks, uint32_t message_modulus,
-    uint32_t carry_modulus, PBS_TYPE pbs_type, bool allocate_gpu_memory);
-
-void cuda_integer_radix_overflowing_sub_kb_64(
-    void *const *streams, uint32_t const *gpu_indexes, uint32_t gpu_count,
-    void *radix_lwe_out, void *radix_lwe_overflowed, void const *radix_lwe_left,
-    void const *radix_lwe_right, int8_t *mem_ptr, void *const *bsks,
-    void *const *ksks, uint32_t num_blocks_in_radix);
-
-void cleanup_cuda_integer_radix_overflowing_sub(void *const *streams,
-                                                uint32_t const *gpu_indexes,
-                                                uint32_t gpu_count,
-                                                int8_t **mem_ptr_void);
-
 void scratch_cuda_integer_scalar_mul_kb_64(
     void *const *streams, uint32_t const *gpu_indexes, uint32_t gpu_count,
     int8_t **mem_ptr, uint32_t glwe_dimension, uint32_t polynomial_size,
@@ -347,10 +397,10 @@ void scratch_cuda_integer_scalar_mul_kb_64(
 
 void cuda_scalar_multiplication_integer_radix_ciphertext_64_inplace(
     void *const *streams, uint32_t const *gpu_indexes, uint32_t gpu_count,
-    void *lwe_array, uint64_t const *decomposed_scalar,
+    CudaRadixCiphertextFFI *lwe_array, uint64_t const *decomposed_scalar,
     uint64_t const *has_at_least_one_set, int8_t *mem_ptr, void *const *bsks,
-    void *const *ksks, uint32_t lwe_dimension, uint32_t polynomial_size,
-    uint32_t message_modulus, uint32_t num_blocks, uint32_t num_scalars);
+    void *const *ksks, uint32_t polynomial_size, uint32_t message_modulus,
+    uint32_t num_scalars);
 
 void cleanup_cuda_integer_radix_scalar_mul(void *const *streams,
                                            uint32_t const *gpu_indexes,
@@ -368,34 +418,15 @@ void scratch_cuda_integer_div_rem_radix_ciphertext_kb_64(
 
 void cuda_integer_div_rem_radix_ciphertext_kb_64(
     void *const *streams, uint32_t const *gpu_indexes, uint32_t gpu_count,
-    void *quotient, void *remainder, void const *numerator, void const *divisor,
-    bool is_signed, int8_t *mem_ptr, void *const *bsks, void *const *ksks,
-    uint32_t num_blocks_in_radix);
+    CudaRadixCiphertextFFI *quotient, CudaRadixCiphertextFFI *remainder,
+    CudaRadixCiphertextFFI const *numerator,
+    CudaRadixCiphertextFFI const *divisor, bool is_signed, int8_t *mem_ptr,
+    void *const *bsks, void *const *ksks);
 
 void cleanup_cuda_integer_div_rem(void *const *streams,
                                   uint32_t const *gpu_indexes,
                                   uint32_t gpu_count, int8_t **mem_ptr_void);
 
-void scratch_cuda_signed_overflowing_add_or_sub_radix_ciphertext_kb_64(
-    void *const *streams, uint32_t const *gpu_indexes, uint32_t gpu_count,
-    int8_t **mem_ptr, uint32_t glwe_dimension, uint32_t polynomial_size,
-    uint32_t big_lwe_dimension, uint32_t small_lwe_dimension, uint32_t ks_level,
-    uint32_t ks_base_log, uint32_t pbs_level, uint32_t pbs_base_log,
-    uint32_t grouping_factor, uint32_t num_blocks, int8_t signed_operation,
-    uint32_t message_modulus, uint32_t carry_modulus, PBS_TYPE pbs_type,
-    bool allocate_gpu_memory);
-
-void cuda_signed_overflowing_add_or_sub_radix_ciphertext_kb_64(
-    void *const *streams, uint32_t const *gpu_indexes, uint32_t gpu_count,
-    void *lhs, void const *rhs, void *overflowed, int8_t signed_operation,
-    int8_t *mem_ptr, void *const *bsks, void *const *ksks,
-    uint32_t num_blocks_in_radix);
-
-void cleanup_signed_overflowing_add_or_sub(void *const *streams,
-                                           uint32_t const *gpu_indexes,
-                                           uint32_t gpu_count,
-                                           int8_t **mem_ptr_void);
-
 void scratch_cuda_integer_compute_prefix_sum_hillis_steele_64(
     void *const *streams, uint32_t const *gpu_indexes, uint32_t gpu_count,
     int8_t **mem_ptr, void const *input_lut, uint32_t lwe_dimension,
@@ -403,12 +434,13 @@ void scratch_cuda_integer_compute_prefix_sum_hillis_steele_64(
     uint32_t ks_base_log, uint32_t pbs_level, uint32_t pbs_base_log,
     uint32_t grouping_factor, uint32_t num_radix_blocks,
     uint32_t message_modulus, uint32_t carry_modulus, PBS_TYPE pbs_type,
-    bool allocate_gpu_memory);
+    uint64_t lut_degree, bool allocate_gpu_memory);
 
 void cuda_integer_compute_prefix_sum_hillis_steele_64(
     void *const *streams, uint32_t const *gpu_indexes, uint32_t gpu_count,
-    void *output_radix_lwe, void *generates_or_propagates, int8_t *mem_ptr,
-    void *const *ksks, void *const *bsks, uint32_t num_blocks, uint32_t shift);
+    CudaRadixCiphertextFFI *output_radix_lwe,
+    CudaRadixCiphertextFFI *generates_or_propagates, int8_t *mem_ptr,
+    void *const *ksks, void *const *bsks, uint32_t num_blocks);
 
 void cleanup_cuda_integer_compute_prefix_sum_hillis_steele_64(
     void *const *streams, uint32_t const *gpu_indexes, uint32_t gpu_count,
@@ -416,9 +448,8 @@ void cleanup_cuda_integer_compute_prefix_sum_hillis_steele_64(
 
 void cuda_integer_reverse_blocks_64_inplace(void *const *streams,
                                             uint32_t const *gpu_indexes,
-                                            uint32_t gpu_count, void *lwe_array,
-                                            uint32_t num_blocks,
-                                            uint32_t lwe_size);
+                                            uint32_t gpu_count,
+                                            CudaRadixCiphertextFFI *lwe_array);
 
 void scratch_cuda_integer_abs_inplace_radix_ciphertext_kb_64(
     void *const *streams, uint32_t const *gpu_indexes, uint32_t gpu_count,
@@ -431,13 +462,51 @@ void scratch_cuda_integer_abs_inplace_radix_ciphertext_kb_64(
 
 void cuda_integer_abs_inplace_radix_ciphertext_kb_64(
     void *const *streams, uint32_t const *gpu_indexes, uint32_t gpu_count,
-    void *ct, int8_t *mem_ptr, bool is_signed, void *const *bsks,
-    void *const *ksks, uint32_t num_blocks);
+    CudaRadixCiphertextFFI *ct, int8_t *mem_ptr, bool is_signed,
+    void *const *bsks, void *const *ksks);
 
 void cleanup_cuda_integer_abs_inplace(void *const *streams,
                                       uint32_t const *gpu_indexes,
                                       uint32_t gpu_count,
                                       int8_t **mem_ptr_void);
 
+void scratch_cuda_integer_are_all_comparisons_block_true_kb_64(
+    void *const *streams, uint32_t const *gpu_indexes, uint32_t gpu_count,
+    int8_t **mem_ptr, uint32_t glwe_dimension, uint32_t polynomial_size,
+    uint32_t big_lwe_dimension, uint32_t small_lwe_dimension, uint32_t ks_level,
+    uint32_t ks_base_log, uint32_t pbs_level, uint32_t pbs_base_log,
+    uint32_t grouping_factor, uint32_t num_radix_blocks,
+    uint32_t message_modulus, uint32_t carry_modulus, PBS_TYPE pbs_type,
+    bool allocate_gpu_memory);
+
+void cuda_integer_are_all_comparisons_block_true_kb_64(
+    void *const *streams, uint32_t const *gpu_indexes, uint32_t gpu_count,
+    CudaRadixCiphertextFFI *lwe_array_out,
+    CudaRadixCiphertextFFI const *lwe_array_in, int8_t *mem_ptr,
+    void *const *bsks, void *const *ksks, uint32_t num_radix_blocks);
+
+void cleanup_cuda_integer_are_all_comparisons_block_true(
+    void *const *streams, uint32_t const *gpu_indexes, uint32_t gpu_count,
+    int8_t **mem_ptr_void);
+
+void scratch_cuda_integer_is_at_least_one_comparisons_block_true_kb_64(
+    void *const *streams, uint32_t const *gpu_indexes, uint32_t gpu_count,
+    int8_t **mem_ptr, uint32_t glwe_dimension, uint32_t polynomial_size,
+    uint32_t big_lwe_dimension, uint32_t small_lwe_dimension, uint32_t ks_level,
+    uint32_t ks_base_log, uint32_t pbs_level, uint32_t pbs_base_log,
+    uint32_t grouping_factor, uint32_t num_radix_blocks,
+    uint32_t message_modulus, uint32_t carry_modulus, PBS_TYPE pbs_type,
+    bool allocate_gpu_memory);
+
+void cuda_integer_is_at_least_one_comparisons_block_true_kb_64(
+    void *const *streams, uint32_t const *gpu_indexes, uint32_t gpu_count,
+    CudaRadixCiphertextFFI *lwe_array_out,
+    CudaRadixCiphertextFFI const *lwe_array_in, int8_t *mem_ptr,
+    void *const *bsks, void *const *ksks, uint32_t num_radix_blocks);
+
+void cleanup_cuda_integer_is_at_least_one_comparisons_block_true(
+    void *const *streams, uint32_t const *gpu_indexes, uint32_t gpu_count,
+    int8_t **mem_ptr_void);
+
 } // extern C
 #endif // CUDA_INTEGER_H

backends/tfhe-cuda-backend/cuda/include/integer/radix_ciphertext.h

@@ -0,0 +1,11 @@
+#ifndef CUDA_RADIX_CIPHERTEXT_H
+#define CUDA_RADIX_CIPHERTEXT_H
+
+void release_radix_ciphertext_async(cudaStream_t const stream,
+                                    uint32_t const gpu_index,
+                                    CudaRadixCiphertextFFI *data);
+
+void reset_radix_ciphertext_blocks(CudaRadixCiphertextFFI *data,
+                                   uint32_t new_num_blocks);
+
+#endif

backends/tfhe-cuda-backend/cuda/include/keyswitch.h

@@ -21,8 +21,8 @@ void cuda_keyswitch_lwe_ciphertext_vector_64(
 
 void scratch_packing_keyswitch_lwe_list_to_glwe_64(
     void *stream, uint32_t gpu_index, int8_t **fp_ks_buffer,
-    uint32_t glwe_dimension, uint32_t polynomial_size, uint32_t num_lwes,
-    bool allocate_gpu_memory);
+    uint32_t lwe_dimension, uint32_t glwe_dimension, uint32_t polynomial_size,
+    uint32_t num_lwes, bool allocate_gpu_memory);
 
 void cuda_packing_keyswitch_lwe_list_to_glwe_64(
     void *stream, uint32_t gpu_index, void *glwe_array_out,

backends/tfhe-cuda-backend/cuda/include/linear_algebra.h

@@ -1,48 +1,52 @@
 #ifndef CUDA_LINALG_H_
 #define CUDA_LINALG_H_
 
+#include "integer/integer.h"
 #include <stdint.h>
 
 extern "C" {
 
-void cuda_negate_lwe_ciphertext_vector_32(void *stream, uint32_t gpu_index,
-                                          void *lwe_array_out,
-                                          void const *lwe_array_in,
-                                          uint32_t input_lwe_dimension,
-                                          uint32_t input_lwe_ciphertext_count);
-void cuda_negate_lwe_ciphertext_vector_64(void *stream, uint32_t gpu_index,
-                                          void *lwe_array_out,
-                                          void const *lwe_array_in,
-                                          uint32_t input_lwe_dimension,
-                                          uint32_t input_lwe_ciphertext_count);
+void cuda_negate_lwe_ciphertext_vector_32(
+    void *stream, uint32_t gpu_index, void *lwe_array_out,
+    void const *lwe_array_in, const uint32_t input_lwe_dimension,
+    const uint32_t input_lwe_ciphertext_count);
+void cuda_negate_lwe_ciphertext_vector_64(
+    void *stream, uint32_t gpu_index, void *lwe_array_out,
+    void const *lwe_array_in, const uint32_t input_lwe_dimension,
+    const uint32_t input_lwe_ciphertext_count);
 void cuda_add_lwe_ciphertext_vector_32(void *stream, uint32_t gpu_index,
-                                       void *lwe_array_out,
-                                       void const *lwe_array_in_1,
-                                       void const *lwe_array_in_2,
-                                       uint32_t input_lwe_dimension,
-                                       uint32_t input_lwe_ciphertext_count);
+                                       CudaRadixCiphertextFFI *output,
+                                       CudaRadixCiphertextFFI const *input_1,
+                                       CudaRadixCiphertextFFI const *input_2);
 void cuda_add_lwe_ciphertext_vector_64(void *stream, uint32_t gpu_index,
-                                       void *lwe_array_out,
-                                       void const *lwe_array_in_1,
-                                       void const *lwe_array_in_2,
-                                       uint32_t input_lwe_dimension,
-                                       uint32_t input_lwe_ciphertext_count);
+                                       CudaRadixCiphertextFFI *output,
+                                       CudaRadixCiphertextFFI const *input_1,
+                                       CudaRadixCiphertextFFI const *input_2);
 void cuda_add_lwe_ciphertext_vector_plaintext_vector_32(
     void *stream, uint32_t gpu_index, void *lwe_array_out,
     void const *lwe_array_in, void const *plaintext_array_in,
-    uint32_t input_lwe_dimension, uint32_t input_lwe_ciphertext_count);
+    const uint32_t input_lwe_dimension,
+    const uint32_t input_lwe_ciphertext_count);
 void cuda_add_lwe_ciphertext_vector_plaintext_vector_64(
     void *stream, uint32_t gpu_index, void *lwe_array_out,
     void const *lwe_array_in, void const *plaintext_array_in,
-    uint32_t input_lwe_dimension, uint32_t input_lwe_ciphertext_count);
+    const uint32_t input_lwe_dimension,
+    const uint32_t input_lwe_ciphertext_count);
 void cuda_mult_lwe_ciphertext_vector_cleartext_vector_32(
     void *stream, uint32_t gpu_index, void *lwe_array_out,
     void const *lwe_array_in, void const *cleartext_array_in,
-    uint32_t input_lwe_dimension, uint32_t input_lwe_ciphertext_count);
+    const uint32_t input_lwe_dimension,
+    const uint32_t input_lwe_ciphertext_count);
 void cuda_mult_lwe_ciphertext_vector_cleartext_vector_64(
     void *stream, uint32_t gpu_index, void *lwe_array_out,
     void const *lwe_array_in, void const *cleartext_array_in,
-    uint32_t input_lwe_dimension, uint32_t input_lwe_ciphertext_count);
+    const uint32_t input_lwe_dimension,
+    const uint32_t input_lwe_ciphertext_count);
+void cuda_add_lwe_ciphertext_vector_plaintext_64(
+    void *stream, uint32_t gpu_index, void *lwe_array_out,
+    void const *lwe_array_in, const uint64_t plaintext_in,
+    const uint32_t input_lwe_dimension,
+    const uint32_t input_lwe_ciphertext_count);
 }
 
 #endif // CUDA_LINALG_H_

backends/tfhe-cuda-backend/cuda/include/pbs/pbs_multibit_utilities.h

@@ -5,12 +5,12 @@
 
 template <typename Torus>
 bool supports_distributed_shared_memory_on_multibit_programmable_bootstrap(
-    uint32_t polynomial_size);
+    uint32_t polynomial_size, uint32_t max_shared_memory);
 
 template <typename Torus>
 bool has_support_to_cuda_programmable_bootstrap_tbc_multi_bit(
     uint32_t num_samples, uint32_t glwe_dimension, uint32_t polynomial_size,
-    uint32_t level_count);
+    uint32_t level_count, uint32_t max_shared_memory);
 
 #if CUDA_ARCH >= 900
 template <typename Torus>
@@ -28,7 +28,7 @@ void cuda_tbc_multi_bit_programmable_bootstrap_lwe_ciphertext_vector(
     pbs_buffer<Torus, MULTI_BIT> *pbs_buffer, uint32_t lwe_dimension,
     uint32_t glwe_dimension, uint32_t polynomial_size, uint32_t grouping_factor,
     uint32_t base_log, uint32_t level_count, uint32_t num_samples,
-    uint32_t lut_count, uint32_t lut_stride);
+    uint32_t num_many_lut, uint32_t lut_stride);
 #endif
 
 template <typename Torus>
@@ -46,7 +46,7 @@ void cuda_cg_multi_bit_programmable_bootstrap_lwe_ciphertext_vector(
     pbs_buffer<Torus, MULTI_BIT> *pbs_buffer, uint32_t lwe_dimension,
     uint32_t glwe_dimension, uint32_t polynomial_size, uint32_t grouping_factor,
     uint32_t base_log, uint32_t level_count, uint32_t num_samples,
-    uint32_t lut_count, uint32_t lut_stride);
+    uint32_t num_many_lut, uint32_t lut_stride);
 
 template <typename Torus>
 void scratch_cuda_multi_bit_programmable_bootstrap(
@@ -63,7 +63,7 @@ void cuda_multi_bit_programmable_bootstrap_lwe_ciphertext_vector(
     pbs_buffer<Torus, MULTI_BIT> *pbs_buffer, uint32_t lwe_dimension,
     uint32_t glwe_dimension, uint32_t polynomial_size, uint32_t grouping_factor,
     uint32_t base_log, uint32_t level_count, uint32_t num_samples,
-    uint32_t lut_count, uint32_t lut_stride);
+    uint32_t num_many_lut, uint32_t lut_stride);
 
 template <typename Torus>
 uint64_t get_buffer_size_full_sm_multibit_programmable_bootstrap_keybundle(
@@ -114,6 +114,8 @@ template <typename Torus> struct pbs_buffer<Torus, PBS_TYPE::MULTI_BIT> {
              uint32_t polynomial_size, uint32_t level_count,
              uint32_t input_lwe_ciphertext_count, uint32_t lwe_chunk_size,
              PBS_VARIANT pbs_variant, bool allocate_gpu_memory) {
+    cuda_set_device(gpu_index);
+
     this->pbs_variant = pbs_variant;
     this->lwe_chunk_size = lwe_chunk_size;
     auto max_shared_memory = cuda_get_max_shared_memory(gpu_index);

backends/tfhe-cuda-backend/cuda/include/pbs/pbs_utilities.h

@@ -9,20 +9,26 @@
 template <typename Torus>
 uint64_t get_buffer_size_full_sm_programmable_bootstrap_step_one(
     uint32_t polynomial_size) {
-  return sizeof(Torus) * polynomial_size +      // accumulator_rotated
-         sizeof(double2) * polynomial_size / 2; // accumulator fft
+  size_t double_count = (sizeof(Torus) == 16) ? 2 : 1;
+  return sizeof(Torus) * polynomial_size + // accumulator_rotated
+         sizeof(double) * 2 * double_count * polynomial_size /
+             2; // accumulator fft
 }
 template <typename Torus>
 uint64_t get_buffer_size_full_sm_programmable_bootstrap_step_two(
     uint32_t polynomial_size) {
-  return sizeof(Torus) * polynomial_size +      // accumulator
-         sizeof(double2) * polynomial_size / 2; // accumulator fft
+  size_t double_count = (sizeof(Torus) == 16) ? 2 : 1;
+  return sizeof(Torus) * polynomial_size + // accumulator
+         sizeof(double) * 2 * double_count * polynomial_size /
+             2; // accumulator fft
 }
 
 template <typename Torus>
 uint64_t
 get_buffer_size_partial_sm_programmable_bootstrap(uint32_t polynomial_size) {
-  return sizeof(double2) * polynomial_size / 2; // accumulator fft
+  size_t double_count = (sizeof(Torus) == 16) ? 2 : 1;
+  return sizeof(double) * 2 * double_count * polynomial_size /
+         2; // accumulator fft
 }
 
 template <typename Torus>
@@ -61,7 +67,7 @@ get_buffer_size_partial_sm_programmable_bootstrap_cg(uint32_t polynomial_size) {
 
 template <typename Torus>
 bool supports_distributed_shared_memory_on_classic_programmable_bootstrap(
-    uint32_t polynomial_size);
+    uint32_t polynomial_size, uint32_t max_shared_memory);
 
 template <typename Torus, PBS_TYPE pbs_type> struct pbs_buffer;
 
@@ -77,10 +83,10 @@ template <typename Torus> struct pbs_buffer<Torus, PBS_TYPE::CLASSICAL> {
              uint32_t polynomial_size, uint32_t level_count,
              uint32_t input_lwe_ciphertext_count, PBS_VARIANT pbs_variant,
              bool allocate_gpu_memory) {
-
+    cuda_set_device(gpu_index);
     this->pbs_variant = pbs_variant;
 
-    auto max_shared_memory = cuda_get_max_shared_memory(0);
+    auto max_shared_memory = cuda_get_max_shared_memory(gpu_index);
 
     if (allocate_gpu_memory) {
       switch (pbs_variant) {
@@ -157,7 +163,7 @@ template <typename Torus> struct pbs_buffer<Torus, PBS_TYPE::CLASSICAL> {
 
         bool supports_dsm =
             supports_distributed_shared_memory_on_classic_programmable_bootstrap<
-                Torus>(polynomial_size);
+                Torus>(polynomial_size, max_shared_memory);
 
         uint64_t full_sm =
             get_buffer_size_full_sm_programmable_bootstrap_tbc<Torus>(
@@ -215,11 +221,159 @@ template <typename Torus> struct pbs_buffer<Torus, PBS_TYPE::CLASSICAL> {
   }
 };
 
+template <PBS_TYPE pbs_type> struct pbs_buffer_128;
+
+template <> struct pbs_buffer_128<PBS_TYPE::CLASSICAL> {
+  int8_t *d_mem;
+
+  __uint128_t *global_accumulator;
+  double *global_join_buffer;
+
+  PBS_VARIANT pbs_variant;
+
+  pbs_buffer_128(cudaStream_t stream, uint32_t gpu_index,
+                 uint32_t glwe_dimension, uint32_t polynomial_size,
+                 uint32_t level_count, uint32_t input_lwe_ciphertext_count,
+                 PBS_VARIANT pbs_variant, bool allocate_gpu_memory) {
+    cuda_set_device(gpu_index);
+    this->pbs_variant = pbs_variant;
+
+    auto max_shared_memory = cuda_get_max_shared_memory(gpu_index);
+    size_t global_join_buffer_size = (glwe_dimension + 1) * level_count *
+                                     input_lwe_ciphertext_count *
+                                     polynomial_size / 2 * sizeof(double) * 4;
+
+    if (allocate_gpu_memory) {
+      switch (pbs_variant) {
+      case PBS_VARIANT::DEFAULT: {
+        uint64_t full_sm_step_one =
+            get_buffer_size_full_sm_programmable_bootstrap_step_one<
+                __uint128_t>(polynomial_size);
+        uint64_t full_sm_step_two =
+            get_buffer_size_full_sm_programmable_bootstrap_step_two<
+                __uint128_t>(polynomial_size);
+        uint64_t partial_sm =
+            get_buffer_size_partial_sm_programmable_bootstrap<__uint128_t>(
+                polynomial_size);
+
+        uint64_t partial_dm_step_one = full_sm_step_one - partial_sm;
+        uint64_t partial_dm_step_two = full_sm_step_two - partial_sm;
+        uint64_t full_dm = full_sm_step_one;
+
+        uint64_t device_mem = 0;
+        if (max_shared_memory < partial_sm) {
+          device_mem = full_dm * input_lwe_ciphertext_count * level_count *
+                       (glwe_dimension + 1);
+        } else if (max_shared_memory < full_sm_step_two) {
+          device_mem =
+              (partial_dm_step_two + partial_dm_step_one * level_count) *
+              input_lwe_ciphertext_count * (glwe_dimension + 1);
+        } else if (max_shared_memory < full_sm_step_one) {
+          device_mem = partial_dm_step_one * input_lwe_ciphertext_count *
+                       level_count * (glwe_dimension + 1);
+        }
+        // Otherwise, both kernels run all in shared memory
+        d_mem = (int8_t *)cuda_malloc_async(device_mem, stream, gpu_index);
+
+        global_join_buffer = (double *)cuda_malloc_async(
+            global_join_buffer_size, stream, gpu_index);
+
+        global_accumulator = (__uint128_t *)cuda_malloc_async(
+            (glwe_dimension + 1) * input_lwe_ciphertext_count *
+                polynomial_size * sizeof(__uint128_t),
+            stream, gpu_index);
+      } break;
+      case PBS_VARIANT::CG: {
+        uint64_t full_sm =
+            get_buffer_size_full_sm_programmable_bootstrap_cg<__uint128_t>(
+                polynomial_size);
+        uint64_t partial_sm =
+            get_buffer_size_partial_sm_programmable_bootstrap_cg<__uint128_t>(
+                polynomial_size);
+
+        uint64_t partial_dm = full_sm - partial_sm;
+        uint64_t full_dm = full_sm;
+        uint64_t device_mem = 0;
+
+        if (max_shared_memory < partial_sm) {
+          device_mem = full_dm * input_lwe_ciphertext_count * level_count *
+                       (glwe_dimension + 1);
+        } else if (max_shared_memory < full_sm) {
+          device_mem = partial_dm * input_lwe_ciphertext_count * level_count *
+                       (glwe_dimension + 1);
+        }
+
+        // Otherwise, both kernels run all in shared memory
+        d_mem = (int8_t *)cuda_malloc_async(device_mem, stream, gpu_index);
+
+        global_join_buffer = (double *)cuda_malloc_async(
+            global_join_buffer_size, stream, gpu_index);
+      } break;
+#if CUDA_ARCH >= 900
+      case PBS_VARIANT::TBC: {
+
+        bool supports_dsm =
+            supports_distributed_shared_memory_on_classic_programmable_bootstrap<
+                __uint128_t>(polynomial_size, max_shared_memory);
+
+        uint64_t full_sm =
+            get_buffer_size_full_sm_programmable_bootstrap_tbc<__uint128_t>(
+                polynomial_size);
+        uint64_t partial_sm =
+            get_buffer_size_partial_sm_programmable_bootstrap_tbc<__uint128_t>(
+                polynomial_size);
+        uint64_t minimum_sm_tbc = 0;
+        if (supports_dsm)
+          minimum_sm_tbc =
+              get_buffer_size_sm_dsm_plus_tbc_classic_programmable_bootstrap<
+                  __uint128_t>(polynomial_size);
+
+        uint64_t partial_dm = full_sm - partial_sm;
+        uint64_t full_dm = full_sm;
+        uint64_t device_mem = 0;
+
+        // There is a minimum amount of memory we need to run the TBC PBS, which
+        // is minimum_sm_tbc. We know that minimum_sm_tbc bytes are available
+        // because otherwise the previous check would have redirected
+        // computation to some other variant. If over that we don't have more
+        // partial_sm bytes, TBC PBS will run on NOSM. If we have partial_sm but
+        // not full_sm bytes, it will run on PARTIALSM. Otherwise, FULLSM.
+        //
+        // NOSM mode actually requires minimum_sm_tbc shared memory bytes.
+        if (max_shared_memory < partial_sm + minimum_sm_tbc) {
+          device_mem = full_dm * input_lwe_ciphertext_count * level_count *
+                       (glwe_dimension + 1);
+        } else if (max_shared_memory < full_sm + minimum_sm_tbc) {
+          device_mem = partial_dm * input_lwe_ciphertext_count * level_count *
+                       (glwe_dimension + 1);
+        }
+
+        // Otherwise, both kernels run all in shared memory
+        d_mem = (int8_t *)cuda_malloc_async(device_mem, stream, gpu_index);
+
+        global_join_buffer = (double *)cuda_malloc_async(
+            global_join_buffer_size, stream, gpu_index);
+      } break;
+#endif
+      default:
+        PANIC("Cuda error (PBS): unsupported implementation variant.")
+      }
+    }
+  }
+
+  void release(cudaStream_t stream, uint32_t gpu_index) {
+    cuda_drop_async(d_mem, stream, gpu_index);
+    cuda_drop_async(global_join_buffer, stream, gpu_index);
+
+    if (pbs_variant == DEFAULT)
+      cuda_drop_async(global_accumulator, stream, gpu_index);
+  }
+};
+
 template <typename Torus>
 uint64_t get_buffer_size_programmable_bootstrap_cg(
     uint32_t glwe_dimension, uint32_t polynomial_size, uint32_t level_count,
-    uint32_t input_lwe_ciphertext_count) {
-  int max_shared_memory = cuda_get_max_shared_memory(0);
+    uint32_t input_lwe_ciphertext_count, uint32_t max_shared_memory) {
   uint64_t full_sm =
       get_buffer_size_full_sm_programmable_bootstrap_cg<Torus>(polynomial_size);
   uint64_t partial_sm =
@@ -245,7 +399,8 @@ template <typename Torus>
 bool has_support_to_cuda_programmable_bootstrap_cg(uint32_t glwe_dimension,
                                                    uint32_t polynomial_size,
                                                    uint32_t level_count,
-                                                   uint32_t num_samples);
+                                                   uint32_t num_samples,
+                                                   uint32_t max_shared_memory);
 
 template <typename Torus>
 void cuda_programmable_bootstrap_cg_lwe_ciphertext_vector(
@@ -255,7 +410,7 @@ void cuda_programmable_bootstrap_cg_lwe_ciphertext_vector(
     Torus const *lwe_input_indexes, double2 const *bootstrapping_key,
     pbs_buffer<Torus, CLASSICAL> *buffer, uint32_t lwe_dimension,
     uint32_t glwe_dimension, uint32_t polynomial_size, uint32_t base_log,
-    uint32_t level_count, uint32_t num_samples, uint32_t lut_count,
+    uint32_t level_count, uint32_t num_samples, uint32_t num_many_lut,
     uint32_t lut_stride);
 
 template <typename Torus>
@@ -266,7 +421,7 @@ void cuda_programmable_bootstrap_lwe_ciphertext_vector(
     Torus const *lwe_input_indexes, double2 const *bootstrapping_key,
     pbs_buffer<Torus, CLASSICAL> *buffer, uint32_t lwe_dimension,
     uint32_t glwe_dimension, uint32_t polynomial_size, uint32_t base_log,
-    uint32_t level_count, uint32_t num_samples, uint32_t lut_count,
+    uint32_t level_count, uint32_t num_samples, uint32_t num_many_lut,
     uint32_t lut_stride);
 
 #if (CUDA_ARCH >= 900)
@@ -278,7 +433,7 @@ void cuda_programmable_bootstrap_tbc_lwe_ciphertext_vector(
     Torus const *lwe_input_indexes, double2 const *bootstrapping_key,
     pbs_buffer<Torus, CLASSICAL> *buffer, uint32_t lwe_dimension,
     uint32_t glwe_dimension, uint32_t polynomial_size, uint32_t base_log,
-    uint32_t level_count, uint32_t num_samples, uint32_t lut_count,
+    uint32_t level_count, uint32_t num_samples, uint32_t num_many_lut,
     uint32_t lut_stride);
 
 template <typename Torus>

backends/tfhe-cuda-backend/cuda/include/pbs/programmable_bootstrap.h

@@ -20,6 +20,11 @@ void cuda_convert_lwe_programmable_bootstrap_key_64(
     uint32_t input_lwe_dim, uint32_t glwe_dim, uint32_t level_count,
     uint32_t polynomial_size);
 
+void cuda_convert_lwe_programmable_bootstrap_key_128(
+    void *stream, uint32_t gpu_index, void *dest, void const *src,
+    uint32_t input_lwe_dim, uint32_t glwe_dim, uint32_t level_count,
+    uint32_t polynomial_size);
+
 void scratch_cuda_programmable_bootstrap_amortized_32(
     void *stream, uint32_t gpu_index, int8_t **pbs_buffer,
     uint32_t glwe_dimension, uint32_t polynomial_size,
@@ -62,6 +67,11 @@ void scratch_cuda_programmable_bootstrap_64(
     uint32_t polynomial_size, uint32_t level_count,
     uint32_t input_lwe_ciphertext_count, bool allocate_gpu_memory);
 
+void scratch_cuda_programmable_bootstrap_128(
+    void *stream, uint32_t gpu_index, int8_t **buffer, uint32_t glwe_dimension,
+    uint32_t polynomial_size, uint32_t level_count,
+    uint32_t input_lwe_ciphertext_count, bool allocate_gpu_memory);
+
 void cuda_programmable_bootstrap_lwe_ciphertext_vector_32(
     void *stream, uint32_t gpu_index, void *lwe_array_out,
     void const *lwe_output_indexes, void const *lut_vector,
@@ -69,7 +79,7 @@ void cuda_programmable_bootstrap_lwe_ciphertext_vector_32(
     void const *lwe_input_indexes, void const *bootstrapping_key,
     int8_t *buffer, uint32_t lwe_dimension, uint32_t glwe_dimension,
     uint32_t polynomial_size, uint32_t base_log, uint32_t level_count,
-    uint32_t num_samples, uint32_t lut_count, uint32_t lut_stride);
+    uint32_t num_samples, uint32_t num_many_lut, uint32_t lut_stride);
 
 void cuda_programmable_bootstrap_lwe_ciphertext_vector_64(
     void *stream, uint32_t gpu_index, void *lwe_array_out,
@@ -78,9 +88,19 @@ void cuda_programmable_bootstrap_lwe_ciphertext_vector_64(
     void const *lwe_input_indexes, void const *bootstrapping_key,
     int8_t *buffer, uint32_t lwe_dimension, uint32_t glwe_dimension,
     uint32_t polynomial_size, uint32_t base_log, uint32_t level_count,
-    uint32_t num_samples, uint32_t lut_count, uint32_t lut_stride);
+    uint32_t num_samples, uint32_t num_many_lut, uint32_t lut_stride);
+
+void cuda_programmable_bootstrap_lwe_ciphertext_vector_128(
+    void *stream, uint32_t gpu_index, void *lwe_array_out,
+    void const *lut_vector, void const *lwe_array_in,
+    void const *bootstrapping_key, int8_t *buffer, uint32_t lwe_dimension,
+    uint32_t glwe_dimension, uint32_t polynomial_size, uint32_t base_log,
+    uint32_t level_count, uint32_t num_samples);
 
 void cleanup_cuda_programmable_bootstrap(void *stream, uint32_t gpu_index,
                                          int8_t **pbs_buffer);
+
+void cleanup_cuda_programmable_bootstrap_128(void *stream, uint32_t gpu_index,
+                                             int8_t **pbs_buffer);
 }
 #endif // CUDA_BOOTSTRAP_H

backends/tfhe-cuda-backend/cuda/include/pbs/programmable_bootstrap_multibit.h

@@ -8,7 +8,7 @@ extern "C" {
 
 bool has_support_to_cuda_programmable_bootstrap_cg_multi_bit(
     uint32_t glwe_dimension, uint32_t polynomial_size, uint32_t level_count,
-    uint32_t num_samples);
+    uint32_t num_samples, uint32_t max_shared_memory);
 
 void cuda_convert_lwe_multi_bit_programmable_bootstrap_key_64(
     void *stream, uint32_t gpu_index, void *dest, void const *src,
@@ -27,7 +27,7 @@ void cuda_multi_bit_programmable_bootstrap_lwe_ciphertext_vector_64(
     void const *lwe_input_indexes, void const *bootstrapping_key,
     int8_t *buffer, uint32_t lwe_dimension, uint32_t glwe_dimension,
     uint32_t polynomial_size, uint32_t grouping_factor, uint32_t base_log,
-    uint32_t level_count, uint32_t num_samples, uint32_t lut_count,
+    uint32_t level_count, uint32_t num_samples, uint32_t num_many_lut,
     uint32_t lut_stride);
 
 void cleanup_cuda_multi_bit_programmable_bootstrap(void *stream,

backends/tfhe-cuda-backend/cuda/src/crypto/ciphertext.cu

@@ -24,7 +24,7 @@ void cuda_convert_lwe_ciphertext_vector_to_cpu_64(void *stream,
 void cuda_glwe_sample_extract_64(void *stream, uint32_t gpu_index,
                                  void *lwe_array_out, void const *glwe_array_in,
                                  uint32_t const *nth_array, uint32_t num_nths,
-                                 uint32_t glwe_dimension,
+                                 uint32_t lwe_per_glwe, uint32_t glwe_dimension,
                                  uint32_t polynomial_size) {
 
   switch (polynomial_size) {
@@ -32,43 +32,43 @@ void cuda_glwe_sample_extract_64(void *stream, uint32_t gpu_index,
     host_sample_extract<uint64_t, AmortizedDegree<256>>(
         static_cast<cudaStream_t>(stream), gpu_index, (uint64_t *)lwe_array_out,
         (uint64_t const *)glwe_array_in, (uint32_t const *)nth_array, num_nths,
-        glwe_dimension);
+        lwe_per_glwe, glwe_dimension);
     break;
   case 512:
     host_sample_extract<uint64_t, AmortizedDegree<512>>(
         static_cast<cudaStream_t>(stream), gpu_index, (uint64_t *)lwe_array_out,
         (uint64_t const *)glwe_array_in, (uint32_t const *)nth_array, num_nths,
-        glwe_dimension);
+        lwe_per_glwe, glwe_dimension);
     break;
   case 1024:
     host_sample_extract<uint64_t, AmortizedDegree<1024>>(
         static_cast<cudaStream_t>(stream), gpu_index, (uint64_t *)lwe_array_out,
         (uint64_t const *)glwe_array_in, (uint32_t const *)nth_array, num_nths,
-        glwe_dimension);
+        lwe_per_glwe, glwe_dimension);
     break;
   case 2048:
     host_sample_extract<uint64_t, AmortizedDegree<2048>>(
         static_cast<cudaStream_t>(stream), gpu_index, (uint64_t *)lwe_array_out,
         (uint64_t const *)glwe_array_in, (uint32_t const *)nth_array, num_nths,
-        glwe_dimension);
+        lwe_per_glwe, glwe_dimension);
     break;
   case 4096:
     host_sample_extract<uint64_t, AmortizedDegree<4096>>(
         static_cast<cudaStream_t>(stream), gpu_index, (uint64_t *)lwe_array_out,
         (uint64_t const *)glwe_array_in, (uint32_t const *)nth_array, num_nths,
-        glwe_dimension);
+        lwe_per_glwe, glwe_dimension);
     break;
   case 8192:
     host_sample_extract<uint64_t, AmortizedDegree<8192>>(
         static_cast<cudaStream_t>(stream), gpu_index, (uint64_t *)lwe_array_out,
         (uint64_t const *)glwe_array_in, (uint32_t const *)nth_array, num_nths,
-        glwe_dimension);
+        lwe_per_glwe, glwe_dimension);
     break;
   case 16384:
     host_sample_extract<uint64_t, AmortizedDegree<16384>>(
         static_cast<cudaStream_t>(stream), gpu_index, (uint64_t *)lwe_array_out,
         (uint64_t const *)glwe_array_in, (uint32_t const *)nth_array, num_nths,
-        glwe_dimension);
+        lwe_per_glwe, glwe_dimension);
     break;
   default:
     PANIC("Cuda error: unsupported polynomial size. Supported "

backends/tfhe-cuda-backend/cuda/src/crypto/ciphertext.cuh

@@ -11,7 +11,7 @@ void cuda_convert_lwe_ciphertext_vector_to_gpu(cudaStream_t stream,
                                                uint32_t gpu_index, T *dest,
                                                T *src, uint32_t number_of_cts,
                                                uint32_t lwe_dimension) {
-  cudaSetDevice(gpu_index);
+  cuda_set_device(gpu_index);
   uint64_t size = number_of_cts * (lwe_dimension + 1) * sizeof(T);
   cuda_memcpy_async_to_gpu(dest, src, size, stream, gpu_index);
 }
@@ -21,14 +21,14 @@ void cuda_convert_lwe_ciphertext_vector_to_cpu(cudaStream_t stream,
                                                uint32_t gpu_index, T *dest,
                                                T *src, uint32_t number_of_cts,
                                                uint32_t lwe_dimension) {
-  cudaSetDevice(gpu_index);
+  cuda_set_device(gpu_index);
   uint64_t size = number_of_cts * (lwe_dimension + 1) * sizeof(T);
   cuda_memcpy_async_to_cpu(dest, src, size, stream, gpu_index);
 }
 
 template <typename Torus, class params>
 __global__ void sample_extract(Torus *lwe_array_out, Torus const *glwe_array_in,
-                               uint32_t const *nth_array,
+                               uint32_t const *nth_array, uint32_t lwe_per_glwe,
                                uint32_t glwe_dimension) {
 
   const int input_id = blockIdx.x;
@@ -39,28 +39,28 @@ __global__ void sample_extract(Torus *lwe_array_out, Torus const *glwe_array_in,
   auto lwe_out = lwe_array_out + input_id * lwe_output_size;
 
   // We assume each GLWE will store the first polynomial_size inputs
-  uint32_t lwe_per_glwe = params::degree;
   auto glwe_in = glwe_array_in + (input_id / lwe_per_glwe) * glwe_input_size;
 
-  // nth is ensured to be in [0, lwe_per_glwe)
-  auto nth = nth_array[input_id] % lwe_per_glwe;
+  // nth is ensured to be in [0, params::degree)
+  auto nth = nth_array[input_id] % params::degree;
 
   sample_extract_mask<Torus, params>(lwe_out, glwe_in, glwe_dimension, nth);
   sample_extract_body<Torus, params>(lwe_out, glwe_in, glwe_dimension, nth);
 }
 
+// lwe_per_glwe LWEs will be extracted per GLWE ciphertext, thus we need to have
+// enough indexes
 template <typename Torus, class params>
-__host__ void host_sample_extract(cudaStream_t stream, uint32_t gpu_index,
-                                  Torus *lwe_array_out,
-                                  Torus const *glwe_array_in,
-                                  uint32_t const *nth_array, uint32_t num_nths,
-                                  uint32_t glwe_dimension) {
-  cudaSetDevice(gpu_index);
-
+__host__ void
+host_sample_extract(cudaStream_t stream, uint32_t gpu_index,
+                    Torus *lwe_array_out, Torus const *glwe_array_in,
+                    uint32_t const *nth_array, uint32_t num_nths,
+                    uint32_t lwe_per_glwe, uint32_t glwe_dimension) {
+  cuda_set_device(gpu_index);
   dim3 grid(num_nths);
   dim3 thds(params::degree / params::opt);
   sample_extract<Torus, params><<<grid, thds, 0, stream>>>(
-      lwe_array_out, glwe_array_in, nth_array, glwe_dimension);
+      lwe_array_out, glwe_array_in, nth_array, lwe_per_glwe, glwe_dimension);
   check_cuda_error(cudaGetLastError());
 }
 

backends/tfhe-cuda-backend/cuda/src/crypto/fast_packing_keyswitch.cuh

@@ -0,0 +1,355 @@
+#ifndef CNCRT_FAST_KS_CUH
+#define CNCRT_FAST_KS_CUH
+
+#undef NDEBUG
+
+#include "device.h"
+#include "gadget.cuh"
+#include "helper_multi_gpu.h"
+#include "keyswitch.cuh"
+#include "polynomial/functions.cuh"
+#include "polynomial/polynomial_math.cuh"
+#include "torus.cuh"
+#include "utils/helper.cuh"
+#include "utils/kernel_dimensions.cuh"
+#include <thread>
+#include <vector>
+
+#define CEIL_DIV(M, N) ((M) + (N)-1) / (N)
+
+const int BLOCK_SIZE_GEMM = 64;
+const int THREADS_GEMM = 8;
+const int BLOCK_SIZE_DECOMP = 8;
+
+template <typename Torus> uint64_t get_shared_mem_size_tgemm() {
+  return BLOCK_SIZE_GEMM * THREADS_GEMM * 2 * sizeof(Torus);
+}
+
+// Initialize decomposition by performing rounding
+// and decomposing one level of an array of Torus LWEs. Only
+// decomposes the mask elements of the incoming LWEs.
+template <typename Torus, typename TorusVec>
+__global__ void decompose_vectorize_init(Torus const *lwe_in, Torus *lwe_out,
+                                         uint32_t lwe_dimension,
+                                         uint32_t num_lwe, uint32_t base_log,
+                                         uint32_t level_count) {
+
+  // index of this LWE ct in the buffer
+  auto lwe_idx = blockIdx.x * blockDim.x + threadIdx.x;
+  // index of the LWE sample in the LWE ct
+  auto lwe_sample_idx = blockIdx.y * blockDim.y + threadIdx.y;
+
+  if (lwe_idx >= num_lwe || lwe_sample_idx >= lwe_dimension)
+    return;
+
+  // Input LWE array is [mask_0, .., mask_lwe_dim, message] and
+  // we only decompose the mask. Thus the stride for reading
+  // is lwe_dimension + 1, while for writing it is lwe_dimension
+  auto read_val_idx = lwe_idx * (lwe_dimension + 1) + lwe_sample_idx;
+  auto write_val_idx = lwe_idx * lwe_dimension + lwe_sample_idx;
+  auto write_state_idx =
+      num_lwe * lwe_dimension + lwe_idx * lwe_dimension + lwe_sample_idx;
+
+  Torus a_i = lwe_in[read_val_idx];
+
+  Torus state = init_decomposer_state(a_i, base_log, level_count);
+
+  Torus mod_b_mask = (1ll << base_log) - 1ll;
+  lwe_out[write_val_idx] = decompose_one<Torus>(state, mod_b_mask, base_log);
+  synchronize_threads_in_block();
+  lwe_out[write_state_idx] = state;
+}
+
+// Continue decomposiion of an array of Torus elements in place. Supposes
+// that the array contains already decomposed elements and
+// computes the new decomposed level in place.
+template <typename Torus, typename TorusVec>
+__global__ void
+decompose_vectorize_step_inplace(Torus *buffer_in, uint32_t lwe_dimension,
+                                 uint32_t num_lwe, uint32_t base_log,
+                                 uint32_t level_count) {
+
+  // index of this LWE ct in the buffer
+  auto lwe_idx = blockIdx.x * blockDim.x + threadIdx.x;
+  // index of the LWE sample in the LWE ct
+  auto lwe_sample_idx = blockIdx.y * blockDim.y + threadIdx.y;
+
+  if (lwe_idx >= num_lwe || lwe_sample_idx >= lwe_dimension)
+    return;
+
+  auto val_idx = lwe_idx * lwe_dimension + lwe_sample_idx;
+  auto state_idx = num_lwe * lwe_dimension + val_idx;
+
+  Torus state = buffer_in[state_idx];
+  synchronize_threads_in_block();
+
+  Torus mod_b_mask = (1ll << base_log) - 1ll;
+
+  buffer_in[val_idx] = decompose_one<Torus>(state, mod_b_mask, base_log);
+  synchronize_threads_in_block();
+  buffer_in[state_idx] = state;
+}
+
+// Multiply matrices A, B of size (M, K), (K, N) respectively
+// with K as the inner dimension.
+//
+// A block of threads processeds blocks of size (BLOCK_SIZE_GEMM,
+// BLOCK_SIZE_GEMM) splitting them in multiple tiles: (BLOCK_SIZE_GEMM,
+// THREADS_GEMM)-shaped tiles of values from A, and a (THREADS_GEMM,
+// BLOCK_SIZE_GEMM)-shaped tiles of values from B.
+//
+// This code is adapted by generalizing the 1d block-tiling
+// kernel from https://github.com/siboehm/SGEMM_CUDA
+// to any matrix dimension
+template <typename Torus, typename TorusVec>
+__global__ void tgemm(int M, int N, int K, const Torus *A, const Torus *B,
+                      int stride_B, Torus *C) {
+
+  const int BM = BLOCK_SIZE_GEMM;
+  const int BN = BLOCK_SIZE_GEMM;
+  const int BK = THREADS_GEMM;
+  const int TM = THREADS_GEMM;
+
+  const uint cRow = blockIdx.y;
+  const uint cCol = blockIdx.x;
+
+  const int threadCol = threadIdx.x % BN;
+  const int threadRow = threadIdx.x / BN;
+
+  // Allocate space for the current block tile in shared memory
+  __shared__ Torus As[BM * BK];
+  __shared__ Torus Bs[BK * BN];
+
+  // Initialize the pointers to the input blocks from A, B
+  // Tiles from these blocks are loaded to shared memory
+  A += cRow * BM * K;
+  B += cCol * BN;
+
+  // Each thread will handle multiple sub-blocks
+  const uint innerColA = threadIdx.x % BK;
+  const uint innerRowA = threadIdx.x / BK;
+  const uint innerColB = threadIdx.x % BN;
+  const uint innerRowB = threadIdx.x / BN;
+
+  // allocate thread-local cache for results in registerfile
+  Torus threadResults[TM] = {0};
+
+  auto row_A = cRow * BM + innerRowA;
+  auto col_B = cCol * BN + innerColB;
+
+  // For each thread, loop over block tiles
+  for (uint bkIdx = 0; bkIdx < K; bkIdx += BK) {
+    auto col_A = bkIdx + innerColA;
+    auto row_B = bkIdx + innerRowB;
+
+    if (row_A < M && col_A < K) {
+      As[innerRowA * BK + innerColA] = A[innerRowA * K + innerColA];
+    } else {
+      As[innerRowA * BK + innerColA] = 0;
+    }
+
+    if (col_B < N && row_B < K) {
+      Bs[innerRowB * BN + innerColB] = B[innerRowB * stride_B + innerColB];
+    } else {
+      Bs[innerRowB * BN + innerColB] = 0;
+    }
+    synchronize_threads_in_block();
+
+    // Advance blocktile for the next iteration of this loop
+    A += BK;
+    B += BK * stride_B;
+
+    // calculate per-thread results
+    for (uint dotIdx = 0; dotIdx < BK; ++dotIdx) {
+      // we make the dotproduct loop the outside loop, which facilitates
+      // reuse of the Bs entry, which we can cache in a tmp var.
+      Torus tmp = Bs[dotIdx * BN + threadCol];
+      for (uint resIdx = 0; resIdx < TM; ++resIdx) {
+        threadResults[resIdx] +=
+            As[(threadRow * TM + resIdx) * BK + dotIdx] * tmp;
+      }
+    }
+    synchronize_threads_in_block();
+  }
+
+  // Initialize the pointer to the output block of size (BLOCK_SIZE_GEMM,
+  // BLOCK_SIZE_GEMM)
+  C += cRow * BM * N + cCol * BN;
+
+  // write out the results
+  for (uint resIdx = 0; resIdx < TM; ++resIdx) {
+    int outRow = cRow * BM + threadRow * TM + resIdx;
+    int outCol = cCol * BN + threadCol;
+
+    if (outRow >= M)
+      continue;
+    if (outCol >= N)
+      continue;
+
+    C[(threadRow * TM + resIdx) * N + threadCol] += threadResults[resIdx];
+  }
+}
+
+// Finish the keyswitching operation and prepare GLWEs for accumulation.
+// 1. Finish the keyswitching computation partially performed with a GEMM:
+//  - negate the dot product between the GLWE and KSK polynomial
+//  - add the GLWE message for the N-th polynomial coeff in the message poly
+// 2. Rotate each of the GLWE . KSK poly dot products to
+//    prepare them for accumulation into a single GLWE
+template <typename Torus>
+__global__ void polynomial_accumulate_monic_monomial_mul_many_neg_and_add_C(
+    Torus *in_glwe_buffer, Torus *out_glwe_buffer, Torus const *lwe_array,
+    uint32_t lwe_dimension, uint32_t num_glwes, uint32_t polynomial_size,
+    uint32_t glwe_dimension) {
+
+  uint32_t glwe_id = blockIdx.x * blockDim.x + threadIdx.x;
+  uint32_t degree = glwe_id; // lwe 0 rotate 0, lwe 1 rotate 1, .. , lwe
+                             // poly_size-1 rotate poly_size-1
+  uint32_t coeffIdx = blockIdx.y * blockDim.y + threadIdx.y;
+
+  if (glwe_id >= num_glwes)
+    return;
+  if (coeffIdx >= polynomial_size)
+    return;
+
+  auto in_poly =
+      in_glwe_buffer + glwe_id * polynomial_size * (glwe_dimension + 1);
+  auto out_result =
+      out_glwe_buffer + glwe_id * polynomial_size * (glwe_dimension + 1);
+  if (coeffIdx == 0) {
+    // Add the message value of the input LWE (`C`) to the N-th coefficient
+    // in the GLWE . KSK dot product
+
+    // The C is added to the first position of the last polynomial in the GLWE
+    // which has (glwe_dimension+1) polynomials
+    // The C value is extracted as the last value of the LWE ct. (of index
+    // glwe_id) the LWEs have (polynomial_size + 1) values
+    in_poly[polynomial_size * glwe_dimension] =
+        lwe_array[glwe_id * (lwe_dimension + 1) + lwe_dimension] -
+        in_poly[polynomial_size * glwe_dimension];
+
+    for (int gi = 1; gi < glwe_dimension; ++gi)
+      in_poly[coeffIdx + gi * polynomial_size] =
+          -in_poly[coeffIdx + gi * polynomial_size];
+
+  } else {
+    // Otherwise simply negate the input coefficient
+    for (int gi = 1; gi < glwe_dimension + 1; ++gi)
+      in_poly[coeffIdx + gi * polynomial_size] =
+          -in_poly[coeffIdx + gi * polynomial_size];
+  }
+  // Negate all the coefficients for rotation for the first poly
+  in_poly[coeffIdx] = -in_poly[coeffIdx];
+
+  // rotate the body
+  polynomial_accumulate_monic_monomial_mul<Torus>(
+      out_result, in_poly, degree, coeffIdx, polynomial_size, 1, true);
+  // rotate the mask too
+  for (int gi = 1; gi < glwe_dimension + 1; ++gi)
+    polynomial_accumulate_monic_monomial_mul<Torus>(
+        out_result + gi * polynomial_size, in_poly + gi * polynomial_size,
+        degree, coeffIdx, polynomial_size, 1, true);
+}
+
+template <typename Torus, typename TorusVec>
+__host__ void host_fast_packing_keyswitch_lwe_list_to_glwe(
+    cudaStream_t stream, uint32_t gpu_index, Torus *glwe_out,
+    Torus const *lwe_array_in, Torus const *fp_ksk_array, int8_t *fp_ks_buffer,
+    uint32_t lwe_dimension, uint32_t glwe_dimension, uint32_t polynomial_size,
+    uint32_t base_log, uint32_t level_count, uint32_t num_lwes) {
+
+  // Optimization of packing keyswitch when packing many LWEs
+
+  cuda_set_device(gpu_index);
+  check_cuda_error(cudaGetLastError());
+
+  int glwe_accumulator_size = (glwe_dimension + 1) * polynomial_size;
+
+  // The fast path of PKS uses the scratch buffer (d_mem) differently than the
+  // old path: it needs to store the decomposed masks in the first half of this
+  // buffer and the keyswitched GLWEs in the second half of the buffer. Thus the
+  // scratch buffer for the fast path must determine the half-size of the
+  // scratch buffer as the max between the size of the GLWE and the size of the
+  // LWE-mask times two (to keep both decomposition state and decomposed
+  // intermediate value)
+  int memory_unit = glwe_accumulator_size > lwe_dimension * 2
+                        ? glwe_accumulator_size
+                        : lwe_dimension * 2;
+
+  // ping pong the buffer between successive calls
+  // split the buffer in two parts of this size
+  auto d_mem_0 = (Torus *)fp_ks_buffer;
+  auto d_mem_1 = d_mem_0 + num_lwes * memory_unit;
+
+  // Set the scratch buffer to 0 as it is used to accumulate
+  // decomposition temporary results
+  cuda_memset_async(d_mem_1, 0, num_lwes * memory_unit * sizeof(Torus), stream,
+                    gpu_index);
+  check_cuda_error(cudaGetLastError());
+
+  // decompose LWEs
+  // don't decompose LWE body - the LWE has lwe_size + 1 elements. The last
+  // element, the body is ignored by rounding down the number of blocks assuming
+  // here that the LWE dimension is a multiple of the block size
+  dim3 grid_decomp(CEIL_DIV(num_lwes, BLOCK_SIZE_DECOMP),
+                   CEIL_DIV(lwe_dimension, BLOCK_SIZE_DECOMP));
+  dim3 threads_decomp(BLOCK_SIZE_DECOMP, BLOCK_SIZE_DECOMP);
+
+  // decompose first level
+  decompose_vectorize_init<Torus, TorusVec>
+      <<<grid_decomp, threads_decomp, 0, stream>>>(lwe_array_in, d_mem_0,
+                                                   lwe_dimension, num_lwes,
+                                                   base_log, level_count);
+  check_cuda_error(cudaGetLastError());
+
+  // gemm to ks the individual LWEs to GLWEs
+  dim3 grid_gemm(CEIL_DIV(glwe_accumulator_size, BLOCK_SIZE_GEMM),
+                 CEIL_DIV(num_lwes, BLOCK_SIZE_GEMM));
+  dim3 threads_gemm(BLOCK_SIZE_GEMM * THREADS_GEMM);
+
+  auto stride_KSK_buffer = glwe_accumulator_size * level_count;
+
+  uint32_t shared_mem_size = get_shared_mem_size_tgemm<Torus>();
+  tgemm<Torus, TorusVec><<<grid_gemm, threads_gemm, shared_mem_size, stream>>>(
+      num_lwes, glwe_accumulator_size, lwe_dimension, d_mem_0, fp_ksk_array,
+      stride_KSK_buffer, d_mem_1);
+  check_cuda_error(cudaGetLastError());
+
+  auto ksk_block_size = glwe_accumulator_size;
+
+  for (int li = 1; li < level_count; ++li) {
+    decompose_vectorize_step_inplace<Torus, TorusVec>
+        <<<grid_decomp, threads_decomp, 0, stream>>>(
+            d_mem_0, lwe_dimension, num_lwes, base_log, level_count);
+    check_cuda_error(cudaGetLastError());
+
+    tgemm<Torus, TorusVec>
+        <<<grid_gemm, threads_gemm, shared_mem_size, stream>>>(
+            num_lwes, glwe_accumulator_size, lwe_dimension, d_mem_0,
+            fp_ksk_array + li * ksk_block_size, stride_KSK_buffer, d_mem_1);
+    check_cuda_error(cudaGetLastError());
+  }
+
+  // should we include the mask in the rotation ??
+  dim3 grid_rotate(CEIL_DIV(num_lwes, BLOCK_SIZE_DECOMP),
+                   CEIL_DIV(polynomial_size, BLOCK_SIZE_DECOMP));
+  dim3 threads_rotate(BLOCK_SIZE_DECOMP, BLOCK_SIZE_DECOMP);
+  // rotate the GLWEs
+  polynomial_accumulate_monic_monomial_mul_many_neg_and_add_C<Torus>
+      <<<grid_rotate, threads_rotate, 0, stream>>>(
+          d_mem_1, d_mem_0, lwe_array_in, lwe_dimension, num_lwes,
+          polynomial_size, glwe_dimension);
+  check_cuda_error(cudaGetLastError());
+
+  dim3 grid_accumulate(
+      CEIL_DIV(polynomial_size * (glwe_dimension + 1), BLOCK_SIZE_DECOMP));
+  dim3 threads_accum(BLOCK_SIZE_DECOMP);
+
+  // accumulate to a single glwe
+  accumulate_glwes<Torus><<<grid_accumulate, threads_accum, 0, stream>>>(
+      glwe_out, d_mem_0, glwe_dimension, polynomial_size, num_lwes);
+
+  check_cuda_error(cudaGetLastError());
+}
+
+#endif

backends/tfhe-cuda-backend/cuda/src/crypto/gadget.cuh

@@ -3,6 +3,7 @@
 
 #include "crypto/torus.cuh"
 #include "device.h"
+#include "fft128/f128.cuh"
 #include <cstdint>
 
 /**
@@ -42,6 +43,13 @@ public:
     }
   }
 
+  __device__ void decompose_and_compress_next_128(double *result) {
+    for (int j = 0; j < num_poly; j++) {
+      auto result_slice = result + j * params::degree / 2 * 4;
+      decompose_and_compress_next_polynomial_128(result_slice, j);
+    }
+  }
+
   // Decomposes a single polynomial
   __device__ void decompose_and_compress_next_polynomial(double2 *result,
                                                          int j) {
@@ -75,10 +83,58 @@ public:
     synchronize_threads_in_block();
   }
 
+  // Decomposes a single polynomial
+  __device__ void decompose_and_compress_next_polynomial_128(double *result,
+                                                             int j) {
+    uint32_t tid = threadIdx.x;
+    auto state_slice = &state[j * params::degree];
+    for (int i = 0; i < params::opt / 2; i++) {
+      auto input1 = &state_slice[tid];
+      auto input2 = &state_slice[tid + params::degree / 2];
+      T res_re = *input1 & mask_mod_b;
+      T res_im = *input2 & mask_mod_b;
+
+      *input1 >>= base_log; // Update state
+      *input2 >>= base_log; // Update state
+
+      T carry_re = ((res_re - 1ll) | *input1) & res_re;
+      T carry_im = ((res_im - 1ll) | *input2) & res_im;
+      carry_re >>= (base_log - 1);
+      carry_im >>= (base_log - 1);
+
+      *input1 += carry_re; // Update state
+      *input2 += carry_im; // Update state
+
+      res_re -= carry_re << base_log;
+      res_im -= carry_im << base_log;
+
+      auto out_re = u128_to_signed_to_f128(res_re);
+      auto out_im = u128_to_signed_to_f128(res_im);
+
+      auto out_re_hi = result + 0 * params::degree / 2;
+      auto out_re_lo = result + 1 * params::degree / 2;
+      auto out_im_hi = result + 2 * params::degree / 2;
+      auto out_im_lo = result + 3 * params::degree / 2;
+
+      out_re_hi[tid] = out_re.hi;
+      out_re_lo[tid] = out_re.lo;
+      out_im_hi[tid] = out_im.hi;
+      out_im_lo[tid] = out_im.lo;
+
+      tid += params::degree / params::opt;
+    }
+    synchronize_threads_in_block();
+  }
+
   __device__ void decompose_and_compress_level(double2 *result, int level) {
     for (int i = 0; i < level_count - level; i++)
       decompose_and_compress_next(result);
   }
+
+  __device__ void decompose_and_compress_level_128(double *result, int level) {
+    for (int i = 0; i < level_count - level; i++)
+      decompose_and_compress_next_128(result);
+  }
 };
 
 template <typename Torus>

backends/tfhe-cuda-backend/cuda/src/crypto/ggsw.cuh

@@ -57,7 +57,7 @@ void batch_fft_ggsw_vector(cudaStream_t *streams, uint32_t *gpu_indexes,
   if (gpu_count != 1)
     PANIC("GPU error (batch_fft_ggsw_vector): multi-GPU execution is not "
           "supported yet.")
-  cudaSetDevice(gpu_indexes[0]);
+  cuda_set_device(gpu_indexes[0]);
 
   int shared_memory_size = sizeof(double) * polynomial_size;
 

backends/tfhe-cuda-backend/cuda/src/crypto/keyswitch.cu

@@ -1,6 +1,8 @@
+#include "fast_packing_keyswitch.cuh"
 #include "keyswitch.cuh"
 #include "keyswitch.h"
 #include <cstdint>
+#include <stdio.h>
 
 /* Perform keyswitch on a batch of 32 bits input LWE ciphertexts.
  * Head out to the equivalent operation on 64 bits for more details.
@@ -53,15 +55,17 @@ void cuda_keyswitch_lwe_ciphertext_vector_64(
 
 void scratch_packing_keyswitch_lwe_list_to_glwe_64(
     void *stream, uint32_t gpu_index, int8_t **fp_ks_buffer,
-    uint32_t glwe_dimension, uint32_t polynomial_size, uint32_t num_lwes,
-    bool allocate_gpu_memory) {
+    uint32_t lwe_dimension, uint32_t glwe_dimension, uint32_t polynomial_size,
+    uint32_t num_lwes, bool allocate_gpu_memory) {
   scratch_packing_keyswitch_lwe_list_to_glwe<uint64_t>(
-      static_cast<cudaStream_t>(stream), gpu_index, fp_ks_buffer,
+      static_cast<cudaStream_t>(stream), gpu_index, fp_ks_buffer, lwe_dimension,
       glwe_dimension, polynomial_size, num_lwes, allocate_gpu_memory);
 }
+
 /* Perform functional packing keyswitch on a batch of 64 bits input LWE
  * ciphertexts.
  */
+
 void cuda_packing_keyswitch_lwe_list_to_glwe_64(
     void *stream, uint32_t gpu_index, void *glwe_array_out,
     void const *lwe_array_in, void const *fp_ksk_array, int8_t *fp_ks_buffer,
@@ -69,7 +73,7 @@ void cuda_packing_keyswitch_lwe_list_to_glwe_64(
     uint32_t output_polynomial_size, uint32_t base_log, uint32_t level_count,
     uint32_t num_lwes) {
 
-  host_packing_keyswitch_lwe_list_to_glwe<uint64_t>(
+  host_fast_packing_keyswitch_lwe_list_to_glwe<uint64_t, ulonglong4>(
       static_cast<cudaStream_t>(stream), gpu_index,
       static_cast<uint64_t *>(glwe_array_out),
       static_cast<const uint64_t *>(lwe_array_in),