Debug with long run tests

- From now on the streams stored in the available cuda server key are the ones to be

.github/actionlint.yaml

@@ -6,6 +6,7 @@ self-hosted-runner:
     - large_windows_16_latest
     - large_ubuntu_16
     - large_ubuntu_16-22.04
+    - v80-desktop
 # 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

@@ -8,9 +8,6 @@ inputs:
   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'
@@ -22,41 +19,58 @@ runs:
     - name: Install dependencies
       shell: bash
       run: |
+        wget https://github.com/Kitware/CMake/releases/download/v"${CMAKE_VERSION}"/cmake-"${CMAKE_VERSION}"-linux-x86_64.sh
+        echo "${CMAKE_SCRIPT_SHA} cmake-${CMAKE_VERSION}-linux-x86_64.sh" > checksum
+        sha256sum -c checksum
+        sudo bash cmake-"${CMAKE_VERSION}"-linux-x86_64.sh --skip-license --prefix=/usr/ --exclude-subdir
         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
+        sudo apt install -y cmake-format libclang-dev
+      env:
+        CMAKE_VERSION: 3.29.6
+        CMAKE_SCRIPT_SHA: "6e4fada5cba3472ae503a11232b6580786802f0879cead2741672bf65d97488a"
 
     - 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
+        TOOLKIT_VERSION="$(echo ${CUDA_VERSION} | sed 's/\(.*\)\.\(.*\)/\1-\2/')"
+        wget https://developer.download.nvidia.com/compute/cuda/repos/ubuntu2204/x86_64/${env.CUDA_KEYRING_PACKAGE}
+        echo "${CUDA_KEYRING_SHA} ${CUDA_KEYRING_PACKAGE}" > checksum
+        sha256sum -c checksum
+        sudo dpkg -i "${CUDA_KEYRING_PACKAGE}"
         sudo apt update
-        sudo apt -y install cuda-toolkit-${TOOLKIT_VERSION}
+        sudo apt -y install cuda-toolkit-"${TOOLKIT_VERSION}"
+      env:
+        CUDA_VERSION: ${{ inputs.cuda-version }}
+        CUDA_KEYRING_PACKAGE: cuda-keyring_1.1-1_all.deb
+        CUDA_KEYRING_SHA: "d93190d50b98ad4699ff40f4f7af50f16a76dac3bb8da1eaaf366d47898ff8df"
 
     - 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}"
+        CUDA_PATH=/usr/local/cuda-"${CUDA_VERSION}"
+        {
+          echo "CUDA_PATH=$CUDA_PATH";
+          echo "LD_LIBRARY_PATH=$CUDA_PATH/lib64:$LD_LIBRARY_PATH";
+          echo "CUDA_MODULE_LOADER=EAGER";
+        } >> "${GITHUB_ENV}"
+        {
+          echo "PATH=$PATH:$CUDA_PATH/bin"; 
+        } >> "${GITHUB_PATH}"
+      env:
+        CUDA_VERSION: ${{ inputs.cuda-version }}
 
     # 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++-${{ inputs.gcc-version }}";
+          echo "CC=/usr/bin/gcc-${GCC_VERSION}";
+          echo "CXX=/usr/bin/g++-${GCC_VERSION}";
+          echo "CUDAHOSTCXX=/usr/bin/g++-${GCC_VERSION}";
         } >> "${GITHUB_ENV}"
+      env:
+        GCC_VERSION: ${{ inputs.gcc-version }}
 
     - name: Check device is detected
       shell: bash

.github/workflows/approve_label.yml

@@ -6,6 +6,9 @@ on:
   pull_request_review:
     types: [submitted]
 
+
+permissions: {}
+
 jobs:
   trigger-tests:
     runs-on: ubuntu-latest
@@ -34,3 +37,10 @@ jobs:
           # We need to use a PAT to be able to trigger `labeled` event for the other workflow.
           github_token: ${{ secrets.FHE_ACTIONS_TOKEN }}
           labels: approved
+
+      - name: Check if maintainer needs to handle label manually
+        if: ${{ failure() }}
+        run: |
+          echo "Pull-request from an external contributor."
+          echo "A maintainer need to manually add/remove the 'approved' label."
+          exit 1

.github/workflows/aws_tfhe_backward_compat_tests.yml

@@ -23,6 +23,9 @@ on:
   workflow_dispatch:
   pull_request:
 
+permissions:
+  contents: read
+
 jobs:
   setup-instance:
     name: Setup instance (backward-compat-tests)
@@ -47,7 +50,7 @@ jobs:
         id: start-github-instance
         if: env.SECRETS_AVAILABLE == 'false'
         run: |
-          echo "runner_group=${{ env.EXTERNAL_CONTRIBUTION_RUNNER }}" >> "$GITHUB_OUTPUT"
+          echo "runner_group=${EXTERNAL_CONTRIBUTION_RUNNER}" >> "$GITHUB_OUTPUT"
 
   backward-compat-tests:
     name: Backward compatibility tests
@@ -83,11 +86,12 @@ jobs:
 
       - name: Get backward compat branch head SHA
         id: backward_compat_sha
+        run: |
+          SHA=$(git ls-remote "${REPO_URL}" refs/heads/"${BACKWARD_COMPAT_BRANCH}" | awk '{print $1}')
+          echo "sha=${SHA}" >> "${GITHUB_OUTPUT}"
         env:
           REPO_URL: "https://github.com/zama-ai/tfhe-backward-compat-data"
-        run: |
-          SHA=$(git ls-remote ${{ env.REPO_URL }} refs/heads/${{ steps.backward_compat_branch.outputs.branch }} | awk '{print $1}')
-          echo "sha=${SHA}" >> "${GITHUB_OUTPUT}"
+          BACKWARD_COMPAT_BRANCH: ${{ steps.backward_compat_branch.outputs.branch }}
 
       - name: Retrieve data from cache
         id: retrieve-data-cache
@@ -101,6 +105,7 @@ jobs:
         uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
         with:
           persist-credentials: 'false'
+          token: ${{ env.CHECKOUT_TOKEN }}
           repository: zama-ai/tfhe-backward-compat-data
           path: tests/tfhe-backward-compat-data
           lfs: 'true'
@@ -121,10 +126,12 @@ jobs:
       - 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}"
+          echo "PULL_REQUEST_MD_LINK=[pull-request](${PR_BASE_URL}${{ github.event.pull_request.number }}), "  >> "${GITHUB_ENV}"
+        env:
+          PR_BASE_URL: ${{ vars.PR_BASE_URL }}
 
       - name: Slack Notification
-        if: ${{ failure() }}
+        if: ${{ failure() || (cancelled() && github.event_name != 'pull_request') }}
         continue-on-error: true
         uses: rtCamp/action-slack-notify@e31e87e03dd19038e411e38ae27cbad084a90661
         env:

.github/workflows/aws_tfhe_fast_tests.yml

@@ -24,6 +24,9 @@ on:
   workflow_dispatch:
   pull_request:
 
+permissions:
+  contents: read
+
 jobs:
   should-run:
     runs-on: ubuntu-latest
@@ -154,7 +157,7 @@ jobs:
         id: start-github-instance
         if: env.SECRETS_AVAILABLE == 'false'
         run: |
-          echo "runner_group=${{ env.EXTERNAL_CONTRIBUTION_RUNNER }}" >> "$GITHUB_OUTPUT"
+          echo "runner_group=${EXTERNAL_CONTRIBUTION_RUNNER}" >> "$GITHUB_OUTPUT"
 
   fast-tests:
     name: Fast CPU tests
@@ -269,7 +272,9 @@ jobs:
       - 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}"
+          echo "PULL_REQUEST_MD_LINK=[pull-request](${PR_BASE_URL}${{ github.event.pull_request.number }}), "  >> "${GITHUB_ENV}"
+        env:
+          PR_BASE_URL: ${{ vars.PR_BASE_URL }}
 
       - name: Slack Notification
         if: ${{ failure() && env.SECRETS_AVAILABLE == 'true' }}
@@ -297,7 +302,7 @@ jobs:
           label: ${{ needs.setup-instance.outputs.runner-name }}
 
       - name: Slack Notification
-        if: ${{ failure() }}
+        if: ${{ failure() || (cancelled() && github.event_name != 'pull_request') }}
         continue-on-error: true
         uses: rtCamp/action-slack-notify@e31e87e03dd19038e411e38ae27cbad084a90661
         env:

.github/workflows/aws_tfhe_integer_tests.yml

@@ -30,6 +30,9 @@ on:
     branches:
       - main
 
+permissions:
+  contents: read
+
 jobs:
   should-run:
     if:
@@ -94,7 +97,7 @@ jobs:
         id: start-github-instance
         if: env.SECRETS_AVAILABLE == 'false'
         run: |
-          echo "runner_group=${{ env.EXTERNAL_CONTRIBUTION_RUNNER }}" >> "$GITHUB_OUTPUT"
+          echo "runner_group=${EXTERNAL_CONTRIBUTION_RUNNER}" >> "$GITHUB_OUTPUT"
 
   unsigned-integer-tests:
     name: Unsigned integer tests
@@ -134,15 +137,17 @@ jobs:
 
       - name: Run unsigned integer tests
         run: |
-          AVX512_SUPPORT=ON NO_BIG_PARAMS=${{ env.NO_BIG_PARAMS }} BIG_TESTS_INSTANCE=TRUE make test_unsigned_integer_ci
+          AVX512_SUPPORT=ON NO_BIG_PARAMS="${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}"
+          echo "PULL_REQUEST_MD_LINK=[pull-request](${PR_BASE_URL}${{ github.event.pull_request.number }}), "  >> "${GITHUB_ENV}"
+        env:
+          PR_BASE_URL: ${{ vars.PR_BASE_URL }}
 
       - name: Slack Notification
-        if: ${{ failure() }}
+        if: ${{ failure() || (cancelled() && github.event_name != 'pull_request') }}
         continue-on-error: true
         uses: rtCamp/action-slack-notify@e31e87e03dd19038e411e38ae27cbad084a90661
         env:

.github/workflows/aws_tfhe_signed_integer_tests.yml

@@ -30,6 +30,9 @@ on:
     branches:
       - main
 
+permissions:
+  contents: read
+
 jobs:
   should-run:
     if:
@@ -95,7 +98,7 @@ jobs:
         id: start-github-instance
         if: env.SECRETS_AVAILABLE == 'false'
         run: |
-          echo "runner_group=${{ env.EXTERNAL_CONTRIBUTION_RUNNER }}" >> "$GITHUB_OUTPUT"
+          echo "runner_group=${EXTERNAL_CONTRIBUTION_RUNNER}" >> "$GITHUB_OUTPUT"
 
   signed-integer-tests:
     name: Signed integer tests
@@ -139,15 +142,17 @@ jobs:
 
       - name: Run signed integer tests
         run: |
-          AVX512_SUPPORT=ON NO_BIG_PARAMS=${{ env.NO_BIG_PARAMS }} BIG_TESTS_INSTANCE=TRUE make test_signed_integer_ci
+          AVX512_SUPPORT=ON NO_BIG_PARAMS="${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}"
+          echo "PULL_REQUEST_MD_LINK=[pull-request](${PR_BASE_URL}${{ github.event.pull_request.number }}), "  >> "${GITHUB_ENV}"
+        env:
+          PR_BASE_URL: ${{ vars.PR_BASE_URL }}
 
       - name: Slack Notification
-        if: ${{ failure() }}
+        if: ${{ failure() || (cancelled() && github.event_name != 'pull_request') }}
         continue-on-error: true
         uses: rtCamp/action-slack-notify@e31e87e03dd19038e411e38ae27cbad084a90661
         env:

.github/workflows/aws_tfhe_tests.yml

@@ -27,6 +27,9 @@ on:
     # Nightly tests @ 1AM after each work day
     - cron: "0 1 * * MON-FRI"
 
+permissions:
+  contents: read
+
 jobs:
   should-run:
     runs-on: ubuntu-latest
@@ -163,7 +166,7 @@ jobs:
         id: start-github-instance
         if: env.SECRETS_AVAILABLE == 'false'
         run: |
-          echo "runner_group=${{ env.EXTERNAL_CONTRIBUTION_RUNNER }}" >> "$GITHUB_OUTPUT"
+          echo "runner_group=${EXTERNAL_CONTRIBUTION_RUNNER}" >> "$GITHUB_OUTPUT"
 
   cpu-tests:
     name: CPU tests
@@ -251,10 +254,12 @@ jobs:
       - 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}"
+          echo "PULL_REQUEST_MD_LINK=[pull-request](${PR_BASE_URL}${{ github.event.pull_request.number }}), "  >> "${GITHUB_ENV}"
+        env:
+          PR_BASE_URL: ${{ vars.PR_BASE_URL }}
 
       - name: Slack Notification
-        if: ${{ failure() }}
+        if: ${{ failure() || (cancelled() && github.event_name != 'pull_request') }}
         continue-on-error: true
         uses: rtCamp/action-slack-notify@e31e87e03dd19038e411e38ae27cbad084a90661
         env:

.github/workflows/aws_tfhe_wasm_tests.yml

@@ -23,6 +23,9 @@ on:
   pull_request:
     types: [ labeled ]
 
+permissions:
+  contents: read
+
 jobs:
   setup-instance:
     name: Setup instance (wasm-tests)
@@ -48,7 +51,7 @@ jobs:
         id: start-github-instance
         if: env.SECRETS_AVAILABLE == 'false'
         run: |
-          echo "runner_group=${{ env.EXTERNAL_CONTRIBUTION_RUNNER }}" >> "$GITHUB_OUTPUT"
+          echo "runner_group=${EXTERNAL_CONTRIBUTION_RUNNER}" >> "$GITHUB_OUTPUT"
 
   wasm-tests:
     name: WASM tests
@@ -120,10 +123,12 @@ jobs:
       - 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}"
+          echo "PULL_REQUEST_MD_LINK=[pull-request](${PR_BASE_URL}${{ github.event.pull_request.number }}), "  >> "${GITHUB_ENV}"
+        env:
+          PR_BASE_URL: ${{ vars.PR_BASE_URL }}
 
       - name: Slack Notification
-        if: ${{ failure() }}
+        if: ${{ failure() || (cancelled() && github.event_name != 'pull_request') }}
         continue-on-error: true
         uses: rtCamp/action-slack-notify@e31e87e03dd19038e411e38ae27cbad084a90661
         env:

.github/workflows/benchmark_boolean.yml

@@ -18,6 +18,9 @@ env:
   SLACK_USERNAME: ${{ secrets.BOT_USERNAME }}
   SLACK_WEBHOOK: ${{ secrets.SLACK_WEBHOOK }}
 
+
+permissions: {}
+
 jobs:
   setup-instance:
     name: Setup instance (boolean-benchmarks)
@@ -45,7 +48,6 @@ jobs:
     concurrency:
       group: ${{ github.workflow_ref }}
       cancel-in-progress: ${{ github.ref != 'refs/heads/main' }}
-    continue-on-error: true
     steps:
       - name: Checkout tfhe-rs repo with tags
         uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
@@ -73,15 +75,17 @@ jobs:
 
       - name: Parse results
         run: |
-          python3 ./ci/benchmark_parser.py target/criterion ${{ env.RESULTS_FILENAME }} \
+          python3 ./ci/benchmark_parser.py target/criterion "${RESULTS_FILENAME}" \
           --database tfhe_rs \
           --hardware "hpc7a.96xlarge" \
-          --project-version "${{ env.COMMIT_HASH }}" \
-          --branch ${{ github.ref_name }} \
-          --commit-date "${{ env.COMMIT_DATE }}" \
-          --bench-date "${{ env.BENCH_DATE }}" \
+          --project-version "${COMMIT_HASH}" \
+          --branch "${REF_NAME}" \
+          --commit-date "${COMMIT_DATE}" \
+          --bench-date "${BENCH_DATE}" \
           --walk-subdirs \
           --name-suffix avx512
+        env:
+          REF_NAME: ${{ github.ref_name }}
 
       - name: Measure key sizes
         run: |
@@ -89,7 +93,7 @@ jobs:
 
       - name: Parse key sizes results
         run: |
-          python3 ./ci/benchmark_parser.py tfhe/boolean_key_sizes.csv ${{ env.RESULTS_FILENAME }} \
+          python3 ./ci/benchmark_parser.py tfhe-benchmark/boolean_key_sizes.csv "${RESULTS_FILENAME}" \
           --object-sizes \
           --append-results
 
@@ -110,11 +114,11 @@ jobs:
       - name: Send data to Slab
         shell: bash
         run: |
-          python3 slab/scripts/data_sender.py ${{ env.RESULTS_FILENAME }} "${{ secrets.JOB_SECRET }}" \
+          python3 slab/scripts/data_sender.py "${RESULTS_FILENAME}" "${{ secrets.JOB_SECRET }}" \
           --slab-url "${{ secrets.SLAB_URL }}"
 
       - name: Slack Notification
-        if: ${{ failure() }}
+        if: ${{ failure() || (cancelled() && github.event_name != 'pull_request') }}
         continue-on-error: true
         uses: rtCamp/action-slack-notify@e31e87e03dd19038e411e38ae27cbad084a90661
         env:

.github/workflows/benchmark_core_crypto.yml

@@ -18,6 +18,9 @@ env:
   SLACK_USERNAME: ${{ secrets.BOT_USERNAME }}
   SLACK_WEBHOOK: ${{ secrets.SLACK_WEBHOOK }}
 
+
+permissions: {}
+
 jobs:
   setup-instance:
     name: Setup instance (core-crypto-benchmarks)
@@ -75,15 +78,17 @@ jobs:
 
       - name: Parse results
         run: |
-          python3 ./ci/benchmark_parser.py target/criterion ${{ env.RESULTS_FILENAME }} \
+          python3 ./ci/benchmark_parser.py target/criterion "${RESULTS_FILENAME}" \
           --database tfhe_rs \
           --hardware "hpc7a.96xlarge" \
-          --project-version "${{ env.COMMIT_HASH }}" \
-          --branch ${{ github.ref_name }} \
-          --commit-date "${{ env.COMMIT_DATE }}" \
-          --bench-date "${{ env.BENCH_DATE }}" \
+          --project-version "${COMMIT_HASH}" \
+          --branch "${REF_NAME}" \
+          --commit-date "${COMMIT_DATE}" \
+          --bench-date "${BENCH_DATE}" \
           --name-suffix avx512 \
           --walk-subdirs
+        env:
+          REF_NAME: ${{ github.ref_name }}
 
       - name: Upload parsed results artifact
         uses: actions/upload-artifact@ea165f8d65b6e75b540449e92b4886f43607fa02
@@ -102,11 +107,11 @@ jobs:
       - name: Send data to Slab
         shell: bash
         run: |
-          python3 slab/scripts/data_sender.py ${{ env.RESULTS_FILENAME }} "${{ secrets.JOB_SECRET }}" \
+          python3 slab/scripts/data_sender.py "${RESULTS_FILENAME}" "${{ secrets.JOB_SECRET }}" \
           --slab-url "${{ secrets.SLAB_URL }}"
 
       - name: Slack Notification
-        if: ${{ failure() }}
+        if: ${{ failure() || (cancelled() && github.event_name != 'pull_request') }}
         continue-on-error: true
         uses: rtCamp/action-slack-notify@e31e87e03dd19038e411e38ae27cbad084a90661
         env:

.github/workflows/benchmark_dex.yml

@@ -0,0 +1,152 @@
+# Run all DEX benchmarks on an AWS instance and return parsed results to Slab CI bot.
+name: DEX benchmarks
+
+on:
+  workflow_dispatch:
+  schedule:
+    # Weekly benchmarks will be triggered each Saturday at 5a.m.
+    - cron: '0 5 * * 6'
+
+env:
+  CARGO_TERM_COLOR: always
+  RESULTS_FILENAME: parsed_benchmark_results_${{ github.sha }}.json
+  ACTION_RUN_URL: ${{ github.server_url }}/${{ github.repository }}/actions/runs/${{ github.run_id }}
+  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 }}
+
+permissions: {}
+
+jobs:
+  setup-instance:
+    name: Setup instance (dex-benchmarks)
+    runs-on: ubuntu-latest
+    if: github.event_name == 'workflow_dispatch' ||
+      (github.event_name == 'schedule' && github.repository == 'zama-ai/tfhe-rs')
+    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: bench
+
+  dex-benchmarks:
+    name: Execute DEX benchmarks
+    needs: setup-instance
+    runs-on: ${{ needs.setup-instance.outputs.runner-name }}
+    concurrency:
+      group: ${{ github.workflow_ref }}
+      cancel-in-progress: ${{ github.ref != 'refs/heads/main' }}
+    timeout-minutes: 720  # 12 hours
+    steps:
+      - name: Checkout tfhe-rs repo with tags
+        uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
+        with:
+          fetch-depth: 0
+          persist-credentials: 'false'
+          token: ${{ secrets.REPO_CHECKOUT_TOKEN }}
+
+      - name: Get benchmark details
+        run: |
+          {
+            echo "BENCH_DATE=$(date --iso-8601=seconds)";
+            echo "COMMIT_DATE=$(git --no-pager show -s --format=%cd --date=iso8601-strict ${{ github.sha }})";
+            echo "COMMIT_HASH=$(git describe --tags --dirty)";
+          } >> "${GITHUB_ENV}"
+
+      - name: Install rust
+        uses: dtolnay/rust-toolchain@888c2e1ea69ab0d4330cbf0af1ecc7b68f368cc1
+        with:
+          toolchain: nightly
+
+      - name: Checkout Slab repo
+        uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
+        with:
+          repository: zama-ai/slab
+          path: slab
+          persist-credentials: 'false'
+          token: ${{ secrets.REPO_CHECKOUT_TOKEN }}
+
+      - name: Run benchmarks
+        run: |
+          make bench_hlapi_dex
+
+      - name: Parse results
+        run: |
+          python3 ./ci/benchmark_parser.py target/criterion "${RESULTS_FILENAME}" \
+          --database tfhe_rs \
+          --hardware "hpc7a.96xlarge" \
+          --project-version "${COMMIT_HASH}" \
+          --branch "${REF_NAME}" \
+          --commit-date "${COMMIT_DATE}" \
+          --bench-date "${BENCH_DATE}" \
+          --walk-subdirs \
+          --name-suffix avx512
+        env:
+          REF_NAME: ${{ github.ref_name }}
+
+      - name: Parse swap request PBS counts
+        run: |
+          python3 ./ci/benchmark_parser.py tfhe-benchmark/dex_swap_request_pbs_count.csv "${RESULTS_FILENAME}" \
+          --object-sizes \
+          --append-results
+
+      - name: Parse swap claim PBS counts
+        run: |
+          python3 ./ci/benchmark_parser.py tfhe-benchmark/dex_swap_claim_pbs_count.csv "${RESULTS_FILENAME}" \
+          --object-sizes \
+          --append-results
+
+      - name: Upload parsed results artifact
+        uses: actions/upload-artifact@ea165f8d65b6e75b540449e92b4886f43607fa02
+        with:
+          name: ${{ github.sha }}_dex
+          path: ${{ env.RESULTS_FILENAME }}
+
+      - name: Send data to Slab
+        shell: bash
+        run: |
+          python3 slab/scripts/data_sender.py "${RESULTS_FILENAME}" "${{ secrets.JOB_SECRET }}" \
+          --slab-url "${{ secrets.SLAB_URL }}"
+
+      - name: Slack Notification
+        if: ${{ failure() || (cancelled() && github.event_name != 'pull_request') }}
+        continue-on-error: true
+        uses: rtCamp/action-slack-notify@e31e87e03dd19038e411e38ae27cbad084a90661
+        env:
+          SLACK_COLOR: ${{ job.status }}
+          SLACK_MESSAGE: "DEX benchmarks finished with status: ${{ job.status }}. (${{ env.ACTION_RUN_URL }})"
+
+  teardown-instance:
+    name: Teardown instance (dex-benchmarks)
+    if: ${{ always() && needs.setup-instance.result == 'success' }}
+    needs: [ setup-instance, dex-benchmarks ]
+    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@e31e87e03dd19038e411e38ae27cbad084a90661
+        env:
+          SLACK_COLOR: ${{ job.status }}
+          SLACK_MESSAGE: "Instance teardown (dex-benchmarks) finished with status: ${{ job.status }}. (${{ env.ACTION_RUN_URL }})"

.github/workflows/benchmark_erc20.yml

@@ -18,6 +18,9 @@ env:
   SLACK_USERNAME: ${{ secrets.BOT_USERNAME }}
   SLACK_WEBHOOK: ${{ secrets.SLACK_WEBHOOK }}
 
+
+permissions: {}
+
 jobs:
   setup-instance:
     name: Setup instance (erc20-benchmarks)
@@ -45,7 +48,6 @@ jobs:
     concurrency:
       group: ${{ github.workflow_ref }}
       cancel-in-progress: ${{ github.ref != 'refs/heads/main' }}
-    continue-on-error: true
     timeout-minutes: 720  # 12 hours
     steps:
       - name: Checkout tfhe-rs repo with tags
@@ -82,19 +84,21 @@ jobs:
 
       - name: Parse results
         run: |
-          python3 ./ci/benchmark_parser.py target/criterion ${{ env.RESULTS_FILENAME }} \
+          python3 ./ci/benchmark_parser.py target/criterion "${RESULTS_FILENAME}" \
           --database tfhe_rs \
           --hardware "hpc7a.96xlarge" \
-          --project-version "${{ env.COMMIT_HASH }}" \
-          --branch ${{ github.ref_name }} \
-          --commit-date "${{ env.COMMIT_DATE }}" \
-          --bench-date "${{ env.BENCH_DATE }}" \
+          --project-version "${COMMIT_HASH}" \
+          --branch "${REF_NAME}" \
+          --commit-date "${COMMIT_DATE}" \
+          --bench-date "${BENCH_DATE}" \
           --walk-subdirs \
           --name-suffix avx512
+        env:
+          REF_NAME: ${{ github.ref_name }}
 
       - name: Parse PBS counts
         run: |
-          python3 ./ci/benchmark_parser.py tfhe/erc20_pbs_count.csv ${{ env.RESULTS_FILENAME }} \
+          python3 ./ci/benchmark_parser.py tfhe-benchmark/erc20_pbs_count.csv "${RESULTS_FILENAME}" \
           --object-sizes \
           --append-results
 
@@ -107,11 +111,11 @@ jobs:
       - name: Send data to Slab
         shell: bash
         run: |
-          python3 slab/scripts/data_sender.py ${{ env.RESULTS_FILENAME }} "${{ secrets.JOB_SECRET }}" \
+          python3 slab/scripts/data_sender.py "${RESULTS_FILENAME}" "${{ secrets.JOB_SECRET }}" \
           --slab-url "${{ secrets.SLAB_URL }}"
 
       - name: Slack Notification
-        if: ${{ failure() }}
+        if: ${{ failure() || (cancelled() && github.event_name != 'pull_request') }}
         continue-on-error: true
         uses: rtCamp/action-slack-notify@e31e87e03dd19038e411e38ae27cbad084a90661
         env:

.github/workflows/benchmark_gpu.yml

@@ -10,13 +10,14 @@ on:
         type: choice
         options:
           - "l40 (n3-L40x1)"
+          - "4-l40 (n3-L40x4)"
+          - "multi-a100-nvlink (n3-A100x8-NVLink)"
           - "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)"
-          - "multi-a100-nvlink (n3-A100x8-NVLink)"
       command:
         description: "Benchmark command to run"
         type: choice
@@ -59,22 +60,33 @@ on:
           - multi_bit
           - both
 
+
+permissions: {}
+
 jobs:
   parse-inputs:
     runs-on: ubuntu-latest
     outputs:
       profile: ${{ steps.parse_profile.outputs.profile }}
       hardware_name: ${{ steps.parse_hardware_name.outputs.name }}
+    env:
+      INPUTS_PROFILE: ${{ inputs.profile }}
     steps:
       - name: Parse profile
         id: parse_profile
         run: |
-          echo "profile=$(echo '${{ inputs.profile }}' | sed 's|\(.*\)[[:space:]](.*)|\1|')" >> "${GITHUB_OUTPUT}"
+          # Use Sed to extract a value from a string, this cannot be done with the ${variable//search/replace} pattern.
+          # shellcheck disable=SC2001
+          PROFILE=$(echo "${INPUTS_PROFILE}" | sed 's|\(.*\)[[:space:]](.*)|\1|')
+          echo "profile=${PROFILE}" >> "${GITHUB_OUTPUT}"
 
       - name: Parse hardware name
         id: parse_hardware_name
         run: |
-          echo "name=$(echo '${{ inputs.profile }}' | sed 's|.*[[:space:]](\(.*\))|\1|')" >> "${GITHUB_OUTPUT}"
+          # Use Sed to extract a value from a string, this cannot be done with the ${variable//search/replace} pattern.
+          # shellcheck disable=SC2001
+          NAME=$(echo "${INPUTS_PROFILE}" | sed 's|.*[[:space:]](\(.*\))|\1|')
+          echo "name=${NAME}" >> "${GITHUB_OUTPUT}"
 
   run-benchmarks:
     name: Run benchmarks
@@ -88,4 +100,12 @@ jobs:
       bench_type: ${{ inputs.bench_type }}
       params_type: ${{ inputs.params_type }}
       all_precisions: ${{ inputs.all_precisions }}
-    secrets: inherit
+    secrets:
+      BOT_USERNAME: ${{ secrets.BOT_USERNAME }}
+      SLACK_CHANNEL: ${{ secrets.SLACK_CHANNEL }}
+      SLACK_WEBHOOK: ${{ secrets.SLACK_WEBHOOK }}
+      REPO_CHECKOUT_TOKEN: ${{ secrets.REPO_CHECKOUT_TOKEN }}
+      JOB_SECRET: ${{ secrets.JOB_SECRET }}
+      SLAB_ACTION_TOKEN: ${{ secrets.SLAB_ACTION_TOKEN }}
+      SLAB_URL: ${{ secrets.SLAB_URL }}
+      SLAB_BASE_URL: ${{ secrets.SLAB_BASE_URL }}

.github/workflows/benchmark_gpu_4090.yml

@@ -22,6 +22,9 @@ on:
     # Weekly benchmarks will be triggered each Friday at 9p.m.
     - cron: "0 21 * * 5"
 
+permissions:
+  contents: read
+
 jobs:
   cuda-integer-benchmarks:
     name: Cuda integer benchmarks (RTX 4090)
@@ -69,15 +72,17 @@ jobs:
 
       - name: Parse results
         run: |
-          python3 ./ci/benchmark_parser.py target/criterion ${{ env.RESULTS_FILENAME }} \
+          python3 ./ci/benchmark_parser.py target/criterion "${RESULTS_FILENAME}" \
           --database tfhe_rs \
           --hardware "rtx4090" \
           --backend gpu \
-          --project-version "${{ env.COMMIT_HASH }}" \
-          --branch ${{ github.ref_name }} \
-          --commit-date "${{ env.COMMIT_DATE }}" \
-          --bench-date "${{ env.BENCH_DATE }}" \
+          --project-version "${COMMIT_HASH}" \
+          --branch "${REF_NAME}" \
+          --commit-date "${COMMIT_DATE}" \
+          --bench-date "${BENCH_DATE}" \
           --walk-subdirs
+        env:
+          REF_NAME: ${{ github.ref_name }}
 
       - name: Upload parsed results artifact
         uses: actions/upload-artifact@ea165f8d65b6e75b540449e92b4886f43607fa02
@@ -88,11 +93,11 @@ jobs:
       - name: Send data to Slab
         shell: bash
         run: |
-          python3 slab/scripts/data_sender.py ${{ env.RESULTS_FILENAME }} "${{ secrets.JOB_SECRET }}" \
+          python3 slab/scripts/data_sender.py "${RESULTS_FILENAME}" "${{ secrets.JOB_SECRET }}" \
           --slab-url "${{ secrets.SLAB_URL }}"
 
       - name: Slack Notification
-        if: ${{ failure() }}
+        if: ${{ failure() || (cancelled() && github.event_name != 'pull_request') }}
         continue-on-error: true
         uses: rtCamp/action-slack-notify@e31e87e03dd19038e411e38ae27cbad084a90661
         env:
@@ -145,14 +150,14 @@ jobs:
 
       - name: Parse results
         run: |
-          python3 ./ci/benchmark_parser.py target/criterion ${{ env.RESULTS_FILENAME }} \
+          python3 ./ci/benchmark_parser.py target/criterion "${RESULTS_FILENAME}" \
           --database tfhe_rs \
           --hardware "rtx4090" \
           --backend gpu \
-          --project-version "${{ env.COMMIT_HASH }}" \
-          --branch ${{ github.ref_name }} \
-          --commit-date "${{ env.COMMIT_DATE }}" \
-          --bench-date "${{ env.BENCH_DATE }}" \
+          --project-version "${COMMIT_HASH}" \
+          --branch "${REF_NAME}" \
+          --commit-date "${COMMIT_DATE}" \
+          --bench-date "${BENCH_DATE}" \
           --walk-subdirs \
       
 
@@ -165,19 +170,11 @@ jobs:
       - name: Send data to Slab
         shell: bash
         run: |
-          echo "Computing HMac on results file"
-          SIGNATURE="$(slab/scripts/hmac_calculator.sh ${{ env.RESULTS_FILENAME }} '${{ secrets.JOB_SECRET }}')"
-          echo "Sending results to Slab..."
-          curl -v -k \
-          -H "Content-Type: application/json" \
-          -H "X-Slab-Repository: ${{ github.repository }}" \
-          -H "X-Slab-Command: store_data_v2" \
-          -H "X-Hub-Signature-256: sha256=${SIGNATURE}" \
-          -d @${{ env.RESULTS_FILENAME }} \
-          ${{ secrets.SLAB_URL }}
+          python3 slab/scripts/data_sender.py "${RESULTS_FILENAME}" "${{ secrets.JOB_SECRET }}" \
+          --slab-url "${{ secrets.SLAB_URL }}"
 
       - name: Slack Notification
-        if: ${{ failure() }}
+        if: ${{ failure() || (cancelled() && github.event_name != 'pull_request') }}
         continue-on-error: true
         uses: rtCamp/action-slack-notify@e31e87e03dd19038e411e38ae27cbad084a90661
         env:

.github/workflows/benchmark_gpu_common.yml

@@ -58,6 +58,9 @@ env:
   SLACK_WEBHOOK: ${{ secrets.SLACK_WEBHOOK }}
   FAST_BENCH: TRUE
 
+
+permissions: {}
+
 jobs:
   prepare-matrix:
     name: Prepare operations matrix
@@ -67,44 +70,56 @@ jobs:
       op_flavor: ${{ steps.set_op_flavor.outputs.op_flavor }}
       bench_type: ${{ steps.set_bench_type.outputs.bench_type }}
       params_type: ${{ steps.set_params_type.outputs.params_type }}
+    env:
+      INPUTS_COMMAND: ${{ inputs.command }}
+      INPUTS_OP_FLAVOR: ${{ inputs.op_flavor }}
     steps:
       - name: Set single command
         if: ${{ !contains(inputs.command, ',')}}
         run: |
-          echo "COMMAND=[\"${{ inputs.command }}\"]" >> "${GITHUB_ENV}"
+          echo "COMMAND=[\"${INPUTS_COMMAND}\"]" >> "${GITHUB_ENV}"
 
       - name: Set multiple commands
         if: ${{ contains(inputs.command, ',')}}
         run: |
-          PARSED_COMMAND=$(echo "${{ inputs.command }}" | sed 's/[[:space:]]*,[[:space:]]*/\\", \\"/g')
+          # Use Sed to extract a value from a string, this cannot be done with the ${variable//search/replace} pattern.
+          # shellcheck disable=SC2001
+          PARSED_COMMAND=$(echo "${INPUTS_COMMAND}" | sed 's/[[:space:]]*,[[:space:]]*/\", \"/g')
           echo "COMMAND=[\"${PARSED_COMMAND}\"]" >> "${GITHUB_ENV}"
 
       - name: Set single operations flavor
         if: ${{ !contains(inputs.op_flavor, ',')}}
         run: |
-          echo "OP_FLAVOR=[\"${{ inputs.op_flavor }}\"]" >> "${GITHUB_ENV}"
+          echo "OP_FLAVOR=[\"${INPUTS_OP_FLAVOR}\"]" >> "${GITHUB_ENV}"
 
       - name: Set multiple operations flavors
         if: ${{ contains(inputs.op_flavor, ',')}}
         run: |
-          PARSED_OP_FLAVOR=$(echo "${{ inputs.op_flavor }}" | sed 's/[[:space:]]*,[[:space:]]*/", "/g')
+          # Use Sed to extract a value from a string, this cannot be done with the ${variable//search/replace} pattern.
+          # shellcheck disable=SC2001
+          PARSED_OP_FLAVOR=$(echo "${INPUTS_OP_FLAVOR}" | sed 's/[[:space:]]*,[[:space:]]*/", "/g')
           echo "OP_FLAVOR=[\"${PARSED_OP_FLAVOR}\"]" >> "${GITHUB_ENV}"
 
       - name: Set benchmark types
         run: |
-          if [[ "${{ inputs.bench_type }}" == "both" ]]; then
+          if [[ "${INPUTS_BENCH_TYPE}" == "both" ]]; then
             echo "BENCH_TYPE=[\"latency\", \"throughput\"]" >> "${GITHUB_ENV}"
           else
-            echo "BENCH_TYPE=[\"${{ inputs.bench_type }}\"]" >> "${GITHUB_ENV}"
+            echo "BENCH_TYPE=[\"${INPUTS_BENCH_TYPE}\"]" >> "${GITHUB_ENV}"
           fi
+        env:
+          INPUTS_BENCH_TYPE: ${{ inputs.bench_type }}
 
       - name: Set parameters types
         run: |
-          if [[ "${{ inputs.params_type }}" == "both" ]]; then
+          if [[ "${INPUTS_PARAMS_TYPE}" == "both" ]]; then
             echo "PARAMS_TYPE=[\"classical\", \"multi_bit\"]" >> "${GITHUB_ENV}"
           else
-            echo "PARAMS_TYPE=[\"${{ inputs.params_type }}\"]" >> "${GITHUB_ENV}"
+            echo "PARAMS_TYPE=[\"${INPUTS_PARAMS_TYPE}\"]" >> "${GITHUB_ENV}"
           fi
+        env:
+          INPUTS_PARAMS_TYPE: ${{ inputs.params_type }}
+
 
       - name: Set command output
         id: set_command
@@ -154,9 +169,11 @@ jobs:
         if: steps.start-remote-instance.outcome == 'failure' &&
           inputs.profile != 'single-h100'
         run: |
-          echo "Remote instance instance has failed to start (profile provided: '${{ inputs.profile }}')"
+          echo "Remote instance instance has failed to start (profile provided: '${INPUTS_PROFILE}')"
           echo "Permanent instance instance cannot be used as a substitute (profile needed: 'single-h100')"
           exit 1
+        env:
+          INPUTS_PROFILE: ${{ inputs.profile }}
 
       # This will allow to fallback on permanent instances running on Hyperstack.
       - name: Use permanent remote instance
@@ -198,7 +215,6 @@ jobs:
     needs: [ prepare-matrix, setup-instance, install-dependencies ]
     runs-on: ${{ needs.setup-instance.outputs.runner-name }}
     timeout-minutes: 1440 # 24 hours
-    continue-on-error: true
     strategy:
       fail-fast: false
       max-parallel: 1
@@ -259,21 +275,30 @@ jobs:
 
       - name: Run benchmarks
         run: |
-          make BENCH_OP_FLAVOR=${{ matrix.op_flavor }} BENCH_TYPE=${{ matrix.bench_type }} BENCH_PARAM_TYPE=${{ matrix.params_type }} bench_${{ matrix.command }}_gpu
+          make BENCH_OP_FLAVOR="${OP_FLAVOR}" BENCH_TYPE="${BENCH_TYPE}" BENCH_PARAM_TYPE="${BENCH_PARAMS_TYPE}" bench_"${BENCH_COMMAND}"_gpu
+        env:
+          OP_FLAVOR: ${{ matrix.op_flavor }}
+          BENCH_TYPE: ${{ matrix.bench_type }}
+          BENCH_PARAMS_TYPE: ${{ matrix.params_type }}
+          BENCH_COMMAND: ${{ matrix.command }}
 
       - name: Parse results
         run: |
-          python3 ./ci/benchmark_parser.py target/criterion ${{ env.RESULTS_FILENAME }} \
+          python3 ./ci/benchmark_parser.py target/criterion "${RESULTS_FILENAME}" \
           --database tfhe_rs \
-          --hardware "${{ inputs.hardware_name }}" \
+          --hardware "${INPUTS_HARDWARE_NAME}" \
           --backend gpu \
-          --project-version "${{ env.COMMIT_HASH }}" \
-          --branch ${{ github.ref_name }} \
-          --commit-date "${{ env.COMMIT_DATE }}" \
-          --bench-date "${{ env.BENCH_DATE }}" \
+          --project-version "${COMMIT_HASH}" \
+          --branch "${REF_NAME}" \
+          --commit-date "${COMMIT_DATE}" \
+          --bench-date "${BENCH_DATE}" \
           --walk-subdirs \
           --name-suffix avx512 \
-          --bench-type ${{ matrix.bench_type }}
+          --bench-type "${BENCH_TYPE}"
+        env:
+          INPUTS_HARDWARE_NAME: ${{ inputs.hardware_name }}
+          REF_NAME: ${{ github.ref_name }}
+          BENCH_TYPE: ${{ matrix.bench_type }}
 
       - name: Upload parsed results artifact
         uses: actions/upload-artifact@ea165f8d65b6e75b540449e92b4886f43607fa02
@@ -292,7 +317,7 @@ jobs:
       - name: Send data to Slab
         shell: bash
         run: |
-          python3 slab/scripts/data_sender.py ${{ env.RESULTS_FILENAME }} "${{ secrets.JOB_SECRET }}" \
+          python3 slab/scripts/data_sender.py "${RESULTS_FILENAME}" "${{ secrets.JOB_SECRET }}" \
           --slab-url "${{ secrets.SLAB_URL }}"
 
   slack-notify:

.github/workflows/benchmark_gpu_dex.yml

@@ -0,0 +1,64 @@
+# Run CUDA DEX benchmarks on a Hyperstack VM and return parsed results to Slab CI bot.
+name: Cuda DEX benchmarks
+
+on:
+  workflow_dispatch:
+    inputs:
+      profile:
+        description: "Instance type"
+        required: true
+        type: choice
+        options:
+          - "l40 (n3-L40x1)"
+          - "4-l40 (n3-L40x4)"
+          - "multi-a100-nvlink (n3-A100x8-NVLink)"
+          - "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)"
+
+permissions: {}
+
+jobs:
+  parse-inputs:
+    runs-on: ubuntu-latest
+    outputs:
+      profile: ${{ steps.parse_profile.outputs.profile }}
+      hardware_name: ${{ steps.parse_hardware_name.outputs.name }}
+    env:
+      INPUTS_PROFILE: ${{ inputs.profile }}
+    steps:
+      - name: Parse profile
+        id: parse_profile
+        run: |
+          # Use Sed to extract a value from a string, this cannot be done with the ${variable//search/replace} pattern.
+          # shellcheck disable=SC2001
+          PROFILE=$(echo "${INPUTS_PROFILE}" | sed 's|\(.*\)[[:space:]](.*)|\1|')
+          echo "profile=${PROFILE}" >> "${GITHUB_OUTPUT}"
+
+      - name: Parse hardware name
+        id: parse_hardware_name
+        run: |
+          # Use Sed to extract a value from a string, this cannot be done with the ${variable//search/replace} pattern.
+          # shellcheck disable=SC2001
+          NAME=$(echo "${INPUTS_PROFILE}" | sed 's|.*[[:space:]](\(.*\))|\1|')
+          echo "name=${NAME}" >> "${GITHUB_OUTPUT}"
+
+  run-benchmarks:
+    name: Run benchmarks
+    needs: parse-inputs
+    uses: ./.github/workflows/benchmark_gpu_dex_common.yml
+    with:
+      profile: ${{ needs.parse-inputs.outputs.profile }}
+      hardware_name: ${{ needs.parse-inputs.outputs.hardware_name }}
+    secrets:
+      BOT_USERNAME: ${{ secrets.BOT_USERNAME }}
+      SLACK_CHANNEL: ${{ secrets.SLACK_CHANNEL }}
+      SLACK_WEBHOOK: ${{ secrets.SLACK_WEBHOOK }}
+      REPO_CHECKOUT_TOKEN: ${{ secrets.REPO_CHECKOUT_TOKEN }}
+      JOB_SECRET: ${{ secrets.JOB_SECRET }}
+      SLAB_ACTION_TOKEN: ${{ secrets.SLAB_ACTION_TOKEN }}
+      SLAB_URL: ${{ secrets.SLAB_URL }}
+      SLAB_BASE_URL: ${{ secrets.SLAB_BASE_URL }}

.github/workflows/benchmark_gpu_dex_common.yml

@@ -0,0 +1,208 @@
+# Run DEX benchmarks on an instance with CUDA and return parsed results to Slab CI bot.
+name: Cuda DEX benchmarks - common
+
+on:
+  workflow_call:
+    inputs:
+      backend:
+        type: string
+        default: hyperstack
+      profile:
+        type: string
+        required: true
+      hardware_name:
+        type: string
+        required: true
+    secrets:
+      REPO_CHECKOUT_TOKEN:
+        required: true
+      SLAB_ACTION_TOKEN:
+        required: true
+      SLAB_BASE_URL:
+        required: true
+      SLAB_URL:
+        required: true
+      JOB_SECRET:
+        required: true
+      SLACK_CHANNEL:
+        required: true
+      BOT_USERNAME:
+        required: true
+      SLACK_WEBHOOK:
+        required: true
+
+env:
+  CARGO_TERM_COLOR: always
+  RESULTS_FILENAME: parsed_benchmark_results_${{ github.sha }}.json
+  PARSE_INTEGER_BENCH_CSV_FILE: tfhe_rs_integer_benches_${{ github.sha }}.csv
+  ACTION_RUN_URL: ${{ github.server_url }}/${{ github.repository }}/actions/runs/${{ github.run_id }}
+  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 }}
+
+permissions: {}
+
+jobs:
+  setup-instance:
+    name: Setup instance (cuda-dex-benchmarks)
+    runs-on: ubuntu-latest
+    if:  github.event_name == 'workflow_dispatch' ||
+      (github.event_name == 'schedule' && github.repository == 'zama-ai/tfhe-rs')
+    outputs:
+      # 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 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 }}
+          slab-url: ${{ secrets.SLAB_BASE_URL }}
+          job-secret: ${{ secrets.JOB_SECRET }}
+          backend: ${{ inputs.backend }}
+          profile: ${{ inputs.profile }}
+
+      - name: Acknowledge remote instance failure
+        if: steps.start-remote-instance.outcome == 'failure' &&
+          inputs.profile != 'single-h100'
+        run: |
+          echo "Remote instance instance has failed to start (profile provided: '${INPUTS_PROFILE}')"
+          echo "Permanent instance instance cannot be used as a substitute (profile needed: 'single-h100')"
+          exit 1
+        env:
+          INPUTS_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-dex-benchmarks:
+    name: Cuda DEX benchmarks (${{ inputs.profile }})
+    needs: setup-instance
+    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
+    steps:
+      - name: Checkout tfhe-rs repo with tags
+        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: ${{ matrix.gcc }}
+
+      - name: Get benchmark details
+        run: |
+          {
+            echo "BENCH_DATE=$(date --iso-8601=seconds)";
+            echo "COMMIT_DATE=$(git --no-pager show -s --format=%cd --date=iso8601-strict ${{ github.sha }})";
+            echo "COMMIT_HASH=$(git describe --tags --dirty)";
+          } >> "${GITHUB_ENV}"
+
+      - name: Install rust
+        uses: dtolnay/rust-toolchain@888c2e1ea69ab0d4330cbf0af1ecc7b68f368cc1
+        with:
+          toolchain: nightly
+
+      - name: Run benchmarks
+        run: |
+          make bench_hlapi_dex_gpu
+
+      - name: Parse results
+        run: |
+          python3 ./ci/benchmark_parser.py target/criterion "${RESULTS_FILENAME}" \
+          --database tfhe_rs \
+          --hardware "${INPUTS_HARDWARE_NAME}" \
+          --backend gpu \
+          --project-version "${COMMIT_HASH}" \
+          --branch "${REF_NAME}" \
+          --commit-date "${COMMIT_DATE}" \
+          --bench-date "${BENCH_DATE}" \
+          --walk-subdirs \
+          --name-suffix avx512
+        env:
+          INPUTS_HARDWARE_NAME: ${{ inputs.hardware_name }}
+          REF_NAME: ${{ github.ref_name }}
+
+      - name: Upload parsed results artifact
+        uses: actions/upload-artifact@ea165f8d65b6e75b540449e92b4886f43607fa02
+        with:
+          name: ${{ github.sha }}_dex_${{ 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: ${{ secrets.REPO_CHECKOUT_TOKEN }}
+
+      - name: Send data to Slab
+        shell: bash
+        run: |
+          python3 slab/scripts/data_sender.py "${RESULTS_FILENAME}" "${{ secrets.JOB_SECRET }}" \
+          --slab-url "${{ secrets.SLAB_URL }}"
+
+  slack-notify:
+    name: Slack Notification
+    needs: [ setup-instance, cuda-dex-benchmarks ]
+    runs-on: ubuntu-latest
+    if: ${{ always() && needs.cuda-dex-benchmarks.result != 'skipped' && failure() }}
+    continue-on-error: true
+    steps:
+      - name: Send message
+        uses: rtCamp/action-slack-notify@e31e87e03dd19038e411e38ae27cbad084a90661
+        env:
+          SLACK_COLOR: ${{ needs.cuda-dex-benchmarks.result }}
+          SLACK_MESSAGE: "Cuda DEX benchmarks (${{ inputs.profile }}) finished with status: ${{ needs.cuda-dex-benchmarks.result }}. (${{ env.ACTION_RUN_URL }})"
+
+  teardown-instance:
+    name: Teardown instance (cuda-dex-${{ inputs.profile }}-benchmarks)
+    if: ${{ always() && needs.setup-instance.outputs.remote-instance-outcome == 'success' }}
+    needs: [ setup-instance, cuda-dex-benchmarks, slack-notify ]
+    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@e31e87e03dd19038e411e38ae27cbad084a90661
+        env:
+          SLACK_COLOR: ${{ job.status }}
+          SLACK_MESSAGE: "Instance teardown (cuda-dex-${{ inputs.profile }}-benchmarks) finished with status: ${{ job.status }}. (${{ env.ACTION_RUN_URL }})"

.github/workflows/benchmark_gpu_dex_weekly.yml

@@ -0,0 +1,61 @@
+# Run CUDA DEX benchmarks on multiple Hyperstack VMs and return parsed results to Slab CI bot.
+name: Cuda DEX weekly benchmarks
+
+on:
+  schedule:
+    # Weekly benchmarks will be triggered each Saturday at 9a.m.
+    - cron: '0 9 * * 6'
+
+permissions: {}
+
+jobs:
+  run-benchmarks-1-h100:
+    name: Run benchmarks (1xH100)
+    if: github.repository == 'zama-ai/tfhe-rs'
+    uses: ./.github/workflows/benchmark_gpu_dex_common.yml
+    with:
+      profile: single-h100
+      hardware_name: n3-H100x1
+    secrets:
+      BOT_USERNAME: ${{ secrets.BOT_USERNAME }}
+      SLACK_CHANNEL: ${{ secrets.SLACK_CHANNEL }}
+      SLACK_WEBHOOK: ${{ secrets.SLACK_WEBHOOK }}
+      REPO_CHECKOUT_TOKEN: ${{ secrets.REPO_CHECKOUT_TOKEN }}
+      JOB_SECRET: ${{ secrets.JOB_SECRET }}
+      SLAB_ACTION_TOKEN: ${{ secrets.SLAB_ACTION_TOKEN }}
+      SLAB_URL: ${{ secrets.SLAB_URL }}
+      SLAB_BASE_URL: ${{ secrets.SLAB_BASE_URL }}
+
+  run-benchmarks-2-h100:
+    name: Run benchmarks (2xH100)
+    if: github.repository == 'zama-ai/tfhe-rs'
+    uses: ./.github/workflows/benchmark_gpu_dex_common.yml
+    with:
+      profile: 2-h100
+      hardware_name: n3-H100x2
+    secrets:
+      BOT_USERNAME: ${{ secrets.BOT_USERNAME }}
+      SLACK_CHANNEL: ${{ secrets.SLACK_CHANNEL }}
+      SLACK_WEBHOOK: ${{ secrets.SLACK_WEBHOOK }}
+      REPO_CHECKOUT_TOKEN: ${{ secrets.REPO_CHECKOUT_TOKEN }}
+      JOB_SECRET: ${{ secrets.JOB_SECRET }}
+      SLAB_ACTION_TOKEN: ${{ secrets.SLAB_ACTION_TOKEN }}
+      SLAB_URL: ${{ secrets.SLAB_URL }}
+      SLAB_BASE_URL: ${{ secrets.SLAB_BASE_URL }}
+
+  run-benchmarks-8-h100:
+    name: Run benchmarks (8xH100)
+    if: github.repository == 'zama-ai/tfhe-rs'
+    uses: ./.github/workflows/benchmark_gpu_dex_common.yml
+    with:
+      profile: multi-h100
+      hardware_name: n3-H100x8
+    secrets:
+      BOT_USERNAME: ${{ secrets.BOT_USERNAME }}
+      SLACK_CHANNEL: ${{ secrets.SLACK_CHANNEL }}
+      SLACK_WEBHOOK: ${{ secrets.SLACK_WEBHOOK }}
+      REPO_CHECKOUT_TOKEN: ${{ secrets.REPO_CHECKOUT_TOKEN }}
+      JOB_SECRET: ${{ secrets.JOB_SECRET }}
+      SLAB_ACTION_TOKEN: ${{ secrets.SLAB_ACTION_TOKEN }}
+      SLAB_URL: ${{ secrets.SLAB_URL }}
+      SLAB_BASE_URL: ${{ secrets.SLAB_BASE_URL }}

.github/workflows/benchmark_gpu_erc20.yml

@@ -10,6 +10,8 @@ on:
         type: choice
         options:
           - "l40 (n3-L40x1)"
+          - "4-l40 (n3-L40x4)"
+          - "multi-a100-nvlink (n3-A100x8-NVLink)"
           - "single-h100 (n3-H100x1)"
           - "2-h100 (n3-H100x2)"
           - "4-h100 (n3-H100x4)"
@@ -17,22 +19,33 @@ on:
           - "multi-h100-nvlink (n3-H100x8-NVLink)"
           - "multi-h100-sxm5 (n3-H100x8-SXM5)"
 
+
+permissions: {}
+
 jobs:
   parse-inputs:
     runs-on: ubuntu-latest
     outputs:
       profile: ${{ steps.parse_profile.outputs.profile }}
       hardware_name: ${{ steps.parse_hardware_name.outputs.name }}
+    env:
+      INPUTS_PROFILE: ${{ inputs.profile }}
     steps:
       - name: Parse profile
         id: parse_profile
         run: |
-          echo "profile=$(echo '${{ inputs.profile }}' | sed 's|\(.*\)[[:space:]](.*)|\1|')" >> "${GITHUB_OUTPUT}"
+          # Use Sed to extract a value from a string, this cannot be done with the ${variable//search/replace} pattern.
+          # shellcheck disable=SC2001
+          PROFILE=$(echo "${INPUTS_PROFILE}" | sed 's|\(.*\)[[:space:]](.*)|\1|')
+          echo "profile=${PROFILE}" >> "${GITHUB_OUTPUT}"
 
       - name: Parse hardware name
         id: parse_hardware_name
         run: |
-          echo "name=$(echo '${{ inputs.profile }}' | sed 's|.*[[:space:]](\(.*\))|\1|')" >> "${GITHUB_OUTPUT}"
+          # Use Sed to extract a value from a string, this cannot be done with the ${variable//search/replace} pattern.
+          # shellcheck disable=SC2001
+          NAME=$(echo "${INPUTS_PROFILE}" | sed 's|.*[[:space:]](\(.*\))|\1|')
+          echo "name=${NAME}" >> "${GITHUB_OUTPUT}"
 
   run-benchmarks:
     name: Run benchmarks
@@ -41,4 +54,12 @@ jobs:
     with:
       profile: ${{ needs.parse-inputs.outputs.profile }}
       hardware_name: ${{ needs.parse-inputs.outputs.hardware_name }}
-    secrets: inherit
+    secrets:
+      BOT_USERNAME: ${{ secrets.BOT_USERNAME }}
+      SLACK_CHANNEL: ${{ secrets.SLACK_CHANNEL }}
+      SLACK_WEBHOOK: ${{ secrets.SLACK_WEBHOOK }}
+      REPO_CHECKOUT_TOKEN: ${{ secrets.REPO_CHECKOUT_TOKEN }}
+      JOB_SECRET: ${{ secrets.JOB_SECRET }}
+      SLAB_ACTION_TOKEN: ${{ secrets.SLAB_ACTION_TOKEN }}
+      SLAB_URL: ${{ secrets.SLAB_URL }}
+      SLAB_BASE_URL: ${{ secrets.SLAB_BASE_URL }}

.github/workflows/benchmark_gpu_erc20_common.yml

@@ -43,6 +43,9 @@ env:
   SLACK_USERNAME: ${{ secrets.BOT_USERNAME }}
   SLACK_WEBHOOK: ${{ secrets.SLACK_WEBHOOK }}
 
+
+permissions: {}
+
 jobs:
   setup-instance:
     name: Setup instance (cuda-erc20-benchmarks)
@@ -73,9 +76,11 @@ jobs:
         if: steps.start-remote-instance.outcome == 'failure' &&
           inputs.profile != 'single-h100'
         run: |
-          echo "Remote instance instance has failed to start (profile provided: '${{ inputs.profile }}')"
+          echo "Remote instance instance has failed to start (profile provided: '${INPUTS_PROFILE}')"
           echo "Permanent instance instance cannot be used as a substitute (profile needed: 'single-h100')"
           exit 1
+        env:
+          INPUTS_PROFILE: ${{ inputs.profile }}
 
       # This will allow to fallback on permanent instances running on Hyperstack.
       - name: Use permanent remote instance
@@ -132,16 +137,19 @@ jobs:
 
       - name: Parse results
         run: |
-          python3 ./ci/benchmark_parser.py target/criterion ${{ env.RESULTS_FILENAME }} \
+          python3 ./ci/benchmark_parser.py target/criterion "${RESULTS_FILENAME}" \
           --database tfhe_rs \
-          --hardware "${{ inputs.hardware_name }}" \
+          --hardware "${INPUTS_HARDWARE_NAME}" \
           --backend gpu \
-          --project-version "${{ env.COMMIT_HASH }}" \
-          --branch ${{ github.ref_name }} \
-          --commit-date "${{ env.COMMIT_DATE }}" \
-          --bench-date "${{ env.BENCH_DATE }}" \
+          --project-version "${COMMIT_HASH}" \
+          --branch "${REF_NAME}" \
+          --commit-date "${COMMIT_DATE}" \
+          --bench-date "${BENCH_DATE}" \
           --walk-subdirs \
           --name-suffix avx512
+        env:
+          INPUTS_HARDWARE_NAME: ${{ inputs.hardware_name }}
+          REF_NAME: ${{ github.ref_name }}
 
       - name: Upload parsed results artifact
         uses: actions/upload-artifact@ea165f8d65b6e75b540449e92b4886f43607fa02
@@ -160,7 +168,7 @@ jobs:
       - name: Send data to Slab
         shell: bash
         run: |
-          python3 slab/scripts/data_sender.py ${{ env.RESULTS_FILENAME }} "${{ secrets.JOB_SECRET }}" \
+          python3 slab/scripts/data_sender.py "${RESULTS_FILENAME}" "${{ secrets.JOB_SECRET }}" \
           --slab-url "${{ secrets.SLAB_URL }}"
 
   slack-notify:

.github/workflows/benchmark_gpu_erc20_weekly.yml

@@ -6,6 +6,9 @@ on:
     # Weekly benchmarks will be triggered each Saturday at 5a.m.
     - cron: '0 5 * * 6'
 
+
+permissions: {}
+
 jobs:
   run-benchmarks-1-h100:
     name: Run benchmarks (1xH100)
@@ -14,7 +17,15 @@ jobs:
     with:
       profile: single-h100
       hardware_name: n3-H100x1
-    secrets: inherit
+    secrets:
+      BOT_USERNAME: ${{ secrets.BOT_USERNAME }}
+      SLACK_CHANNEL: ${{ secrets.SLACK_CHANNEL }}
+      SLACK_WEBHOOK: ${{ secrets.SLACK_WEBHOOK }}
+      REPO_CHECKOUT_TOKEN: ${{ secrets.REPO_CHECKOUT_TOKEN }}
+      JOB_SECRET: ${{ secrets.JOB_SECRET }}
+      SLAB_ACTION_TOKEN: ${{ secrets.SLAB_ACTION_TOKEN }}
+      SLAB_URL: ${{ secrets.SLAB_URL }}
+      SLAB_BASE_URL: ${{ secrets.SLAB_BASE_URL }}
 
   run-benchmarks-2-h100:
     name: Run benchmarks (2xH100)
@@ -23,7 +34,15 @@ jobs:
     with:
       profile: 2-h100
       hardware_name: n3-H100x2
-    secrets: inherit
+    secrets:
+      BOT_USERNAME: ${{ secrets.BOT_USERNAME }}
+      SLACK_CHANNEL: ${{ secrets.SLACK_CHANNEL }}
+      SLACK_WEBHOOK: ${{ secrets.SLACK_WEBHOOK }}
+      REPO_CHECKOUT_TOKEN: ${{ secrets.REPO_CHECKOUT_TOKEN }}
+      JOB_SECRET: ${{ secrets.JOB_SECRET }}
+      SLAB_ACTION_TOKEN: ${{ secrets.SLAB_ACTION_TOKEN }}
+      SLAB_URL: ${{ secrets.SLAB_URL }}
+      SLAB_BASE_URL: ${{ secrets.SLAB_BASE_URL }}
 
   run-benchmarks-8-h100:
     name: Run benchmarks (8xH100)
@@ -32,4 +51,12 @@ jobs:
     with:
       profile: multi-h100
       hardware_name: n3-H100x8
-    secrets: inherit
+    secrets:
+      BOT_USERNAME: ${{ secrets.BOT_USERNAME }}
+      SLACK_CHANNEL: ${{ secrets.SLACK_CHANNEL }}
+      SLACK_WEBHOOK: ${{ secrets.SLACK_WEBHOOK }}
+      REPO_CHECKOUT_TOKEN: ${{ secrets.REPO_CHECKOUT_TOKEN }}
+      JOB_SECRET: ${{ secrets.JOB_SECRET }}
+      SLAB_ACTION_TOKEN: ${{ secrets.SLAB_ACTION_TOKEN }}
+      SLAB_URL: ${{ secrets.SLAB_URL }}
+      SLAB_BASE_URL: ${{ secrets.SLAB_BASE_URL }}

.github/workflows/benchmark_gpu_weekly.yml

@@ -6,6 +6,9 @@ on:
     # Weekly benchmarks will be triggered each Saturday at 1a.m.
     - cron: '0 1 * * 6'
 
+
+permissions: {}
+
 jobs:
   run-benchmarks-1-h100:
     name: Run integer benchmarks (1xH100)
@@ -18,7 +21,15 @@ jobs:
       op_flavor: default
       bench_type: latency
       all_precisions: true
-    secrets: inherit
+    secrets:
+      BOT_USERNAME: ${{ secrets.BOT_USERNAME }}
+      SLACK_CHANNEL: ${{ secrets.SLACK_CHANNEL }}
+      SLACK_WEBHOOK: ${{ secrets.SLACK_WEBHOOK }}
+      REPO_CHECKOUT_TOKEN: ${{ secrets.REPO_CHECKOUT_TOKEN }}
+      JOB_SECRET: ${{ secrets.JOB_SECRET }}
+      SLAB_ACTION_TOKEN: ${{ secrets.SLAB_ACTION_TOKEN }}
+      SLAB_URL: ${{ secrets.SLAB_URL }}
+      SLAB_BASE_URL: ${{ secrets.SLAB_BASE_URL }}
 
   run-benchmarks-2-h100:
     name: Run integer benchmarks (2xH100)
@@ -31,7 +42,15 @@ jobs:
       op_flavor: default
       bench_type: latency
       all_precisions: true
-    secrets: inherit
+    secrets:
+      BOT_USERNAME: ${{ secrets.BOT_USERNAME }}
+      SLACK_CHANNEL: ${{ secrets.SLACK_CHANNEL }}
+      SLACK_WEBHOOK: ${{ secrets.SLACK_WEBHOOK }}
+      REPO_CHECKOUT_TOKEN: ${{ secrets.REPO_CHECKOUT_TOKEN }}
+      JOB_SECRET: ${{ secrets.JOB_SECRET }}
+      SLAB_ACTION_TOKEN: ${{ secrets.SLAB_ACTION_TOKEN }}
+      SLAB_URL: ${{ secrets.SLAB_URL }}
+      SLAB_BASE_URL: ${{ secrets.SLAB_BASE_URL }}
 
   run-benchmarks-8-h100:
     name: Run integer benchmarks (8xH100)
@@ -44,7 +63,15 @@ jobs:
       op_flavor: default
       bench_type: latency
       all_precisions: true
-    secrets: inherit
+    secrets:
+      BOT_USERNAME: ${{ secrets.BOT_USERNAME }}
+      SLACK_CHANNEL: ${{ secrets.SLACK_CHANNEL }}
+      SLACK_WEBHOOK: ${{ secrets.SLACK_WEBHOOK }}
+      REPO_CHECKOUT_TOKEN: ${{ secrets.REPO_CHECKOUT_TOKEN }}
+      JOB_SECRET: ${{ secrets.JOB_SECRET }}
+      SLAB_ACTION_TOKEN: ${{ secrets.SLAB_ACTION_TOKEN }}
+      SLAB_URL: ${{ secrets.SLAB_URL }}
+      SLAB_BASE_URL: ${{ secrets.SLAB_BASE_URL }}
 
   run-benchmarks-l40:
     name: Run integer benchmarks (L40)
@@ -57,7 +84,15 @@ jobs:
       op_flavor: default
       bench_type: latency
       all_precisions: true
-    secrets: inherit
+    secrets:
+      BOT_USERNAME: ${{ secrets.BOT_USERNAME }}
+      SLACK_CHANNEL: ${{ secrets.SLACK_CHANNEL }}
+      SLACK_WEBHOOK: ${{ secrets.SLACK_WEBHOOK }}
+      REPO_CHECKOUT_TOKEN: ${{ secrets.REPO_CHECKOUT_TOKEN }}
+      JOB_SECRET: ${{ secrets.JOB_SECRET }}
+      SLAB_ACTION_TOKEN: ${{ secrets.SLAB_ACTION_TOKEN }}
+      SLAB_URL: ${{ secrets.SLAB_URL }}
+      SLAB_BASE_URL: ${{ secrets.SLAB_BASE_URL }}
 
   run-benchmarks-1-h100-core-crypto:
     name: Run core-crypto benchmarks (1xH100)
@@ -68,4 +103,12 @@ jobs:
       hardware_name: n3-H100x1
       command: pbs,pbs128,ks,ks_pbs
       bench_type: latency
-    secrets: inherit
+    secrets:
+      BOT_USERNAME: ${{ secrets.BOT_USERNAME }}
+      SLACK_CHANNEL: ${{ secrets.SLACK_CHANNEL }}
+      SLACK_WEBHOOK: ${{ secrets.SLACK_WEBHOOK }}
+      REPO_CHECKOUT_TOKEN: ${{ secrets.REPO_CHECKOUT_TOKEN }}
+      JOB_SECRET: ${{ secrets.JOB_SECRET }}
+      SLAB_ACTION_TOKEN: ${{ secrets.SLAB_ACTION_TOKEN }}
+      SLAB_URL: ${{ secrets.SLAB_URL }}
+      SLAB_BASE_URL: ${{ secrets.SLAB_BASE_URL }}

.github/workflows/benchmark_hpu_integer.yml

@@ -0,0 +1,88 @@
+# Run all integer benchmarks on a permanent HPU instance and return parsed results to Slab CI bot.
+name: Hpu Integer Benchmarks
+
+on:
+  workflow_dispatch:
+
+env:
+  CARGO_TERM_COLOR: always
+  RESULTS_FILENAME: parsed_benchmark_results_${{ github.sha }}.json
+  ACTION_RUN_URL: ${{ github.server_url }}/${{ github.repository }}/actions/runs/${{ github.run_id }}
+  RUST_BACKTRACE: "full"
+  RUST_MIN_STACK: "8388608"
+
+permissions: {}
+
+jobs:
+  integer-benchmarks-hpu:
+    name: Execute integer & erc20 benchmarks for HPU backend
+    runs-on: v80-desktop
+    concurrency:
+      group: ${{ github.workflow }}_${{ github.ref }}
+      cancel-in-progress: ${{ github.ref != 'refs/heads/main' }}
+    timeout-minutes: 1440  # 24 hours
+    steps:
+      # Needed as long as hw_regmap repository is private
+      - name: Configure SSH
+        uses: webfactory/ssh-agent@a6f90b1f127823b31d4d4a8d96047790581349bd # v0.9.1
+        with:
+          ssh-private-key: ${{ secrets.SSH_PRIVATE_KEY }}
+
+      - name: Checkout tfhe-rs repo with tags
+        uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
+        with:
+          fetch-depth: 0
+          persist-credentials: 'false'
+          token: ${{ secrets.REPO_CHECKOUT_TOKEN }}
+
+      - name: Get benchmark details
+        run: |
+          {
+            echo "BENCH_DATE=$(date --iso-8601=seconds)";
+            echo "COMMIT_DATE=$(git --no-pager show -s --format=%cd --date=iso8601-strict ${{ github.sha }})";
+            echo "COMMIT_HASH=$(git describe --tags --dirty)";
+          } >> "${GITHUB_ENV}"
+
+      - name: Install rust
+        uses: dtolnay/rust-toolchain@888c2e1ea69ab0d4330cbf0af1ecc7b68f368cc1
+        with:
+          toolchain: nightly
+
+      - name: Checkout Slab repo
+        uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
+        with:
+          repository: zama-ai/slab
+          path: slab
+          persist-credentials: 'false'
+          token: ${{ secrets.REPO_CHECKOUT_TOKEN }}
+
+      - name: Run benchmarks
+        run: |
+          make bench_integer_hpu
+          make bench_hlapi_erc20_hpu
+
+      - name: Parse results
+        run: |
+          python3 ./ci/benchmark_parser.py target/criterion "${RESULTS_FILENAME}" \
+          --database tfhe_rs \
+          --hardware "hpu_x1" \
+          --backend hpu \
+          --project-version "${COMMIT_HASH}" \
+          --branch "${REF_NAME}" \
+          --commit-date "${COMMIT_DATE}" \
+          --bench-date "${BENCH_DATE}" \
+          --walk-subdirs
+        env:
+          REF_NAME: ${{ github.ref_name }}
+
+      - name: Upload parsed results artifact
+        uses: actions/upload-artifact@ea165f8d65b6e75b540449e92b4886f43607fa02
+        with:
+          name: ${{ github.sha }}_integer_benchmarks
+          path: ${{ env.RESULTS_FILENAME }}
+
+      - name: Send data to Slab
+        shell: bash
+        run: |
+          python3 slab/scripts/data_sender.py "${RESULTS_FILENAME}" "${{ secrets.JOB_SECRET }}" \
+          --slab-url "${{ secrets.SLAB_URL }}"

.github/workflows/benchmark_integer.yml

@@ -36,6 +36,9 @@ env:
   SLACK_WEBHOOK: ${{ secrets.SLACK_WEBHOOK }}
   FAST_BENCH: TRUE
 
+
+permissions: {}
+
 jobs:
   prepare-matrix:
     name: Prepare operations matrix
@@ -60,11 +63,13 @@ jobs:
         if: github.event_name == 'workflow_dispatch'
         run: |
           echo "OP_FLAVOR=[\"default\"]" >> "${GITHUB_ENV}"
-          if [[ "${{ inputs.bench_type }}" == "both" ]]; then
+          if [[ "${INPUTS_BENCH_TYPE}" == "both" ]]; then
             echo "BENCH_TYPE=[\"latency\", \"throughput\"]" >> "${GITHUB_ENV}"
           else
-            echo "BENCH_TYPE=[\"${{ inputs.bench_type }}\"]" >> "${GITHUB_ENV}"
+            echo "BENCH_TYPE=[\"${INPUTS_BENCH_TYPE}\"]" >> "${GITHUB_ENV}"
           fi
+        env:
+          INPUTS_BENCH_TYPE: ${{ inputs.bench_type }}
 
       - name: Default benchmark type
         if: github.event_name != 'workflow_dispatch'
@@ -106,7 +111,6 @@ jobs:
     concurrency:
       group: ${{ github.workflow_ref }}
       cancel-in-progress: ${{ github.ref != 'refs/heads/main' }}
-    continue-on-error: true
     timeout-minutes: 1440  # 24 hours
     strategy:
       max-parallel: 1
@@ -150,26 +154,35 @@ jobs:
 
       - name: Run benchmarks with AVX512
         run: |
-          make BENCH_OP_FLAVOR=${{ matrix.op_flavor }} BENCH_TYPE=${{ matrix.bench_type }} bench_${{ matrix.command }}
+          make BENCH_OP_FLAVOR="${OP_FLAVOR}" BENCH_TYPE="${BENCH_TYPE}" bench_"${BENCH_COMMAND}"
+        env:
+          OP_FLAVOR: ${{ matrix.op_flavor }}
+          BENCH_TYPE: ${{ matrix.bench_type }}
+          BENCH_COMMAND: ${{ matrix.command }}
 
       # Run these benchmarks only once per benchmark type
       - name: Run compression benchmarks with AVX512
         if: matrix.op_flavor == 'default' && matrix.command == 'integer'
         run: |
-          make BENCH_TYPE=${{ matrix.bench_type }} bench_integer_compression
+          make BENCH_TYPE="${BENCH_TYPE}" bench_integer_compression
+        env:
+          BENCH_TYPE: ${{ matrix.bench_type }}
 
       - name: Parse results
         run: |
-          python3 ./ci/benchmark_parser.py target/criterion ${{ env.RESULTS_FILENAME }} \
+          python3 ./ci/benchmark_parser.py target/criterion "${RESULTS_FILENAME}" \
           --database tfhe_rs \
           --hardware "hpc7a.96xlarge" \
-          --project-version "${{ env.COMMIT_HASH }}" \
-          --branch ${{ github.ref_name }} \
-          --commit-date "${{ env.COMMIT_DATE }}" \
-          --bench-date "${{ env.BENCH_DATE }}" \
+          --project-version "${COMMIT_HASH}" \
+          --branch "${REF_NAME}" \
+          --commit-date "${COMMIT_DATE}" \
+          --bench-date "${BENCH_DATE}" \
           --walk-subdirs \
           --name-suffix avx512 \
-          --bench-type ${{ matrix.bench_type }}
+          --bench-type "${BENCH_TYPE}"
+        env:
+          REF_NAME: ${{ github.ref_name }}
+          BENCH_TYPE: ${{ matrix.bench_type }}
 
       - name: Upload parsed results artifact
         uses: actions/upload-artifact@ea165f8d65b6e75b540449e92b4886f43607fa02
@@ -180,11 +193,11 @@ jobs:
       - name: Send data to Slab
         shell: bash
         run: |
-          python3 slab/scripts/data_sender.py ${{ env.RESULTS_FILENAME }} "${{ secrets.JOB_SECRET }}" \
+          python3 slab/scripts/data_sender.py "${RESULTS_FILENAME}" "${{ secrets.JOB_SECRET }}" \
           --slab-url "${{ secrets.SLAB_URL }}"
 
       - name: Slack Notification
-        if: ${{ failure() }}
+        if: ${{ failure() || (cancelled() && github.event_name != 'pull_request') }}
         continue-on-error: true
         uses: rtCamp/action-slack-notify@e31e87e03dd19038e411e38ae27cbad084a90661
         env:

.github/workflows/benchmark_shortint.yml

@@ -22,6 +22,9 @@ env:
   SLACK_USERNAME: ${{ secrets.BOT_USERNAME }}
   SLACK_WEBHOOK: ${{ secrets.SLACK_WEBHOOK }}
 
+
+permissions: {}
+
 jobs:
   prepare-matrix:
     name: Prepare operations matrix
@@ -72,7 +75,6 @@ jobs:
     concurrency:
       group: ${{ github.workflow_ref }}
       cancel-in-progress: ${{ github.ref != 'refs/heads/main' }}
-    continue-on-error: true
     strategy:
       max-parallel: 1
       matrix:
@@ -108,21 +110,23 @@ jobs:
 
       - name: Run benchmarks with AVX512
         run: |
-          make BENCH_OP_FLAVOR=${{ matrix.op_flavor }} bench_shortint
+          make BENCH_OP_FLAVOR="${OP_FLAVOR}" bench_shortint
+        env:
+          OP_FLAVOR: ${{ matrix.op_flavor }}
 
       - name: Parse results
         run: |
-          COMMIT_DATE="$(git --no-pager show -s --format=%cd --date=iso8601-strict ${{ github.sha }})"
-          COMMIT_HASH="$(git describe --tags --dirty)"
-          python3 ./ci/benchmark_parser.py target/criterion ${{ env.RESULTS_FILENAME }} \
+          python3 ./ci/benchmark_parser.py target/criterion "${RESULTS_FILENAME}" \
           --database tfhe_rs \
           --hardware "hpc7a.96xlarge" \
           --project-version "${COMMIT_HASH}" \
-          --branch ${{ github.ref_name }} \
+          --branch "${REF_NAME}" \
           --commit-date "${COMMIT_DATE}" \
-          --bench-date "${{ env.BENCH_DATE }}" \
+          --bench-date "${BENCH_DATE}" \
           --walk-subdirs \
           --name-suffix avx512
+        env:
+          REF_NAME: ${{ github.ref_name }}
 
       # This small benchmark needs to be executed only once.
       - name: Measure key sizes
@@ -133,7 +137,7 @@ jobs:
       - name: Parse key sizes results
         if: matrix.op_flavor == 'default'
         run: |
-          python3 ./ci/benchmark_parser.py tfhe/shortint_key_sizes.csv ${{ env.RESULTS_FILENAME }} \
+          python3 ./ci/benchmark_parser.py tfhe-benchmark/shortint_key_sizes.csv "${RESULTS_FILENAME}" \
           --object-sizes \
           --append-results
 
@@ -146,11 +150,11 @@ jobs:
       - name: Send data to Slab
         shell: bash
         run: |
-          python3 slab/scripts/data_sender.py ${{ env.RESULTS_FILENAME }} "${{ secrets.JOB_SECRET }}" \
+          python3 slab/scripts/data_sender.py "${RESULTS_FILENAME}" "${{ secrets.JOB_SECRET }}" \
           --slab-url "${{ secrets.SLAB_URL }}"
 
       - name: Slack Notification
-        if: ${{ failure() }}
+        if: ${{ failure() || (cancelled() && github.event_name != 'pull_request') }}
         continue-on-error: true
         uses: rtCamp/action-slack-notify@e31e87e03dd19038e411e38ae27cbad084a90661
         env:

.github/workflows/benchmark_signed_integer.yml

@@ -36,6 +36,9 @@ env:
   SLACK_WEBHOOK: ${{ secrets.SLACK_WEBHOOK }}
   FAST_BENCH: TRUE
 
+
+permissions: {}
+
 jobs:
   prepare-matrix:
     name: Prepare operations matrix
@@ -60,11 +63,13 @@ jobs:
         if: github.event_name == 'workflow_dispatch'
         run: |
           echo "OP_FLAVOR=[\"default\"]" >> "${GITHUB_ENV}"
-          if [[ "${{ inputs.bench_type }}" == "both" ]]; then
+          if [[ "${INPUTS_BENCH_TYPE}" == "both" ]]; then
             echo "BENCH_TYPE=[\"latency\", \"throughput\"]" >> "${GITHUB_ENV}"
           else
-            echo "BENCH_TYPE=[\"${{ inputs.bench_type }}\"]" >> "${GITHUB_ENV}"
+            echo "BENCH_TYPE=[\"${INPUTS_BENCH_TYPE}\"]" >> "${GITHUB_ENV}"
           fi
+        env:
+          INPUTS_BENCH_TYPE: ${{ inputs.bench_type }}
 
       - name: Default benchmark type
         if: github.event_name != 'workflow_dispatch'
@@ -106,7 +111,6 @@ jobs:
     concurrency:
       group: ${{ github.workflow_ref }}
       cancel-in-progress: ${{ github.ref != 'refs/heads/main' }}
-    continue-on-error: true
     timeout-minutes: 1440  # 24 hours
     strategy:
       max-parallel: 1
@@ -150,20 +154,27 @@ jobs:
 
       - name: Run benchmarks with AVX512
         run: |
-          make BENCH_OP_FLAVOR=${{ matrix.op_flavor }} BENCH_TYPE=${{ matrix.bench_type }} bench_signed_${{ matrix.command }}
+          make BENCH_OP_FLAVOR="${OP_FLAVOR}" BENCH_TYPE="${BENCH_TYPE}" bench_signed_"${BENCH_COMMAND}"
+        env:
+          OP_FLAVOR: ${{ matrix.op_flavor }}
+          BENCH_TYPE: ${{ matrix.bench_type }}
+          BENCH_COMMAND: ${{ matrix.command }}
 
       - name: Parse results
         run: |
-          python3 ./ci/benchmark_parser.py target/criterion ${{ env.RESULTS_FILENAME }} \
+          python3 ./ci/benchmark_parser.py target/criterion "${RESULTS_FILENAME}" \
           --database tfhe_rs \
           --hardware "hpc7a.96xlarge" \
-          --project-version "${{ env.COMMIT_HASH }}" \
-          --branch ${{ github.ref_name }} \
-          --commit-date "${{ env.COMMIT_DATE }}" \
-          --bench-date "${{ env.BENCH_DATE }}" \
+          --project-version "${COMMIT_HASH}" \
+          --branch "${REF_NAME}" \
+          --commit-date "${COMMIT_DATE}" \
+          --bench-date "${BENCH_DATE}" \
           --walk-subdirs \
           --name-suffix avx512 \
-          --bench-type ${{ matrix.bench_type }}
+          --bench-type "${BENCH_TYPE}"
+        env:
+          REF_NAME: ${{ github.ref_name }}
+          BENCH_TYPE: ${{ matrix.bench_type }}
 
       - name: Upload parsed results artifact
         uses: actions/upload-artifact@ea165f8d65b6e75b540449e92b4886f43607fa02
@@ -174,11 +185,11 @@ jobs:
       - name: Send data to Slab
         shell: bash
         run: |
-          python3 slab/scripts/data_sender.py ${{ env.RESULTS_FILENAME }} "${{ secrets.JOB_SECRET }}" \
+          python3 slab/scripts/data_sender.py "${RESULTS_FILENAME}" "${{ secrets.JOB_SECRET }}" \
           --slab-url "${{ secrets.SLAB_URL }}"
 
       - name: Slack Notification
-        if: ${{ failure() }}
+        if: ${{ failure() || (cancelled() && github.event_name != 'pull_request') }}
         continue-on-error: true
         uses: rtCamp/action-slack-notify@e31e87e03dd19038e411e38ae27cbad084a90661
         env:

.github/workflows/benchmark_tfhe_fft.yml

@@ -23,6 +23,9 @@ on:
     # Job will be triggered each Thursday at 11p.m.
     - cron: '0 23 * * 4'
 
+
+permissions: {}
+
 jobs:
   setup-ec2:
     name: Setup EC2 instance (fft-benchmarks)
@@ -53,6 +56,8 @@ jobs:
         uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
         with:
           fetch-depth: 0
+          persist-credentials: 'false'
+          token: ${{ secrets.REPO_CHECKOUT_TOKEN }}
 
       - name: Get benchmark details
         run: |
@@ -74,14 +79,16 @@ jobs:
 
       - name: Parse AVX512 results
         run: |
-          python3 ./ci/fft_benchmark_parser.py target/criterion ${{ env.RESULTS_FILENAME }} \
+          python3 ./ci/fft_benchmark_parser.py target/criterion "${RESULTS_FILENAME}" \
           --database concrete_fft \
           --hardware "hpc7a.96xlarge" \
-          --project-version "${{ env.COMMIT_HASH }}" \
-          --branch ${{ github.ref_name }} \
-          --commit-date "${{ env.COMMIT_DATE }}" \
-          --bench-date "${{ env.BENCH_DATE }}" \
+          --project-version "${COMMIT_HASH}" \
+          --branch "${REF_NAME}" \
+          --commit-date "${COMMIT_DATE}" \
+          --bench-date "${BENCH_DATE}" \
           --name-suffix avx512
+        env:
+          REF_NAME: ${{ github.ref_name }}
 
       - name: Upload parsed results artifact
         uses: actions/upload-artifact@ea165f8d65b6e75b540449e92b4886f43607fa02
@@ -100,19 +107,11 @@ jobs:
       - name: Send data to Slab
         shell: bash
         run: |
-          echo "Computing HMac on downloaded artifact"
-          SIGNATURE="$(slab/scripts/hmac_calculator.sh ${{ env.RESULTS_FILENAME }} '${{ secrets.JOB_SECRET }}')"
-          echo "Sending results to Slab..."
-          curl -v -k \
-          -H "Content-Type: application/json" \
-          -H "X-Slab-Repository: ${{ github.repository }}" \
-          -H "X-Slab-Command: store_data_v2" \
-          -H "X-Hub-Signature-256: sha256=${SIGNATURE}" \
-          -d @${{ env.RESULTS_FILENAME }} \
-          ${{ secrets.SLAB_URL }}
+          python3 slab/scripts/data_sender.py "${RESULTS_FILENAME}" "${{ secrets.JOB_SECRET }}" \
+          --slab-url "${{ secrets.SLAB_URL }}"
 
       - name: Slack Notification
-        if: ${{ failure() }}
+        if: ${{ failure() || (cancelled() && github.event_name != 'pull_request') }}
         continue-on-error: true
         uses: rtCamp/action-slack-notify@e31e87e03dd19038e411e38ae27cbad084a90661
         env:

.github/workflows/benchmark_tfhe_ntt.yml

@@ -23,6 +23,9 @@ on:
     # Job will be triggered each Friday at 11p.m.
     - cron: "0 23 * * 5"
 
+
+permissions: {}
+
 jobs:
   setup-ec2:
     name: Setup EC2 instance (ntt-benchmarks)
@@ -53,6 +56,8 @@ jobs:
         uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
         with:
           fetch-depth: 0
+          persist-credentials: 'false'
+          token: ${{ secrets.REPO_CHECKOUT_TOKEN }}
 
       - name: Get benchmark details
         run: |
@@ -74,14 +79,16 @@ jobs:
 
       - name: Parse results
         run: |
-          python3 ./ci/ntt_benchmark_parser.py target/criterion ${{ env.RESULTS_FILENAME }} \
+          python3 ./ci/ntt_benchmark_parser.py target/criterion "${RESULTS_FILENAME}" \
           --database concrete_ntt \
           --hardware "hpc7a.96xlarge" \
-          --project-version "${{ env.COMMIT_HASH }}" \
-          --branch ${{ github.ref_name }} \
-          --commit-date "${{ env.COMMIT_DATE }}" \
-          --bench-date "${{ env.BENCH_DATE }}" \
+          --project-version "${COMMIT_HASH}" \
+          --branch "${REF_NAME}" \
+          --commit-date "${COMMIT_DATE}" \
+          --bench-date "${BENCH_DATE}" \
           --name-suffix avx512
+        env:
+          REF_NAME: ${{ github.ref_name }}
 
       - name: Upload parsed results artifact
         uses: actions/upload-artifact@ea165f8d65b6e75b540449e92b4886f43607fa02
@@ -100,19 +107,11 @@ jobs:
       - name: Send data to Slab
         shell: bash
         run: |
-          echo "Computing HMac on downloaded artifact"
-          SIGNATURE="$(slab/scripts/hmac_calculator.sh ${{ env.RESULTS_FILENAME }} '${{ secrets.JOB_SECRET }}')"
-          echo "Sending results to Slab..."
-          curl -v -k \
-          -H "Content-Type: application/json" \
-          -H "X-Slab-Repository: ${{ github.repository }}" \
-          -H "X-Slab-Command: store_data_v2" \
-          -H "X-Hub-Signature-256: sha256=${SIGNATURE}" \
-          -d @${{ env.RESULTS_FILENAME }} \
-          ${{ secrets.SLAB_URL }}
+          python3 slab/scripts/data_sender.py "${RESULTS_FILENAME}" "${{ secrets.JOB_SECRET }}" \
+          --slab-url "${{ secrets.SLAB_URL }}"
 
       - name: Slack Notification
-        if: ${{ failure() }}
+        if: ${{ failure() || (cancelled() && github.event_name != 'pull_request') }}
         continue-on-error: true
         uses: rtCamp/action-slack-notify@e31e87e03dd19038e411e38ae27cbad084a90661
         env:

.github/workflows/benchmark_tfhe_zk_pok.yml

@@ -30,6 +30,9 @@ env:
   SLACK_WEBHOOK: ${{ secrets.SLACK_WEBHOOK }}
   BENCH_TYPE: ${{ inputs.bench_type || 'latency' }}
 
+
+permissions: {}
+
 jobs:
   should-run:
     runs-on: ubuntu-latest
@@ -42,6 +45,8 @@ 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
@@ -114,22 +119,24 @@ jobs:
 
       - name: Run benchmarks
         run: |
-          make BENCH_TYPE=${{ env.BENCH_TYPE }} bench_tfhe_zk_pok
+          make BENCH_TYPE="${BENCH_TYPE}" bench_tfhe_zk_pok
 
       - name: Parse results
         run: |
-          python3 ./ci/benchmark_parser.py target/criterion ${{ env.RESULTS_FILENAME }} \
+          python3 ./ci/benchmark_parser.py target/criterion "${RESULTS_FILENAME}" \
           --database tfhe_rs \
           --crate tfhe-zk-pok \
           --hardware "hpc7a.96xlarge" \
           --backend cpu \
-          --project-version "${{ env.COMMIT_HASH }}" \
-          --branch ${{ github.ref_name }} \
-          --commit-date "${{ env.COMMIT_DATE }}" \
-          --bench-date "${{ env.BENCH_DATE }}" \
+          --project-version "${COMMIT_HASH}" \
+          --branch "${REF_NAME}" \
+          --commit-date "${COMMIT_DATE}" \
+          --bench-date "${BENCH_DATE}" \
           --walk-subdirs \
           --name-suffix avx512 \
-          --bench-type ${{ env.BENCH_TYPE }}
+          --bench-type "${BENCH_TYPE}"
+        env:
+          REF_NAME: ${{ github.ref_name }}
 
       - name: Upload parsed results artifact
         uses: actions/upload-artifact@ea165f8d65b6e75b540449e92b4886f43607fa02
@@ -148,11 +155,11 @@ jobs:
       - name: Send data to Slab
         shell: bash
         run: |
-          python3 slab/scripts/data_sender.py ${{ env.RESULTS_FILENAME }} "${{ secrets.JOB_SECRET }}" \
+          python3 slab/scripts/data_sender.py "${RESULTS_FILENAME}" "${{ secrets.JOB_SECRET }}" \
           --slab-url "${{ secrets.SLAB_URL }}"
 
       - name: Slack Notification
-        if: ${{ failure() }}
+        if: ${{ failure() || (cancelled() && github.event_name != 'pull_request') }}
         continue-on-error: true
         uses: rtCamp/action-slack-notify@e31e87e03dd19038e411e38ae27cbad084a90661
         env:

.github/workflows/benchmark_wasm_client.yml

@@ -21,6 +21,9 @@ env:
   SLACK_USERNAME: ${{ secrets.BOT_USERNAME }}
   SLACK_WEBHOOK: ${{ secrets.SLACK_WEBHOOK }}
 
+
+permissions: {}
+
 jobs:
   should-run:
     runs-on: ubuntu-latest
@@ -143,15 +146,17 @@ jobs:
       - name: Parse results
         run: |
           make parse_wasm_benchmarks
-          python3 ./ci/benchmark_parser.py tfhe/wasm_pk_gen.csv ${{ env.RESULTS_FILENAME }} \
+          python3 ./ci/benchmark_parser.py tfhe-benchmark/wasm_pk_gen.csv "${RESULTS_FILENAME}" \
           --database tfhe_rs \
           --hardware "m6i.4xlarge" \
-          --project-version "${{ env.COMMIT_HASH }}" \
-          --branch ${{ github.ref_name }} \
-          --commit-date "${{ env.COMMIT_DATE }}" \
-          --bench-date "${{ env.BENCH_DATE }}" \
+          --project-version "${COMMIT_HASH}" \
+          --branch "${REF_NAME}" \
+          --commit-date "${COMMIT_DATE}" \
+          --bench-date "${BENCH_DATE}" \
           --key-gen
-          rm tfhe/wasm_pk_gen.csv
+          rm tfhe-benchmark/wasm_pk_gen.csv
+        env:
+          REF_NAME: ${{ github.ref_name }}
 
       # Run these benchmarks only once
       - name: Measure public key and ciphertext sizes in HL Api
@@ -162,7 +167,7 @@ jobs:
       - name: Parse key and ciphertext sizes results
         if:  matrix.browser == 'chrome'
         run: |
-          python3 ./ci/benchmark_parser.py tfhe/hlapi_cpk_and_cctl_sizes.csv ${{ env.RESULTS_FILENAME }} \
+          python3 ./ci/benchmark_parser.py tfhe-benchmark/hlapi_cpk_and_cctl_sizes.csv "${RESULTS_FILENAME}" \
           --key-gen \
           --append-results
 
@@ -183,11 +188,11 @@ jobs:
       - name: Send data to Slab
         shell: bash
         run: |
-          python3 slab/scripts/data_sender.py ${{ env.RESULTS_FILENAME }} "${{ secrets.JOB_SECRET }}" \
+          python3 slab/scripts/data_sender.py "${RESULTS_FILENAME}" "${{ secrets.JOB_SECRET }}" \
           --slab-url "${{ secrets.SLAB_URL }}"
 
       - name: Slack Notification
-        if: ${{ failure() }}
+        if: ${{ failure() || (cancelled() && github.event_name != 'pull_request') }}
         continue-on-error: true
         uses: rtCamp/action-slack-notify@e31e87e03dd19038e411e38ae27cbad084a90661
         env:

.github/workflows/benchmark_zk_pke.yml

@@ -31,6 +31,9 @@ env:
   SLACK_USERNAME: ${{ secrets.BOT_USERNAME }}
   SLACK_WEBHOOK: ${{ secrets.SLACK_WEBHOOK }}
 
+
+permissions: {}
+
 jobs:
   should-run:
     runs-on: ubuntu-latest
@@ -74,11 +77,13 @@ jobs:
       - name: Set benchmark types
         if: github.event_name == 'workflow_dispatch'
         run: |
-          if [[ "${{ inputs.bench_type }}" == "both" ]]; then
+          if [[ "${INPUTS_BENCH_TYPE}" == "both" ]]; then
             echo "BENCH_TYPE=[\"latency\", \"throughput\"]" >> "${GITHUB_ENV}"
           else
-            echo "BENCH_TYPE=[\"${{ inputs.bench_type }}\"]" >> "${GITHUB_ENV}"
+            echo "BENCH_TYPE=[\"${INPUTS_BENCH_TYPE}\"]" >> "${GITHUB_ENV}"
           fi
+        env:
+          INPUTS_BENCH_TYPE: ${{ inputs.bench_type }}
 
       - name: Default benchmark type
         if: github.event_name != 'workflow_dispatch'
@@ -156,25 +161,30 @@ jobs:
 
       - name: Run benchmarks with AVX512
         run: |
-          make BENCH_TYPE=${{ matrix.bench_type }} bench_integer_zk
+          make BENCH_TYPE="${BENCH_TYPE}" bench_integer_zk
+        env:
+          BENCH_TYPE: ${{ matrix.bench_type }}
 
       - name: Parse results
         run: |
-          python3 ./ci/benchmark_parser.py target/criterion ${{ env.RESULTS_FILENAME }} \
+          python3 ./ci/benchmark_parser.py target/criterion "${RESULTS_FILENAME}" \
           --database tfhe_rs \
           --hardware "hpc7a.96xlarge" \
           --backend cpu \
-          --project-version "${{ env.COMMIT_HASH }}" \
-          --branch ${{ github.ref_name }} \
-          --commit-date "${{ env.COMMIT_DATE }}" \
-          --bench-date "${{ env.BENCH_DATE }}" \
+          --project-version "${COMMIT_HASH}" \
+          --branch "${REF_NAME}" \
+          --commit-date "${COMMIT_DATE}" \
+          --bench-date "${BENCH_DATE}" \
           --walk-subdirs \
           --name-suffix avx512 \
-          --bench-type ${{ matrix.bench_type }}
+          --bench-type "${BENCH_TYPE}"
+        env:
+          REF_NAME: ${{ github.ref_name }}
+          BENCH_TYPE: ${{ matrix.bench_type }}
 
       - name: Parse CRS sizes results
         run: |
-          python3 ./ci/benchmark_parser.py tfhe/pke_zk_crs_sizes.csv ${{ env.RESULTS_FILENAME }} \
+          python3 ./ci/benchmark_parser.py tfhe-benchmark/pke_zk_crs_sizes.csv "${RESULTS_FILENAME}" \
           --object-sizes \
           --append-results
 
@@ -195,11 +205,11 @@ jobs:
       - name: Send data to Slab
         shell: bash
         run: |
-          python3 slab/scripts/data_sender.py ${{ env.RESULTS_FILENAME }} "${{ secrets.JOB_SECRET }}" \
+          python3 slab/scripts/data_sender.py "${RESULTS_FILENAME}" "${{ secrets.JOB_SECRET }}" \
           --slab-url "${{ secrets.SLAB_URL }}"
 
       - name: Slack Notification
-        if: ${{ failure() }}
+        if: ${{ failure() || (cancelled() && github.event_name != 'pull_request') }}
         continue-on-error: true
         uses: rtCamp/action-slack-notify@e31e87e03dd19038e411e38ae27cbad084a90661
         env:

.github/workflows/cargo_build.yml

@@ -8,11 +8,15 @@ env:
   RUSTFLAGS: "-C target-cpu=native"
   RUST_BACKTRACE: "full"
   RUST_MIN_STACK: "8388608"
+  CHECKOUT_TOKEN: ${{ secrets.REPO_CHECKOUT_TOKEN || secrets.GITHUB_TOKEN }}
 
 concurrency:
   group: ${{ github.workflow }}-${{ github.head_ref }}
   cancel-in-progress: true
 
+permissions:
+  contents: read
+
 jobs:
   cargo-builds:
     runs-on: ${{ matrix.os }}
@@ -26,6 +30,9 @@ jobs:
 
     steps:
       - uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
+        with:
+          persist-credentials: 'false'
+          token: ${{ env.CHECKOUT_TOKEN }}
 
       - name: Install latest stable
         uses: dtolnay/rust-toolchain@888c2e1ea69ab0d4330cbf0af1ecc7b68f368cc1
@@ -87,5 +94,10 @@ jobs:
         run: |
           make build_tfhe_coverage
 
+      - name: Run Hpu pcc checks
+        if: ${{ contains(matrix.os, 'ubuntu') }}
+        run: |
+          make pcc_hpu
+
       # The wasm build check is a bit annoying to set-up here and is done during the tests in
       # aws_tfhe_tests.yml

.github/workflows/cargo_build_tfhe_fft.yml

@@ -6,11 +6,15 @@ on:
 
 env:
   CARGO_TERM_COLOR: always
+  CHECKOUT_TOKEN: ${{ secrets.REPO_CHECKOUT_TOKEN || secrets.GITHUB_TOKEN }}
 
 concurrency:
   group: ${{ github.workflow }}-${{ github.head_ref }}
   cancel-in-progress: true
 
+permissions:
+  contents: read
+
 jobs:
   cargo-builds-fft:
     runs-on: ${{ matrix.runner_type }}
@@ -22,6 +26,9 @@ jobs:
 
     steps:
       - uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
+        with:
+          persist-credentials: 'false'
+          token: ${{ env.CHECKOUT_TOKEN }}
 
       - name: Install Rust
         uses: actions-rs/toolchain@16499b5e05bf2e26879000db0c1d13f7e13fa3af

.github/workflows/cargo_build_tfhe_ntt.yml

@@ -6,11 +6,15 @@ on:
 
 env:
   CARGO_TERM_COLOR: always
+  CHECKOUT_TOKEN: ${{ secrets.REPO_CHECKOUT_TOKEN || secrets.GITHUB_TOKEN }}
 
 concurrency:
   group: ${{ github.workflow }}-${{ github.head_ref }}
   cancel-in-progress: true
 
+permissions:
+  contents: read
+
 jobs:
   cargo-builds-ntt:
     runs-on: ${{ matrix.os }}
@@ -20,6 +24,9 @@ jobs:
       fail-fast: false
     steps:
       - uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
+        with:
+          persist-credentials: 'false'
+          token: ${{ env.CHECKOUT_TOKEN }}
 
       - name: Install Rust
         uses: actions-rs/toolchain@16499b5e05bf2e26879000db0c1d13f7e13fa3af

.github/workflows/cargo_test_fft.yml

@@ -10,11 +10,15 @@ on:
 env:
   CARGO_TERM_COLOR: always
   IS_PULL_REQUEST: ${{ github.event_name == 'pull_request' }}
+  CHECKOUT_TOKEN: ${{ secrets.REPO_CHECKOUT_TOKEN || secrets.GITHUB_TOKEN }}
 
 concurrency:
   group: ${{ github.workflow }}-${{ github.head_ref }}
   cancel-in-progress: true
 
+permissions:
+  contents: read
+
 jobs:
   should-run:
     runs-on: ubuntu-latest
@@ -28,6 +32,7 @@ jobs:
         with:
           fetch-depth: 0
           persist-credentials: 'false'
+          token: ${{ env.CHECKOUT_TOKEN }}
 
       - name: Check for file changes
         id: changed-files
@@ -46,10 +51,13 @@ jobs:
     runs-on: ${{ matrix.runner_type }}
     strategy:
       matrix:
-        runner_type: [ubuntu-latest, macos-latest, windows-latest]
+        runner_type: [ ubuntu-latest, macos-latest, windows-latest ]
       fail-fast: false
     steps:
       - uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
+        with:
+          persist-credentials: 'false'
+          token: ${{ env.CHECKOUT_TOKEN }}
 
       - name: Install Rust
         uses: actions-rs/toolchain@16499b5e05bf2e26879000db0c1d13f7e13fa3af
@@ -74,9 +82,12 @@ jobs:
     runs-on: ${{ matrix.runner_type }}
     strategy:
       matrix:
-        runner_type: [ubuntu-latest, macos-latest, windows-latest]
+        runner_type: [ ubuntu-latest, macos-latest, windows-latest ]
     steps:
       - uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
+        with:
+          persist-credentials: 'false'
+          token: ${{ env.CHECKOUT_TOKEN }}
 
       - name: Install Rust
         uses: actions-rs/toolchain@16499b5e05bf2e26879000db0c1d13f7e13fa3af
@@ -98,6 +109,9 @@ jobs:
     runs-on: ubuntu-latest
     steps:
       - uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
+        with:
+          persist-credentials: 'false'
+          token: ${{ env.CHECKOUT_TOKEN }}
 
       - name: Test node js
         run: |

.github/workflows/cargo_test_ntt.yml

@@ -10,12 +10,16 @@ on:
 env:
   CARGO_TERM_COLOR: always
   IS_PULL_REQUEST: ${{ github.event_name == 'pull_request' }}
+  CHECKOUT_TOKEN: ${{ secrets.REPO_CHECKOUT_TOKEN || secrets.GITHUB_TOKEN }}
 
 concurrency:
   group: ${{ github.workflow }}-${{ github.head_ref }}
   cancel-in-progress: true
 
-jobs:  
+permissions:
+  contents: read
+
+jobs:
   should-run:
     runs-on: ubuntu-latest
     permissions:
@@ -28,6 +32,7 @@ jobs:
         with:
           fetch-depth: 0
           persist-credentials: 'false'
+          token: ${{ env.CHECKOUT_TOKEN }}
 
       - name: Check for file changes
         id: changed-files
@@ -46,10 +51,13 @@ jobs:
     runs-on: ${{ matrix.os }}
     strategy:
       matrix:
-        os: [ubuntu-latest, macos-latest, windows-latest]
+        os: [ ubuntu-latest, macos-latest, windows-latest ]
       fail-fast: false
     steps:
       - uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
+        with:
+          persist-credentials: 'false'
+          token: ${{ env.CHECKOUT_TOKEN }}
 
       - name: Install Rust
         uses: actions-rs/toolchain@16499b5e05bf2e26879000db0c1d13f7e13fa3af
@@ -69,9 +77,12 @@ jobs:
     runs-on: ${{ matrix.os }}
     strategy:
       matrix:
-        os: [ubuntu-latest, macos-latest, windows-latest]
+        os: [ ubuntu-latest, macos-latest, windows-latest ]
     steps:
       - uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
+        with:
+          persist-credentials: 'false'
+          token: ${{ env.CHECKOUT_TOKEN }}
 
       - name: Install Rust
         uses: actions-rs/toolchain@16499b5e05bf2e26879000db0c1d13f7e13fa3af

.github/workflows/check_commit.yml

@@ -3,6 +3,10 @@ name: Check commit and PR compliance
 on:
   pull_request:
 
+permissions:
+  contents: read
+  pull-requests: read # Permission needed to scan commits in a pull-request
+
 jobs:
   check-commit-pr:
     name: Check commit and PR

.github/workflows/ci_lint.yml

@@ -5,9 +5,13 @@ on:
   pull_request:
 
 env:
-  ACTIONLINT_VERSION: 1.6.27
+  ACTIONLINT_VERSION: 1.7.7
+  ACTIONLINT_CHECKSUM: "023070a287cd8cccd71515fedc843f1985bf96c436b7effaecce67290e7e0757"
   CHECKOUT_TOKEN: ${{ secrets.REPO_CHECKOUT_TOKEN || secrets.GITHUB_TOKEN }}
 
+permissions:
+  contents: read
+
 jobs:
   lint-check:
     name: Lint and checks
@@ -21,17 +25,24 @@ jobs:
 
       - name: Get actionlint
         run: |
-          bash <(curl https://raw.githubusercontent.com/rhysd/actionlint/main/scripts/download-actionlint.bash) ${{ env.ACTIONLINT_VERSION }}
-          echo "f2ee6d561ce00fa93aab62a7791c1a0396ec7e8876b2a8f2057475816c550782  actionlint" > checksum
+          wget "https://github.com/rhysd/actionlint/releases/download/v${{ env.ACTIONLINT_VERSION }}/actionlint_${{ env.ACTIONLINT_VERSION }}_linux_amd64.tar.gz"
+          echo "${{ env.ACTIONLINT_CHECKSUM }} actionlint_${{ env.ACTIONLINT_VERSION }}_linux_amd64.tar.gz" > checksum
           sha256sum -c checksum
+          tar -xf actionlint_${{ env.ACTIONLINT_VERSION }}_linux_amd64.tar.gz actionlint
           ln -s "$(pwd)/actionlint" /usr/local/bin/
 
       - name: Lint workflows
         run: |
           make lint_workflow
 
+      - name: Check workflows security
+        run: |
+          make check_workflow_security
+        env:
+          GH_TOKEN: ${{ env.CHECKOUT_TOKEN }}
+
       - name: Ensure SHA pinned actions
-        uses: zgosalvez/github-actions-ensure-sha-pinned-actions@4830be28ce81da52ec70d65c552a7403821d98d4 # v3.0.23
+        uses: zgosalvez/github-actions-ensure-sha-pinned-actions@fc87bb5b5a97953d987372e74478de634726b3e5 # v3.0.25
         with:
           allowlist: |
             slsa-framework/slsa-github-generator

.github/workflows/code_coverage.yml

@@ -10,12 +10,16 @@ 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.
   workflow_dispatch:
   # Code coverage workflow is only run via workflow_dispatch event since execution duration is not stabilized yet.
 
+permissions:
+  contents: read
+
 jobs:
   setup-instance:
     name: Setup instance (code-coverage)
@@ -45,6 +49,9 @@ jobs:
     steps:
       - name: Checkout tfhe-rs
         uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
+        with:
+          persist-credentials: 'false'
+          token: ${{ env.CHECKOUT_TOKEN }}
 
       - name: Install latest stable
         uses: dtolnay/rust-toolchain@888c2e1ea69ab0d4330cbf0af1ecc7b68f368cc1
@@ -83,7 +90,7 @@ jobs:
           make test_shortint_cov
 
       - name: Upload tfhe coverage to Codecov
-        uses: codecov/codecov-action@0565863a31f2c772f9f0395002a31e3f06189574
+        uses: codecov/codecov-action@18283e04ce6e62d37312384ff67231eb8fd56d24
         if: steps.changed-files.outputs.tfhe_any_changed == 'true'
         with:
           token: ${{ secrets.CODECOV_TOKEN }}
@@ -97,7 +104,7 @@ jobs:
           make test_integer_cov
 
       - name: Upload tfhe coverage to Codecov
-        uses: codecov/codecov-action@0565863a31f2c772f9f0395002a31e3f06189574
+        uses: codecov/codecov-action@18283e04ce6e62d37312384ff67231eb8fd56d24
         if: steps.changed-files.outputs.tfhe_any_changed == 'true'
         with:
           token: ${{ secrets.CODECOV_TOKEN }}
@@ -106,7 +113,7 @@ jobs:
           files: integer/cobertura.xml
 
       - name: Slack Notification
-        if: ${{ failure() }}
+        if: ${{ failure() || (cancelled() && github.event_name != 'pull_request') }}
         continue-on-error: true
         uses: rtCamp/action-slack-notify@e31e87e03dd19038e411e38ae27cbad084a90661
         env:

.github/workflows/csprng_randomness_tests.yml

@@ -21,6 +21,9 @@ on:
   pull_request:
     types: [ labeled ]
 
+permissions:
+  contents: read
+
 jobs:
   setup-instance:
     name: Setup instance (csprng-randomness-tests)
@@ -46,7 +49,7 @@ jobs:
         id: start-github-instance
         if: env.SECRETS_AVAILABLE == 'false'
         run: |
-          echo "runner_group=${{ env.EXTERNAL_CONTRIBUTION_RUNNER }}" >> "$GITHUB_OUTPUT"
+          echo "runner_group=${EXTERNAL_CONTRIBUTION_RUNNER}" >> "$GITHUB_OUTPUT"
 
   csprng-randomness-tests:
     name: CSPRNG randomness tests
@@ -72,7 +75,7 @@ jobs:
           make dieharder_csprng
 
       - name: Slack Notification
-        if: ${{ failure() }}
+        if: ${{ failure() || (cancelled() && github.event_name != 'pull_request') }}
         continue-on-error: true
         uses: rtCamp/action-slack-notify@e31e87e03dd19038e411e38ae27cbad084a90661
         env:

.github/workflows/data_pr_close.yml

@@ -25,6 +25,9 @@ on:
 # the script will always return 0 because of the "echo EOF".
 
 
+
+permissions: {}
+
 jobs:
   auto_close_job:
     if: ${{ contains(github.event.pull_request.labels.*.name, 'data_PR') }}
@@ -39,14 +42,17 @@ jobs:
           curl --fail-with-body --no-progress-meter -L -X GET \
           -H "Accept: application/vnd.github+json" \
           -H "X-GitHub-Api-Version: 2022-11-28"  \
-          ${{ env.TARGET_REPO_API_URL }}/pulls\?head=${{ github.repository_owner }}:${{ env.PR_BRANCH }} | jq -e '.[0]' | sed 's/null/{ "message": "corresponding PR not found" }/'
+          "${TARGET_REPO_API_URL}"/pulls\?head="${REPO_OWNER}":"${PR_BRANCH}" | jq -e '.[0]' | sed 's/null/{ "message": "corresponding PR not found" }/'
           RES="$?"
           echo EOF
         } >> "${GITHUB_ENV}"
         exit $RES
+      env:
+        REPO_OWNER: ${{ github.repository_owner }}
 
     - name: Comment on the PR to indicate the reason of the close
       run: |
+        BODY="'{ \"body\": \"PR ${CLOSE_TYPE}d because the corresponding PR in main repo was ${CLOSE_TYPE}d: ${REPO}#${EVENT_NUMBER}\" }'"
         {
           set +e
           set -o pipefail
@@ -55,12 +61,16 @@ jobs:
           -H "Accept: application/vnd.github+json" \
           -H "Authorization: Bearer ${{ secrets.FHE_ACTIONS_TOKEN }}" \
           -H "X-GitHub-Api-Version: 2022-11-28" \
-          ${{ fromJson(env.TARGET_REPO_PR).comments_url }} \
-          -d '{ "body": "PR ${{ env.CLOSE_TYPE }}d because the corresponding PR in main repo was ${{ env.CLOSE_TYPE }}d: ${{ github.repository }}#${{ github.event.number  }}" }'
+          "${COMMENTS_URL}" \
+          -d "${BODY}"
           RES="$?"
           echo EOF
         } >> "${GITHUB_ENV}"
         exit $RES
+      env:
+        REPO: ${{ github.repository }}
+        EVENT_NUMBER: ${{ github.event.number }}
+        COMMENTS_URL: ${{ fromJson(env.TARGET_REPO_PR).comments_url }}
 
     - name: Merge the Pull Request in the data repo
       if: ${{ github.event.pull_request.merged }}
@@ -73,12 +83,14 @@ jobs:
           -H "Accept: application/vnd.github+json" \
           -H "Authorization: Bearer ${{ secrets.FHE_ACTIONS_TOKEN }}" \
           -H "X-GitHub-Api-Version: 2022-11-28" \
-          ${{ fromJson(env.TARGET_REPO_PR).url }}/merge \
+          "${TARGET_REPO_PR_URL}"/merge \
           -d '{ "merge_method": "rebase" }'
           RES="$?"
           echo EOF
         } >> "${GITHUB_ENV}"
         exit $RES
+      env:
+        TARGET_REPO_PR_URL: ${{ fromJson(env.TARGET_REPO_PR).url }}
 
     - name: Close the Pull Request in the data repo
       if: ${{ !github.event.pull_request.merged }}
@@ -91,12 +103,14 @@ jobs:
           -H "Accept: application/vnd.github+json" \
           -H "Authorization: Bearer ${{ secrets.FHE_ACTIONS_TOKEN }}" \
           -H "X-GitHub-Api-Version: 2022-11-28" \
-          ${{ fromJson(env.TARGET_REPO_PR).url }} \
+          "${TARGET_REPO_PR_URL}" \
           -d '{ "state": "closed" }'
           RES="$?"
           echo EOF
         } >> "${GITHUB_ENV}"
         exit $RES
+      env:
+        TARGET_REPO_PR_URL: ${{ fromJson(env.TARGET_REPO_PR).url }}
 
     - name: Delete the associated branch in the data repo
       run: |
@@ -108,7 +122,7 @@ jobs:
           -H "Accept: application/vnd.github+json" \
           -H "Authorization: Bearer ${{ secrets.FHE_ACTIONS_TOKEN }}" \
           -H "X-GitHub-Api-Version: 2022-11-28" \
-          ${{ env.TARGET_REPO_API_URL }}/git/refs/heads/${{ env.PR_BRANCH }}
+          "${TARGET_REPO_API_URL}"/git/refs/heads/"${PR_BRANCH}"
           RES="$?"
           echo EOF
         } >> "${GITHUB_ENV}"

.github/workflows/gpu_4090_tests.yml

@@ -22,6 +22,9 @@ on:
     # Nightly tests @ 1AM after each work day
     - cron: "0 1 * * MON-FRI"
 
+permissions:
+  contents: read
+
 jobs:
   cuda-tests-linux:
     name: CUDA tests (RTX 4090)
@@ -77,7 +80,7 @@ jobs:
           github_token: ${{ secrets.GITHUB_TOKEN }}
 
       - name: Slack Notification
-        if: ${{ failure() }}
+        if: ${{ failure() || (cancelled() && github.event_name != 'pull_request') }}
         continue-on-error: true
         uses: rtCamp/action-slack-notify@e31e87e03dd19038e411e38ae27cbad084a90661
         env:

.github/workflows/gpu_fast_h100_tests.yml

@@ -25,6 +25,9 @@ on:
   pull_request:
     types: [ labeled ]
 
+permissions:
+  contents: read
+
 jobs:
   should-run:
     runs-on: ubuntu-latest
@@ -102,7 +105,7 @@ jobs:
         id: start-github-instance
         if: env.SECRETS_AVAILABLE == 'false'
         run: |
-          echo "runner_group=${{ env.EXTERNAL_CONTRIBUTION_RUNNER }}" >> "$GITHUB_OUTPUT"
+          echo "runner_group=${EXTERNAL_CONTRIBUTION_RUNNER}" >> "$GITHUB_OUTPUT"
 
   cuda-tests-linux:
     name: CUDA H100 tests
@@ -169,7 +172,9 @@ jobs:
       - 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 }}), "  >> "${GITHUB_ENV}"
+          echo "PULL_REQUEST_MD_LINK=[pull-request](${PR_BASE_URL}${{ github.event.pull_request.number }}), "  >> "${GITHUB_ENV}"
+        env:
+          PR_BASE_URL: ${{ vars.PR_BASE_URL }}
 
       - name: Send message
         if: env.SECRETS_AVAILABLE == 'true'

.github/workflows/gpu_fast_tests.yml

@@ -24,6 +24,9 @@ on:
   workflow_dispatch:
   pull_request:
 
+permissions:
+  contents: read
+
 jobs:
   should-run:
     runs-on: ubuntu-latest
@@ -87,7 +90,7 @@ jobs:
         id: start-github-instance
         if: env.SECRETS_AVAILABLE == 'false'
         run: |
-          echo "runner_group=${{ env.EXTERNAL_CONTRIBUTION_RUNNER }}" >> "$GITHUB_OUTPUT"
+          echo "runner_group=${EXTERNAL_CONTRIBUTION_RUNNER}" >> "$GITHUB_OUTPUT"
 
   cuda-tests-linux:
     name: CUDA tests
@@ -153,7 +156,9 @@ jobs:
       - 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 }}), "  >> "${GITHUB_ENV}"
+          echo "PULL_REQUEST_MD_LINK=[pull-request](${PR_BASE_URL}${{ github.event.pull_request.number }}), "  >> "${GITHUB_ENV}"
+        env:
+          PR_BASE_URL: ${{ vars.PR_BASE_URL }}
 
       - name: Send message
         if: env.SECRETS_AVAILABLE == 'true'

.github/workflows/gpu_full_h100_tests.yml

@@ -15,6 +15,9 @@ env:
 on:
   workflow_dispatch:
 
+
+permissions: {}
+
 jobs:
   setup-instance:
     name: Setup instance (cuda-h100-tests)
@@ -62,18 +65,6 @@ jobs:
             cuda: "12.2"
             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:

.github/workflows/gpu_full_multi_gpu_tests.yml

@@ -25,6 +25,9 @@ on:
   pull_request:
     types: [ labeled ]
 
+permissions:
+  contents: read
+
 jobs:
   should-run:
     runs-on: ubuntu-latest
@@ -89,7 +92,7 @@ jobs:
         id: start-github-instance
         if: env.SECRETS_AVAILABLE == 'false'
         run: |
-          echo "runner_group=${{ env.EXTERNAL_CONTRIBUTION_RUNNER }}" >> "$GITHUB_OUTPUT"
+          echo "runner_group=${EXTERNAL_CONTRIBUTION_RUNNER}" >> "$GITHUB_OUTPUT"
 
   cuda-tests-linux:
     name: CUDA multi-GPU tests
@@ -146,7 +149,7 @@ jobs:
 
       - name: Run High Level API Tests
         run: |
-          BIG_TESTS_INSTANCE=FALSE make test_high_level_api_gpu
+          make test_high_level_api_gpu
 
   slack-notify:
     name: Slack Notification
@@ -158,7 +161,9 @@ jobs:
       - 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 }}), "  >> "${GITHUB_ENV}"
+          echo "PULL_REQUEST_MD_LINK=[pull-request](${PR_BASE_URL}${{ github.event.pull_request.number }}), "  >> "${GITHUB_ENV}"
+        env:
+          PR_BASE_URL: ${{ vars.PR_BASE_URL }}
 
       - name: Send message
         if: env.SECRETS_AVAILABLE == 'true'

.github/workflows/gpu_integer_long_run_tests.yml

@@ -10,6 +10,7 @@ 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.
@@ -18,6 +19,9 @@ on:
     # Nightly tests will be triggered each evening 8p.m.
     - cron: "0 20 * * *"
 
+permissions:
+  contents: read
+
 jobs:
   setup-instance:
     name: Setup instance (gpu-tests)
@@ -57,6 +61,9 @@ jobs:
     steps:
       - name: Checkout tfhe-rs
         uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
+        with:
+          persist-credentials: 'false'
+          token: ${{ env.CHECKOUT_TOKEN }}
 
       - name: Setup Hyperstack dependencies
         uses: ./.github/actions/gpu_setup

.github/workflows/gpu_pcc.yml

@@ -1,4 +1,4 @@
-# Perfom tfhe-cuda-backend post-commit checks on an AWS instance
+# Perform tfhe-cuda-backend post-commit checks on an AWS instance
 name: Cuda - Post-commit Checks
 
 env:
@@ -17,10 +17,15 @@ env:
   # Secrets will be available only to zama-ai organization members
   SECRETS_AVAILABLE: ${{ secrets.JOB_SECRET != '' }}
   EXTERNAL_CONTRIBUTION_RUNNER: "large_ubuntu_16-22.04"
+  CUDA_KEYRING_PACKAGE: cuda-keyring_1.1-1_all.deb
+  CUDA_KEYRING_SHA: "d93190d50b98ad4699ff40f4f7af50f16a76dac3bb8da1eaaf366d47898ff8df"
 
 on:
   pull_request:
 
+permissions:
+  contents: read
+
 jobs:
   setup-instance:
     name: Setup instance (cuda-pcc)
@@ -45,7 +50,7 @@ jobs:
         id: start-github-instance
         if: env.SECRETS_AVAILABLE == 'false'
         run: |
-          echo "runner_group=${{ env.EXTERNAL_CONTRIBUTION_RUNNER }}" >> "$GITHUB_OUTPUT"
+          echo "runner_group=${EXTERNAL_CONTRIBUTION_RUNNER}" >> "$GITHUB_OUTPUT"
 
   cuda-pcc:
     name: CUDA post-commit checks
@@ -77,8 +82,10 @@ jobs:
         shell: bash
         run: |
           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
+          wget https://developer.download.nvidia.com/compute/cuda/repos/ubuntu2204/x86_64/"${CUDA_KEYRING_PACKAGE}"
+          echo "${CUDA_KEYRING_SHA} ${CUDA_KEYRING_PACKAGE}" > checksum
+          sha256sum -c checksum
+          sudo dpkg -i "${CUDA_KEYRING_PACKAGE}"
           sudo apt update
           sudo apt -y install "cuda-toolkit-${TOOLKIT_VERSION}" cmake-format
 
@@ -113,10 +120,16 @@ jobs:
         run: |
           make pcc_gpu
 
+      - name: Check build with hpu enabled
+        run: |
+          make clippy_gpu_hpu
+
       - 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}"
+          echo "PULL_REQUEST_MD_LINK=[pull-request](${PR_BASE_URL}${{ github.event.pull_request.number }}), "  >> "${GITHUB_ENV}"
+        env:
+          PR_BASE_URL: ${{ vars.PR_BASE_URL }}
 
       - name: Slack Notification
         if: ${{ failure() && env.SECRETS_AVAILABLE == 'true' }}

.github/workflows/gpu_signed_integer_classic_tests.yml

@@ -25,6 +25,9 @@ on:
   pull_request:
     types: [ labeled ]
 
+permissions:
+  contents: read
+
 jobs:
   should-run:
     runs-on: ubuntu-latest
@@ -89,7 +92,7 @@ jobs:
         id: start-github-instance
         if: env.SECRETS_AVAILABLE == 'false'
         run: |
-          echo "runner_group=${{ env.EXTERNAL_CONTRIBUTION_RUNNER }}" >> "$GITHUB_OUTPUT"
+          echo "runner_group=${EXTERNAL_CONTRIBUTION_RUNNER}" >> "$GITHUB_OUTPUT"
 
   cuda-tests-linux:
     name: CUDA signed integer tests with classical PBS
@@ -141,7 +144,9 @@ jobs:
       - 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 }}), "  >> "${GITHUB_ENV}"
+          echo "PULL_REQUEST_MD_LINK=[pull-request](${PR_BASE_URL}${{ github.event.pull_request.number }}), "  >> "${GITHUB_ENV}"
+        env:
+          PR_BASE_URL: ${{ vars.PR_BASE_URL }}
 
       - name: Send message
         if: env.SECRETS_AVAILABLE == 'true'

.github/workflows/gpu_signed_integer_h100_tests.yml

@@ -25,6 +25,8 @@ on:
   pull_request:
     types: [ labeled ]
 
+permissions:
+  contents: read
 
 jobs:
   should-run:
@@ -103,7 +105,7 @@ jobs:
         id: start-github-instance
         if: env.SECRETS_AVAILABLE == 'false'
         run: |
-          echo "runner_group=${{ env.EXTERNAL_CONTRIBUTION_RUNNER }}" >> "$GITHUB_OUTPUT"
+          echo "runner_group=${EXTERNAL_CONTRIBUTION_RUNNER}" >> "$GITHUB_OUTPUT"
 
   cuda-tests-linux:
     name: CUDA H100 signed integer tests
@@ -156,7 +158,9 @@ jobs:
       - 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 }}), "  >> "${GITHUB_ENV}"
+          echo "PULL_REQUEST_MD_LINK=[pull-request](${PR_BASE_URL}${{ github.event.pull_request.number }}), "  >> "${GITHUB_ENV}"
+        env:
+          PR_BASE_URL: ${{ vars.PR_BASE_URL }}
 
       - name: Send message
         if: env.SECRETS_AVAILABLE == 'true'

.github/workflows/gpu_signed_integer_tests.yml

@@ -29,6 +29,9 @@ on:
     # Nightly tests @ 1AM after each work day
     - cron: "0 1 * * MON-FRI"
 
+permissions:
+  contents: read
+
 jobs:
   should-run:
     runs-on: ubuntu-latest
@@ -93,7 +96,7 @@ jobs:
         id: start-github-instance
         if: env.SECRETS_AVAILABLE == 'false'
         run: |
-          echo "runner_group=${{ env.EXTERNAL_CONTRIBUTION_RUNNER }}" >> "$GITHUB_OUTPUT"
+          echo "runner_group=${EXTERNAL_CONTRIBUTION_RUNNER}" >> "$GITHUB_OUTPUT"
 
   cuda-signed-integer-tests:
     name: CUDA signed integer tests
@@ -153,7 +156,9 @@ jobs:
       - 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 }}), "  >> "${GITHUB_ENV}"
+          echo "PULL_REQUEST_MD_LINK=[pull-request](${PR_BASE_URL}${{ github.event.pull_request.number }}), "  >> "${GITHUB_ENV}"
+        env:
+          PR_BASE_URL: ${{ vars.PR_BASE_URL }}
 
       - name: Send message
         if: env.SECRETS_AVAILABLE == 'true'

.github/workflows/gpu_unsigned_integer_classic_tests.yml

@@ -25,6 +25,8 @@ on:
   pull_request:
     types: [ labeled ]
 
+permissions:
+  contents: read
 
 jobs:
   should-run:
@@ -90,7 +92,7 @@ jobs:
         id: start-github-instance
         if: env.SECRETS_AVAILABLE == 'false'
         run: |
-          echo "runner_group=${{ env.EXTERNAL_CONTRIBUTION_RUNNER }}" >> "$GITHUB_OUTPUT"
+          echo "runner_group=${EXTERNAL_CONTRIBUTION_RUNNER}" >> "$GITHUB_OUTPUT"
 
   cuda-tests-linux:
     name: CUDA unsigned integer tests with classical PBS
@@ -142,7 +144,9 @@ jobs:
       - 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 }}), "  >> "${GITHUB_ENV}"
+          echo "PULL_REQUEST_MD_LINK=[pull-request](${PR_BASE_URL}${{ github.event.pull_request.number }}), "  >> "${GITHUB_ENV}"
+        env:
+          PR_BASE_URL: ${{ vars.PR_BASE_URL }}
 
       - name: Send message
         if: env.SECRETS_AVAILABLE == 'true'

.github/workflows/gpu_unsigned_integer_h100_tests.yml

@@ -25,6 +25,9 @@ on:
   pull_request:
     types: [ labeled ]
 
+permissions:
+  contents: read
+
 jobs:
   should-run:
     runs-on: ubuntu-latest
@@ -102,7 +105,7 @@ jobs:
         id: start-github-instance
         if: env.SECRETS_AVAILABLE == 'false'
         run: |
-          echo "runner_group=${{ env.EXTERNAL_CONTRIBUTION_RUNNER }}" >> "$GITHUB_OUTPUT"
+          echo "runner_group=${EXTERNAL_CONTRIBUTION_RUNNER}" >> "$GITHUB_OUTPUT"
 
   cuda-tests-linux:
     name: CUDA H100 unsigned integer tests
@@ -155,7 +158,9 @@ jobs:
       - 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 }}), "  >> "${GITHUB_ENV}"
+          echo "PULL_REQUEST_MD_LINK=[pull-request](${PR_BASE_URL}${{ github.event.pull_request.number }}), "  >> "${GITHUB_ENV}"
+        env:
+          PR_BASE_URL: ${{ vars.PR_BASE_URL }}
 
       - name: Send message
         if: env.SECRETS_AVAILABLE == 'true'

.github/workflows/gpu_unsigned_integer_tests.yml

@@ -29,6 +29,9 @@ on:
     # Nightly tests @ 1AM after each work day
     - cron: "0 1 * * MON-FRI"
 
+permissions:
+  contents: read
+
 jobs:
   should-run:
     runs-on: ubuntu-latest
@@ -93,7 +96,7 @@ jobs:
         id: start-github-instance
         if: env.SECRETS_AVAILABLE == 'false'
         run: |
-          echo "runner_group=${{ env.EXTERNAL_CONTRIBUTION_RUNNER }}" >> "$GITHUB_OUTPUT"
+          echo "runner_group=${EXTERNAL_CONTRIBUTION_RUNNER}" >> "$GITHUB_OUTPUT"
 
   cuda-unsigned-integer-tests:
     name: CUDA unsigned integer tests
@@ -153,7 +156,9 @@ jobs:
       - 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 }}), "  >> "${GITHUB_ENV}"
+          echo "PULL_REQUEST_MD_LINK=[pull-request](${PR_BASE_URL}${{ github.event.pull_request.number }}), "  >> "${GITHUB_ENV}"
+        env:
+          PR_BASE_URL: ${{ vars.PR_BASE_URL }}
 
       - name: Send message
         if: env.SECRETS_AVAILABLE == 'true'

.github/workflows/hpu_hlapi_tests.yml

@@ -0,0 +1,73 @@
+# Test tfhe-fft
+name: Cargo Test HLAPI HPU
+
+on:
+  pull_request:
+  push:
+    branches:
+      - main
+
+env:
+  CARGO_TERM_COLOR: always
+  IS_PULL_REQUEST: ${{ github.event_name == 'pull_request' }}
+  CHECKOUT_TOKEN: ${{ secrets.REPO_CHECKOUT_TOKEN || secrets.GITHUB_TOKEN }}
+
+concurrency:
+  group: ${{ github.workflow }}-${{ github.head_ref }}
+  cancel-in-progress: true
+
+
+permissions: { }
+
+jobs:
+  should-run:
+    runs-on: ubuntu-latest
+    permissions:
+      pull-requests: read
+    outputs:
+      hpu_test: ${{ env.IS_PULL_REQUEST == 'false' || steps.changed-files.outputs.hpu_any_changed }}
+    steps:
+      - name: Checkout tfhe-rs
+        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@ed68ef82c095e0d48ec87eccea555d944a631a4c # v46.0.5
+        with:
+          files_yaml: |
+            hpu:
+              - tfhe/Cargo.toml
+              - Makefile
+              - backends/tfhe-hpu-backend/**
+              - mockups/tfhe-hpu-mockup/**
+
+  cargo-tests-hpu:
+    needs: should-run
+    if: needs.should-run.outputs.hpu_test == 'true'
+    runs-on: large_ubuntu_16
+    steps:
+      - uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
+        with:
+          persist-credentials: 'false'
+          token: ${{ env.CHECKOUT_TOKEN }}
+
+      - name: Install Rust
+        uses: actions-rs/toolchain@16499b5e05bf2e26879000db0c1d13f7e13fa3af
+        with:
+          toolchain: stable
+          override: true
+
+      - name: Install Just
+        run: |
+          cargo install just
+
+      - name: Test HLAPI HPU
+        run: |
+          source setup_hpu.sh
+          just -f mockups/tfhe-hpu-mockup/Justfile  BUILD_PROFILE=release mockup &
+          make HPU_CONFIG=sim test_high_level_api_hpu
+     

.github/workflows/integer_long_run_tests.yml

@@ -18,6 +18,9 @@ on:
     # Weekly tests will be triggered each Friday at 9p.m.
     - cron: "0 21 * * 5"
 
+
+permissions: {}
+
 jobs:
   setup-instance:
     name: Setup instance (cpu-tests)
@@ -63,7 +66,7 @@ jobs:
           make test_integer_long_run
 
       - name: Slack Notification
-        if: ${{ failure() }}
+        if: ${{ failure() || (cancelled() && github.event_name != 'pull_request') }}
         continue-on-error: true
         uses: rtCamp/action-slack-notify@e31e87e03dd19038e411e38ae27cbad084a90661
         env:

.github/workflows/m1_tests.yml

@@ -27,6 +27,9 @@ concurrency:
   group: ${{ github.workflow_ref }}
   cancel-in-progress: true
 
+permissions:
+  contents: read
+
 jobs:
   cargo-builds-m1:
     if: ${{ (github.event_name == 'schedule' &&  github.repository == 'zama-ai/tfhe-rs') ||

.github/workflows/make_release.yml

@@ -28,6 +28,12 @@ on:
 env:
   ACTION_RUN_URL: ${{ github.server_url }}/${{ github.repository }}/actions/runs/${{ github.run_id }}
   NPM_TAG: ""
+  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 }}
+
+permissions: {}
 
 jobs:
   verify_tag:
@@ -94,7 +100,7 @@ jobs:
         run: |
           echo "NPM_TAG=latest" >> "${GITHUB_ENV}"
       - name: Download artifact
-        uses: actions/download-artifact@95815c38cf2ff2164869cbab79da8d1f422bc89e # v4.2.1
+        uses: actions/download-artifact@d3f86a106a0bac45b974a628896c90dbdf5c8093 # v4.3.0
         with:
           name: crate
           path: target/package
@@ -104,7 +110,10 @@ jobs:
           CRATES_TOKEN: ${{ secrets.CARGO_REGISTRY_TOKEN }}
           DRY_RUN: ${{ inputs.dry_run && '--dry-run' || '' }}
         run: |
-          cargo publish -p tfhe --token ${{ env.CRATES_TOKEN }} ${{ env.DRY_RUN }}
+          # DRY_RUN expansion cannot be double quoted when variable contains empty string otherwise cargo publish 
+          # would fail. This is safe since DRY_RUN is handled in the env section above.
+          # shellcheck disable=SC2086
+          cargo publish -p tfhe --token "${CRATES_TOKEN}" ${DRY_RUN}
 
       - name: Generate hash
         id: published_hash
@@ -116,11 +125,7 @@ jobs:
         uses: rtCamp/action-slack-notify@e31e87e03dd19038e411e38ae27cbad084a90661 # v2.3.3
         env:
           SLACK_COLOR: failure
-          SLACK_CHANNEL: ${{ secrets.SLACK_CHANNEL }}
-          SLACK_ICON: https://pbs.twimg.com/profile_images/1274014582265298945/OjBKP9kn_400x400.png
           SLACK_MESSAGE: "SLSA tfhe crate - hash comparison failure: (${{ env.ACTION_RUN_URL }})"
-          SLACK_USERNAME: ${{ secrets.BOT_USERNAME }}
-          SLACK_WEBHOOK: ${{ secrets.SLACK_WEBHOOK }}
 
       - name: Build web package
         if: ${{ inputs.push_web_package }}
@@ -156,13 +161,9 @@ jobs:
           provenance: true
 
       - name: Slack Notification
-        if: ${{ failure() }}
+        if: ${{ failure() || (cancelled() && github.event_name != 'pull_request') }}
         continue-on-error: true
         uses: rtCamp/action-slack-notify@e31e87e03dd19038e411e38ae27cbad084a90661 # v2.3.3
         env:
           SLACK_COLOR: ${{ job.status }}
-          SLACK_CHANNEL: ${{ secrets.SLACK_CHANNEL }}
-          SLACK_ICON: https://pbs.twimg.com/profile_images/1274014582265298945/OjBKP9kn_400x400.png
           SLACK_MESSAGE: "tfhe release failed: (${{ env.ACTION_RUN_URL }})"
-          SLACK_USERNAME: ${{ secrets.BOT_USERNAME }}
-          SLACK_WEBHOOK: ${{ secrets.SLACK_WEBHOOK }}

.github/workflows/make_release_cuda.yml

@@ -15,6 +15,8 @@ env:
   SLACK_USERNAME: ${{ secrets.BOT_USERNAME }}
   SLACK_WEBHOOK: ${{ secrets.SLACK_WEBHOOK }}
 
+permissions: {}
+
 jobs:
   verify_tag:
     uses: ./.github/workflows/verify_tagged_commit.yml
@@ -157,7 +159,10 @@ jobs:
           CRATES_TOKEN: ${{ secrets.CARGO_REGISTRY_TOKEN }}
           DRY_RUN: ${{ inputs.dry_run && '--dry-run' || '' }}
         run: |
-          cargo publish -p tfhe-cuda-backend --token ${{ env.CRATES_TOKEN }} ${{ env.DRY_RUN }}
+          # DRY_RUN expansion cannot be double quoted when variable contains empty string otherwise cargo publish 
+          # would fail. This is safe since DRY_RUN is handled in the env section above.
+          # shellcheck disable=SC2086
+          cargo publish -p tfhe-cuda-backend --token "${CRATES_TOKEN}" ${DRY_RUN}
 
       - name: Generate hash
         id: published_hash
@@ -169,14 +174,10 @@ jobs:
         uses: rtCamp/action-slack-notify@e31e87e03dd19038e411e38ae27cbad084a90661 # v2.3.3
         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() }}
+        if: ${{ failure() || (cancelled() && github.event_name != 'pull_request') }}
         continue-on-error: true
         uses: rtCamp/action-slack-notify@e31e87e03dd19038e411e38ae27cbad084a90661 # v2.3.3
         env:

.github/workflows/make_release_hpu.yml

@@ -0,0 +1,105 @@
+name: Publish HPU 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 }}
+
+permissions: {}
+
+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
+    needs: verify_tag
+    outputs:
+      hash: ${{ steps.hash.outputs.hash }}
+    steps:
+      - name: Checkout
+        uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683 # v4.2.2
+        with:
+          fetch-depth: 0
+          persist-credentials: 'false'
+          token: ${{ secrets.REPO_CHECKOUT_TOKEN }}
+      - name: Prepare package
+        run: |
+          cargo package -p tfhe-hpu-backend
+      - uses: actions/upload-artifact@ea165f8d65b6e75b540449e92b4886f43607fa02 # v4.6.2
+        with:
+          name: crate
+          path: target/package/*.crate
+      - name: generate hash
+        id: hash
+        run: cd target/package && echo "hash=$(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-hpu-backend Release
+    runs-on: ubuntu-latest
+    needs: [verify_tag, package] # for comparing hashes
+    steps:
+      - name: Checkout
+        uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683 # v4.2.2
+        with:
+          fetch-depth: 0
+          persist-credentials: 'false'
+          token: ${{ secrets.REPO_CHECKOUT_TOKEN }}
+
+      - name: Publish crate.io package
+        env:
+          CRATES_TOKEN: ${{ secrets.CARGO_REGISTRY_TOKEN }}
+          DRY_RUN: ${{ inputs.dry_run && '--dry-run' || '' }}
+        run: |
+          # DRY_RUN expansion cannot be double quoted when variable contains empty string otherwise cargo publish 
+          # would fail. This is safe since DRY_RUN is handled in the env section above.
+          # shellcheck disable=SC2086
+          cargo publish -p tfhe-hpu-backend --token "${CRATES_TOKEN}" ${DRY_RUN}
+
+      - name: Generate hash
+        id: published_hash
+        run: cd target/package && echo "pub_hash=$(sha256sum ./*.crate | base64 -w0)" >> "${GITHUB_OUTPUT}"
+
+      - name: Slack notification (hashes comparison)
+        if: ${{ needs.package.outputs.hash != steps.published_hash.outputs.pub_hash }}
+        continue-on-error: true
+        uses: rtCamp/action-slack-notify@e31e87e03dd19038e411e38ae27cbad084a90661 # v2.3.3
+        env:
+          SLACK_COLOR: failure
+          SLACK_MESSAGE: "SLSA tfhe-hpu-backend crate - hash comparison failure: (${{ env.ACTION_RUN_URL }})"
+
+      - name: Slack Notification
+        if: ${{ failure() || (cancelled() && github.event_name != 'pull_request') }}
+        continue-on-error: true
+        uses: rtCamp/action-slack-notify@e31e87e03dd19038e411e38ae27cbad084a90661 # v2.3.3
+        env:
+          SLACK_COLOR: ${{ job.status }}
+          SLACK_MESSAGE: "tfhe-hpu-backend release failed: (${{ env.ACTION_RUN_URL }})"

.github/workflows/make_release_tfhe_csprng.yml

@@ -10,6 +10,12 @@ on:
 
 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 }}
+
+permissions: {}
 
 jobs:
   verify_tag:
@@ -27,6 +33,8 @@ jobs:
         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-csprng
@@ -64,9 +72,10 @@ jobs:
         uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683 # v4.2.2
         with:
           fetch-depth: 0
-          token: ${{ secrets.FHE_ACTIONS_TOKEN }}
+          persist-credentials: 'false'
+          token: ${{ secrets.REPO_CHECKOUT_TOKEN }}
       - name: Download artifact
-        uses: actions/download-artifact@95815c38cf2ff2164869cbab79da8d1f422bc89e # v4.2.1
+        uses: actions/download-artifact@d3f86a106a0bac45b974a628896c90dbdf5c8093 # v4.3.0
         with:
           name: crate-tfhe-csprng
           path: target/package
@@ -75,7 +84,10 @@ jobs:
           CRATES_TOKEN: ${{ secrets.CARGO_REGISTRY_TOKEN }}
           DRY_RUN: ${{ inputs.dry_run && '--dry-run' || '' }}
         run: |
-          cargo publish -p tfhe-csprng --token ${{ env.CRATES_TOKEN }} ${{ env.DRY_RUN }}
+          # DRY_RUN expansion cannot be double quoted when variable contains empty string otherwise cargo publish 
+          # would fail. This is safe since DRY_RUN is handled in the env section above.
+          # shellcheck disable=SC2086
+          cargo publish -p tfhe-csprng --token "${CRATES_TOKEN}" ${DRY_RUN}
       - name: Generate hash
         id: published_hash
         run: cd target/package && echo "pub_hash=$(sha256sum ./*.crate | base64 -w0)" >> "${GITHUB_OUTPUT}"
@@ -85,19 +97,11 @@ jobs:
         uses: rtCamp/action-slack-notify@e31e87e03dd19038e411e38ae27cbad084a90661 # v2.3.3
         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() }}
+        if: ${{ failure() || (cancelled() && github.event_name != 'pull_request') }}
         continue-on-error: true
         uses: rtCamp/action-slack-notify@e31e87e03dd19038e411e38ae27cbad084a90661 # v2.3.3
         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

@@ -11,6 +11,12 @@ on:
 
 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 }}
+
+permissions: {}
 
 jobs:
   verify_tag:
@@ -29,7 +35,8 @@ jobs:
         uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683 # v4.2.2
         with:
           fetch-depth: 0
-          token: ${{ secrets.FHE_ACTIONS_TOKEN }}
+          persist-credentials: 'false'
+          token: ${{ secrets.REPO_CHECKOUT_TOKEN }}
       - name: Prepare package
         run: |
           cargo package -p tfhe-fft
@@ -65,14 +72,18 @@ jobs:
         uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683 # v4.2.2
         with:
           fetch-depth: 0
-          token: ${{ secrets.FHE_ACTIONS_TOKEN }}
+          persist-credentials: 'false'
+          token: ${{ secrets.REPO_CHECKOUT_TOKEN }}
 
       - name: Publish crate.io package
         env:
           CRATES_TOKEN: ${{ secrets.CARGO_REGISTRY_TOKEN }}
           DRY_RUN: ${{ inputs.dry_run && '--dry-run' || '' }}
         run: |
-          cargo publish -p tfhe-fft --token ${{ env.CRATES_TOKEN }} ${{ env.DRY_RUN }}
+          # DRY_RUN expansion cannot be double quoted when variable contains empty string otherwise cargo publish 
+          # would fail. This is safe since DRY_RUN is handled in the env section above.
+          # shellcheck disable=SC2086
+          cargo publish -p tfhe-fft --token "${CRATES_TOKEN}" ${DRY_RUN}
 
       - name: Generate hash
         id: published_hash
@@ -84,20 +95,12 @@ jobs:
         uses: rtCamp/action-slack-notify@e31e87e03dd19038e411e38ae27cbad084a90661 # v2.3.3
         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() }}
+        if: ${{ failure() || (cancelled() && github.event_name != 'pull_request') }}
         continue-on-error: true
         uses: rtCamp/action-slack-notify@e31e87e03dd19038e411e38ae27cbad084a90661 # v2.3.3
         env:
           SLACK_COLOR: ${{ job.status }}
-          SLACK_CHANNEL: ${{ secrets.SLACK_CHANNEL }}
-          SLACK_ICON: https://pbs.twimg.com/profile_images/1274014582265298945/OjBKP9kn_400x400.png
           SLACK_MESSAGE: "tfhe-fft release failed: (${{ env.ACTION_RUN_URL }})"
-          SLACK_USERNAME: ${{ secrets.BOT_USERNAME }}
-          SLACK_WEBHOOK: ${{ secrets.SLACK_WEBHOOK }}

.github/workflows/make_release_tfhe_ntt.yml

@@ -11,6 +11,12 @@ on:
 
 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 }}
+
+permissions: {}
 
 jobs:
   verify_tag:
@@ -29,7 +35,8 @@ jobs:
         uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683 # v4.2.2
         with:
           fetch-depth: 0
-          token: ${{ secrets.FHE_ACTIONS_TOKEN }}
+          persist-credentials: 'false'
+          token: ${{ secrets.REPO_CHECKOUT_TOKEN }}
       - name: Prepare package
         run: |
           cargo package -p tfhe-ntt
@@ -65,13 +72,18 @@ jobs:
         uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683 # v4.2.2
         with:
           fetch-depth: 0
+          persist-credentials: 'false'
+          token: ${{ secrets.REPO_CHECKOUT_TOKEN }}
 
       - name: Publish crate.io package
         env:
           CRATES_TOKEN: ${{ secrets.CARGO_REGISTRY_TOKEN }}
           DRY_RUN: ${{ inputs.dry_run && '--dry-run' || '' }}
         run: |
-          cargo publish -p tfhe-ntt --token ${{ env.CRATES_TOKEN }} ${{ env.DRY_RUN }}
+          # DRY_RUN expansion cannot be double quoted when variable contains empty string otherwise cargo publish 
+          # would fail. This is safe since DRY_RUN is handled in the env section above.
+          # shellcheck disable=SC2086
+          cargo publish -p tfhe-ntt --token "${CRATES_TOKEN}" ${DRY_RUN}
 
       - name: Generate hash
         id: published_hash
@@ -83,20 +95,12 @@ jobs:
         uses: rtCamp/action-slack-notify@e31e87e03dd19038e411e38ae27cbad084a90661 # v2.3.3
         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() }}
+        if: ${{ failure() || (cancelled() && github.event_name != 'pull_request') }}
         continue-on-error: true
         uses: rtCamp/action-slack-notify@e31e87e03dd19038e411e38ae27cbad084a90661 # v2.3.3
         env:
           SLACK_COLOR: ${{ job.status }}
-          SLACK_CHANNEL: ${{ secrets.SLACK_CHANNEL }}
-          SLACK_ICON: https://pbs.twimg.com/profile_images/1274014582265298945/OjBKP9kn_400x400.png
           SLACK_MESSAGE: "tfhe-ntt release failed: (${{ env.ACTION_RUN_URL }})"
-          SLACK_USERNAME: ${{ secrets.BOT_USERNAME }}
-          SLACK_WEBHOOK: ${{ secrets.SLACK_WEBHOOK }}

.github/workflows/make_release_tfhe_versionable.yml

@@ -10,6 +10,8 @@ env:
   SLACK_USERNAME: ${{ secrets.BOT_USERNAME }}
   SLACK_WEBHOOK: ${{ secrets.SLACK_WEBHOOK }}
 
+permissions: {}
+
 jobs:
   verify_tag:
     uses: ./.github/workflows/verify_tagged_commit.yml
@@ -27,6 +29,8 @@ jobs:
         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-versionable-derive
@@ -64,7 +68,7 @@ jobs:
           persist-credentials: 'false'
           token: ${{ secrets.REPO_CHECKOUT_TOKEN }}
       - name: Download artifact
-        uses: actions/download-artifact@95815c38cf2ff2164869cbab79da8d1f422bc89e # v4.2.1
+        uses: actions/download-artifact@d3f86a106a0bac45b974a628896c90dbdf5c8093 # v4.3.0
         with:
           name: crate-tfhe-versionable-derive
           path: target/package
@@ -72,7 +76,7 @@ jobs:
         env:
           CRATES_TOKEN: ${{ secrets.CARGO_REGISTRY_TOKEN }}
         run: |
-          cargo publish -p tfhe-versionable-derive --token ${{ env.CRATES_TOKEN }}
+          cargo publish -p tfhe-versionable-derive --token "${CRATES_TOKEN}"
       - name: Generate hash
         id: published_hash
         run: cd target/package && echo "pub_hash=$(sha256sum ./*.crate | base64 -w0)" >> "${GITHUB_OUTPUT}"
@@ -84,7 +88,7 @@ jobs:
           SLACK_COLOR: failure
           SLACK_MESSAGE: "SLSA tfhe-versionable-derive - hash comparison failure: (${{ env.ACTION_RUN_URL }})"
       - name: Slack Notification
-        if: ${{ failure() }}
+        if: ${{ failure() || (cancelled() && github.event_name != 'pull_request') }}
         continue-on-error: true
         uses: rtCamp/action-slack-notify@e31e87e03dd19038e411e38ae27cbad084a90661 # v2.3.3
         env:
@@ -102,6 +106,8 @@ jobs:
         uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
         with:
           fetch-depth: 0
+          persist-credentials: 'false'
+          token: ${{ secrets.REPO_CHECKOUT_TOKEN }}
       - name: Prepare package
         run: |
           cargo package -p tfhe-versionable
@@ -136,8 +142,10 @@ jobs:
         uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
         with:
           fetch-depth: 0
+          persist-credentials: 'false'
+          token: ${{ secrets.REPO_CHECKOUT_TOKEN }}
       - name: Download artifact
-        uses: actions/download-artifact@95815c38cf2ff2164869cbab79da8d1f422bc89e # v4.2.1
+        uses: actions/download-artifact@d3f86a106a0bac45b974a628896c90dbdf5c8093 # v4.3.0
         with:
           name: crate-tfhe-versionable
           path: target/package
@@ -145,7 +153,7 @@ jobs:
         env:
           CRATES_TOKEN: ${{ secrets.CARGO_REGISTRY_TOKEN }}
         run: |
-          cargo publish -p tfhe-versionable --token ${{ env.CRATES_TOKEN }}
+          cargo publish -p tfhe-versionable --token "${CRATES_TOKEN}"
       - name: Generate hash
         id: published_hash
         run: cd target/package && echo "pub_hash=$(sha256sum ./*.crate | base64 -w0)" >> "${GITHUB_OUTPUT}"
@@ -157,7 +165,7 @@ jobs:
           SLACK_COLOR: failure
           SLACK_MESSAGE: "SLSA tfhe-versionable - hash comparison failure: (${{ env.ACTION_RUN_URL }})"
       - name: Slack Notification
-        if: ${{ failure() }}
+        if: ${{ failure() || (cancelled() && github.event_name != 'pull_request') }}
         continue-on-error: true
         uses: rtCamp/action-slack-notify@e31e87e03dd19038e411e38ae27cbad084a90661 # v2.3.3
         env:

.github/workflows/make_release_zk_pok.yml

@@ -10,6 +10,12 @@ on:
 
 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 }}
+
+permissions: {}
 
 jobs:
   package:
@@ -21,6 +27,8 @@ jobs:
           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-zk-pok
@@ -64,7 +72,7 @@ jobs:
           persist-credentials: 'false'
           token: ${{ secrets.REPO_CHECKOUT_TOKEN }}
       - name: Download artifact
-        uses: actions/download-artifact@95815c38cf2ff2164869cbab79da8d1f422bc89e # v4.2.1
+        uses: actions/download-artifact@d3f86a106a0bac45b974a628896c90dbdf5c8093 # v4.3.0
         with:
           name: crate-zk-pok
           path: target/package
@@ -73,7 +81,10 @@ jobs:
           CRATES_TOKEN: ${{ secrets.CARGO_REGISTRY_TOKEN }}
           DRY_RUN: ${{ inputs.dry_run && '--dry-run' || '' }}
         run: |
-          cargo publish -p tfhe-zk-pok --token ${{ env.CRATES_TOKEN }} ${{ env.DRY_RUN }}
+          # DRY_RUN expansion cannot be double quoted when variable contains empty string otherwise cargo publish 
+          # would fail. This is safe since DRY_RUN is handled in the env section above.
+          # shellcheck disable=SC2086
+          cargo publish -p tfhe-zk-pok --token "${CRATES_TOKEN}" ${DRY_RUN}
       - name: Verify hash
         id: published_hash
         run: cd target/package && echo "pub_hash=$(sha256sum ./*.crate | base64 -w0)" >> "${GITHUB_OUTPUT}"
@@ -83,19 +94,11 @@ jobs:
         uses: rtCamp/action-slack-notify@e31e87e03dd19038e411e38ae27cbad084a90661 # v2.3.3
         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() }}
+        if: ${{ failure() || (cancelled() && github.event_name != 'pull_request') }}
         continue-on-error: true
         uses: rtCamp/action-slack-notify@e31e87e03dd19038e411e38ae27cbad084a90661 # v2.3.3
         env:
           SLACK_COLOR: ${{ job.status }}
-          SLACK_CHANNEL: ${{ secrets.SLACK_CHANNEL }}
-          SLACK_ICON: https://pbs.twimg.com/profile_images/1274014582265298945/OjBKP9kn_400x400.png
           SLACK_MESSAGE: "tfhe-zk-pok release failed: (${{ env.ACTION_RUN_URL }})"
-          SLACK_USERNAME: ${{ secrets.BOT_USERNAME }}
-          SLACK_WEBHOOK: ${{ secrets.SLACK_WEBHOOK }}

.github/workflows/parameters_check.yml

@@ -12,12 +12,17 @@ on:
       - "main"
   workflow_dispatch:
 
+permissions: {}
+
 jobs:
   params-curves-security-check:
     runs-on: large_ubuntu_16-22.04
     steps:
       - name: Checkout tfhe-rs
         uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
+        with:
+          persist-credentials: 'false'
+          token: ${{ secrets.REPO_CHECKOUT_TOKEN }}
 
       - name: Checkout lattice-estimator
         uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
@@ -25,6 +30,7 @@ jobs:
           repository: malb/lattice-estimator
           path: lattice_estimator
           ref: 'e80ec6bbbba212428b0e92d0467c18629cf9ed67'
+          persist-credentials: 'false'
 
       - name: Install Sage
         run: |

.github/workflows/placeholder_workflow.yml

@@ -4,6 +4,8 @@ name: Placeholder Workflow
 on:
   workflow_dispatch:
 
+permissions: {}
+
 jobs:
   placeholder:
     name: Placeholder

.github/workflows/sync_on_push.yml

@@ -7,6 +7,8 @@ on:
       - 'main'
   workflow_dispatch:
 
+permissions: {}
+
 jobs:
   sync-repo:
     if: ${{ github.repository == 'zama-ai/tfhe-rs' }}

.github/workflows/verify_tagged_commit.yml

@@ -9,6 +9,8 @@ on:
       READ_ORG_TOKEN:
         required: true
 
+permissions: {}
+
 jobs:
   checks:
     runs-on: ubuntu-latest
@@ -26,7 +28,10 @@ jobs:
 
       - name: Actor authorized
         run: |
-          if [ "${{ steps.actor_check.outputs.authorized }}" == "false" ]; then
-            echo "Actor '${{ github.triggering_actor }}' is not authorized to perform release"
+          if [ "${ACTOR_CHECK_OUTPUT}" == "false" ]; then
+            echo "Actor '${TRIGGERING_ACTOR}' is not authorized to perform release"
             exit 1
           fi
+        env:
+          TRIGGERING_ACTOR: ${{ github.triggering_actor }}
+          ACTOR_CHECK_OUTPUT: ${{ steps.actor_check.outputs.authorized }}

.lfsconfig

@@ -0,0 +1,2 @@
+[lfs]
+  fetchexclude = *

CODEOWNERS

@@ -0,0 +1,18 @@
+# Specifying a path without code owners means that path won't have owners and is akin to a negation
+# i.e. the `core_crypto` dir is owned and needs owner approval/review, but not the `gpu` sub dir
+# See https://docs.github.com/en/repositories/managing-your-repositorys-settings-and-features/customizing-your-repository/about-code-owners#example-of-a-codeowners-file
+/tfhe/src/core_crypto/                  @IceTDrinker
+/tfhe/src/core_crypto/gpu
+
+/tfhe/src/shortint/                     @mayeul-zama
+
+/tfhe/src/integer/                      @tmontaigu
+/tfhe/src/integer/gpu
+
+/tfhe/src/high_level_api/               @tmontaigu
+
+/Makefile                               @IceTDrinker @soonum
+
+/.github/                               @soonum
+
+/CODEOWNERS                             @IceTDrinker

CONTRIBUTING.md

@@ -102,8 +102,7 @@ For example, if you made changes in `tfhe/src/integer/*`, you can test them with
 ## 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/).
+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
 
@@ -145,12 +144,15 @@ sequenceDiagram
     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
+{% hint style="info" %}
+
+## 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
+{% endhint %}
 
 ## 8. Data versioning
 

Cargo.toml

@@ -2,15 +2,19 @@
 resolver = "2"
 members = [
     "tfhe",
+    "tfhe-benchmark",
     "tfhe-fft",
     "tfhe-ntt",
     "tfhe-zk-pok",
     "tasks",
     "tfhe-csprng",
     "backends/tfhe-cuda-backend",
+    "backends/tfhe-hpu-backend",
     "utils/tfhe-versionable",
     "utils/tfhe-versionable-derive",
+    "utils/param_dedup",
     "tests",
+    "mockups/tfhe-hpu-mockup",
 ]
 
 exclude = [

Makefile

@@ -2,6 +2,7 @@ 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)
+CARGO_BUILD_JOBS=default
 CPU_COUNT=$(shell ./scripts/cpu_count.sh)
 RS_BUILD_TOOLCHAIN:=stable
 CARGO_RS_BUILD_TOOLCHAIN:=+$(RS_BUILD_TOOLCHAIN)
@@ -55,6 +56,9 @@ REGEX_PATTERN?=''
 TFHECUDA_SRC=backends/tfhe-cuda-backend/cuda
 TFHECUDA_BUILD=$(TFHECUDA_SRC)/build
 
+# tfhe-hpu-backend
+HPU_CONFIG=v80
+
 # Exclude these files from coverage reports
 define COVERAGE_EXCLUDED_FILES
 --exclude-files apps/trivium/src/trivium/* \
@@ -163,6 +167,12 @@ install_typos_checker: install_rs_build_toolchain
 	cargo $(CARGO_RS_BUILD_TOOLCHAIN) install typos-cli || \
 	( echo "Unable to install typos-cli, unknown error." && exit 1 )
 
+.PHONY: install_zizmor # Install zizmor workflow security checker
+install_zizmor: install_rs_build_toolchain
+	@zizmor --version > /dev/null 2>&1 || \
+	cargo $(CARGO_RS_BUILD_TOOLCHAIN) install zizmor || \
+	( echo "Unable to install zizmor, unknown error." && exit 1 )
+
 .PHONY: setup_venv # Setup Python virtualenv for wasm tests
 setup_venv:
 	python3 -m venv venv
@@ -284,7 +294,7 @@ check_typos: install_typos_checker
 .PHONY: clippy_gpu # Run clippy lints on tfhe with "gpu" enabled
 clippy_gpu: install_rs_check_toolchain
 	RUSTFLAGS="$(RUSTFLAGS)" cargo "$(CARGO_RS_CHECK_TOOLCHAIN)" clippy \
-		--features=boolean,shortint,integer,internal-keycache,gpu,pbs-stats,extended-types \
+		--features=boolean,shortint,integer,internal-keycache,gpu,pbs-stats,extended-types,zk-pok \
 		--all-targets \
 		-p $(TFHE_SPEC) -- --no-deps -D warnings
 
@@ -295,6 +305,20 @@ check_gpu: install_rs_check_toolchain
 		--all-targets \
 		-p $(TFHE_SPEC)
 
+.PHONY: clippy_hpu # Run clippy lints on tfhe with "hpu" enabled
+clippy_hpu: install_rs_check_toolchain
+	RUSTFLAGS="$(RUSTFLAGS)" cargo "$(CARGO_RS_CHECK_TOOLCHAIN)" clippy \
+		--features=boolean,shortint,integer,internal-keycache,hpu,pbs-stats,extended-types \
+		--all-targets \
+		-p $(TFHE_SPEC) -- --no-deps -D warnings
+
+.PHONY: clippy_gpu_hpu # Run clippy lints on tfhe with "gpu" and "hpu" enabled
+clippy_gpu_hpu: install_rs_check_toolchain
+	RUSTFLAGS="$(RUSTFLAGS)" cargo "$(CARGO_RS_CHECK_TOOLCHAIN)" clippy \
+		--features=boolean,shortint,integer,internal-keycache,gpu,hpu,pbs-stats,extended-types,zk-pok \
+		--all-targets \
+		-p $(TFHE_SPEC) -- --no-deps -D warnings
+
 .PHONY: fix_newline # Fix newline at end of file issues to be UNIX compliant
 fix_newline: check_linelint_installed
 	linelint -a .
@@ -307,6 +331,10 @@ check_newline: check_linelint_installed
 lint_workflow: check_actionlint_installed
 	actionlint
 
+.PHONY: check_workflow_security # Run zizmor security checker on GitHub workflows
+check_workflow_security: install_zizmor
+	zizmor --persona pedantic .
+
 .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 \
@@ -444,10 +472,15 @@ clippy_tfhe_lints: install_cargo_dylint # the toolchain is selected with toolcha
 	rustup toolchain install && \
 	cargo clippy --all-targets -- --no-deps -D warnings
 
+.PHONY: clippy_param_dedup # Run clippy lints on param_dedup tool
+clippy_param_dedup: install_rs_check_toolchain
+	RUSTFLAGS="$(RUSTFLAGS)" cargo "$(CARGO_RS_CHECK_TOOLCHAIN)" clippy --all-targets \
+		-p param_dedup -- --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_tfhe_lints clippy_ws_tests
+clippy_versionable clippy_tfhe_lints clippy_ws_tests clippy_bench clippy_param_dedup
 
 .PHONY: clippy_fast # Run main clippy targets
 clippy_fast: clippy_rustdoc clippy clippy_all_targets clippy_c_api clippy_js_wasm_api clippy_tasks \
@@ -458,6 +491,11 @@ clippy_cuda_backend: install_rs_check_toolchain
 	RUSTFLAGS="$(RUSTFLAGS)" cargo "$(CARGO_RS_CHECK_TOOLCHAIN)" clippy --all-targets \
 		-p tfhe-cuda-backend -- --no-deps -D warnings
 
+.PHONY: clippy_hpu_backend # Run clippy lints on the tfhe-hpu-backend
+clippy_hpu_backend: install_rs_check_toolchain
+	RUSTFLAGS="$(RUSTFLAGS)" cargo "$(CARGO_RS_CHECK_TOOLCHAIN)" clippy --all-targets \
+		-p tfhe-hpu-backend -- --no-deps -D warnings
+
 .PHONY: check_rust_bindings_did_not_change # Check rust bindings are up to date for tfhe-cuda-backend
 check_rust_bindings_did_not_change:
 	cargo build -p tfhe-cuda-backend && "$(MAKE)" fmt_gpu && \
@@ -687,6 +725,28 @@ test_signed_integer_multi_bit_gpu_ci: install_rs_check_toolchain install_cargo_n
 		--cargo-profile "$(CARGO_PROFILE)" --multi-bit --backend "gpu" \
 		--signed-only --tfhe-package "$(TFHE_SPEC)"
 
+.PHONY: test_integer_hpu_ci # Run the tests for integer ci on hpu backend
+test_integer_hpu_ci: install_rs_check_toolchain install_cargo_nextest
+	cargo test --release -p $(TFHE_SPEC) --features hpu-v80 --test hpu
+
+.PHONY: test_integer_hpu_mockup_ci # Run the tests for integer ci on hpu backend and mockup
+test_integer_hpu_mockup_ci: install_rs_check_toolchain install_cargo_nextest
+	source ./setup_hpu.sh --config sim ; \
+	cargo build --release --bin hpu_mockup; \
+    coproc target/release/hpu_mockup --params mockups/tfhe-hpu-mockup/params/tuniform_64b_pfail64_psi64.toml > mockup.log; \
+	HPU_TEST_ITER=1 \
+	cargo test --profile devo -p $(TFHE_SPEC) --features hpu --test hpu -- u32 && \
+	kill %1
+
+.PHONY: test_integer_hpu_mockup_ci_fast # Run the quick tests for integer ci on hpu backend and mockup.
+test_integer_hpu_mockup_ci_fast: install_rs_check_toolchain install_cargo_nextest
+	source ./setup_hpu.sh --config sim ; \
+	cargo build --profile devo --bin hpu_mockup; \
+    coproc target/devo/hpu_mockup --params mockups/tfhe-hpu-mockup/params/tuniform_64b_fast.toml > mockup.log; \
+	HPU_TEST_ITER=1 \
+	cargo test --profile devo -p $(TFHE_SPEC) --features hpu --test hpu -- u32 && \
+	kill %1
+
 .PHONY: test_boolean # Run the tests of the boolean module
 test_boolean: install_rs_build_toolchain
 	RUSTFLAGS="$(RUSTFLAGS)" cargo $(CARGO_RS_BUILD_TOOLCHAIN) test --profile $(CARGO_PROFILE) \
@@ -839,9 +899,25 @@ test_high_level_api: install_rs_build_toolchain
 
 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=integer,internal-keycache,gpu -p $(TFHE_SPEC) \
+		--test-threads=4 --features=integer,internal-keycache,gpu,zk-pok -p $(TFHE_SPEC) \
 		-E "test(/high_level_api::.*gpu.*/)"
 
+test_high_level_api_hpu: install_rs_build_toolchain install_cargo_nextest
+ifeq ($(HPU_CONFIG), v80)
+	RUSTFLAGS="$(RUSTFLAGS)" cargo $(CARGO_RS_BUILD_TOOLCHAIN) nextest run --cargo-profile $(CARGO_PROFILE) \
+		--build-jobs=$(CARGO_BUILD_JOBS) \
+		--test-threads=1 \
+		--features=integer,internal-keycache,hpu,hpu-v80 -p $(TFHE_SPEC) \
+		-E "test(/high_level_api::.*hpu.*/)"
+else
+	RUSTFLAGS="$(RUSTFLAGS)" cargo $(CARGO_RS_BUILD_TOOLCHAIN) nextest run --cargo-profile $(CARGO_PROFILE) \
+		--build-jobs=$(CARGO_BUILD_JOBS) \
+		--test-threads=1 \
+		--features=integer,internal-keycache,hpu -p $(TFHE_SPEC) \
+		-E "test(/high_level_api::.*hpu.*/)"
+endif
+
+
 .PHONY: test_strings # Run the tests for strings ci
 test_strings: install_rs_build_toolchain
 	RUSTFLAGS="$(RUSTFLAGS)" cargo $(CARGO_RS_BUILD_TOOLCHAIN) test --profile $(CARGO_PROFILE) \
@@ -959,8 +1035,8 @@ lint_doc: install_rs_check_toolchain
 lint_docs: lint_doc
 
 .PHONY: format_doc_latex # Format the documentation latex equations to avoid broken rendering.
-format_doc_latex:
-	RUSTFLAGS="" cargo xtask format_latex_doc
+format_doc_latex: install_rs_build_toolchain
+	RUSTFLAGS="" cargo "$(CARGO_RS_BUILD_TOOLCHAIN)" xtask format_latex_doc
 	@"$(MAKE)" --no-print-directory fmt
 	@printf "\n===============================\n\n"
 	@printf "Please manually inspect changes made by format_latex_doc, rustfmt can break equations \
@@ -968,8 +1044,8 @@ format_doc_latex:
 	@printf "\n===============================\n"
 
 .PHONY: check_md_docs_are_tested # Checks that the rust codeblocks in our .md files are tested
-check_md_docs_are_tested:
-	RUSTFLAGS="" cargo xtask check_tfhe_docs_are_tested
+check_md_docs_are_tested: install_rs_build_toolchain
+	RUSTFLAGS="" cargo "$(CARGO_RS_BUILD_TOOLCHAIN)" xtask check_tfhe_docs_are_tested
 
 .PHONY: check_intra_md_links # Checks broken internal links in Markdown docs
 check_intra_md_links: install_mlc
@@ -997,7 +1073,7 @@ 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=experimental,boolean,shortint,integer,internal-keycache,gpu \
+		--features=experimental,boolean,shortint,integer,internal-keycache,gpu,zk-pok \
 		-p $(TFHE_SPEC)
 	mkdir -p "$(TFHECUDA_BUILD)" && \
 		cd "$(TFHECUDA_BUILD)" && \
@@ -1073,6 +1149,24 @@ dieharder_csprng: install_dieharder build_tfhe_csprng
 # Benchmarks
 #
 
+.PHONY: clippy_bench # Run clippy lints on tfhe-benchmark
+clippy_bench: install_rs_check_toolchain
+	RUSTFLAGS="$(RUSTFLAGS)" cargo "$(CARGO_RS_CHECK_TOOLCHAIN)" clippy --all-targets \
+		--features=boolean,shortint,integer,internal-keycache,nightly-avx512,pbs-stats,zk-pok \
+		-p tfhe-benchmark -- --no-deps -D warnings
+
+.PHONY: clippy_bench_gpu # Run clippy lints on tfhe-benchmark
+clippy_bench_gpu: install_rs_check_toolchain
+	RUSTFLAGS="$(RUSTFLAGS)" cargo "$(CARGO_RS_CHECK_TOOLCHAIN)" clippy --all-targets \
+		--features=gpu,shortint,integer,internal-keycache,nightly-avx512,pbs-stats,zk-pok \
+		-p tfhe-benchmark -- --no-deps -D warnings
+
+.PHONY: clippy_bench_hpu # Run clippy lints on tfhe-benchmark
+clippy_bench_hpu: install_rs_check_toolchain
+	RUSTFLAGS="$(RUSTFLAGS)" cargo "$(CARGO_RS_CHECK_TOOLCHAIN)" clippy --all-targets \
+		--features=hpu,shortint,integer,internal-keycache,pbs-stats\
+		-p tfhe-benchmark -- --no-deps -D warnings
+
 .PHONY: print_doc_bench_parameters # Print parameters used in doc benchmarks
 print_doc_bench_parameters:
 	RUSTFLAGS="" cargo run --example print_doc_bench_parameters \
@@ -1083,49 +1177,57 @@ 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=integer,internal-keycache,nightly-avx512,pbs-stats -p $(TFHE_SPEC) --
+	--features=integer,internal-keycache,nightly-avx512,pbs-stats -p tfhe-benchmark --
 
 .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=integer,internal-keycache,nightly-avx512,pbs-stats -p $(TFHE_SPEC) --
+	--features=integer,internal-keycache,nightly-avx512,pbs-stats -p tfhe-benchmark --
 
 .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=integer,gpu,internal-keycache,nightly-avx512,pbs-stats -p $(TFHE_SPEC) --
+	--features=integer,gpu,internal-keycache,nightly-avx512,pbs-stats -p tfhe-benchmark --
 
 .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) --
+	--features=integer,gpu,internal-keycache,nightly-avx512,pbs-stats -p tfhe-benchmark --
+
+.PHONY: bench_integer_hpu # Run benchmarks for integer on HPU backend
+bench_integer_hpu: install_rs_check_toolchain
+	source ./setup_hpu.sh --config $(HPU_CONFIG) ; \
+	RUSTFLAGS="$(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=integer,internal-keycache,pbs-stats,hpu,hpu-v80 -p tfhe-benchmark -- --quick
 
 .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=integer,internal-keycache,nightly-avx512,pbs-stats -p $(TFHE_SPEC) --
+	--features=integer,internal-keycache,nightly-avx512,pbs-stats -p tfhe-benchmark --
 
 .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=integer,internal-keycache,gpu,pbs-stats -p $(TFHE_SPEC) --
-	
+	--features=integer,internal-keycache,gpu,pbs-stats -p tfhe-benchmark --
+
 .PHONY: bench_integer_zk_gpu
 bench_integer_zk_gpu: install_rs_check_toolchain
 	RUSTFLAGS="$(RUSTFLAGS)" __TFHE_RS_BENCH_TYPE=$(BENCH_TYPE) \
 	cargo $(CARGO_RS_CHECK_TOOLCHAIN) bench \
 	--bench zk-pke-bench \
-	--features=integer,internal-keycache,gpu,pbs-stats,zk-pok -p $(TFHE_SPEC) --
+	--features=integer,internal-keycache,gpu,pbs-stats,zk-pok -p tfhe-benchmark --
 
 .PHONY: bench_integer_multi_bit # Run benchmarks for unsigned integer using multi-bit parameters
 bench_integer_multi_bit: install_rs_check_toolchain
@@ -1133,7 +1235,7 @@ 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=integer,internal-keycache,nightly-avx512,pbs-stats -p $(TFHE_SPEC) --
+	--features=integer,internal-keycache,nightly-avx512,pbs-stats -p tfhe-benchmark --
 
 .PHONY: bench_signed_integer_multi_bit # Run benchmarks for signed integer using multi-bit parameters
 bench_signed_integer_multi_bit: install_rs_check_toolchain
@@ -1141,7 +1243,7 @@ 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=integer,internal-keycache,nightly-avx512,pbs-stats -p $(TFHE_SPEC) --
+	--features=integer,internal-keycache,nightly-avx512,pbs-stats -p tfhe-benchmark --
 
 .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
@@ -1149,7 +1251,7 @@ 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=integer,gpu,internal-keycache,nightly-avx512,pbs-stats -p $(TFHE_SPEC) --
+	--features=integer,gpu,internal-keycache,nightly-avx512,pbs-stats -p tfhe-benchmark --
 
 .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
@@ -1157,7 +1259,7 @@ bench_signed_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-signed-bench \
-	--features=integer,gpu,internal-keycache,nightly-avx512,pbs-stats -p $(TFHE_SPEC) --
+	--features=integer,gpu,internal-keycache,nightly-avx512,pbs-stats -p tfhe-benchmark --
 
 .PHONY: bench_integer_zk # Run benchmarks for integer encryption with ZK proofs
 bench_integer_zk: install_rs_check_toolchain
@@ -1165,91 +1267,83 @@ bench_integer_zk: install_rs_check_toolchain
 	cargo $(CARGO_RS_CHECK_TOOLCHAIN) bench \
 	--bench zk-pke-bench \
 	--features=integer,internal-keycache,zk-pok,nightly-avx512,pbs-stats \
-	-p $(TFHE_SPEC) --
+	-p tfhe-benchmark --
 
 .PHONY: bench_shortint # Run benchmarks for shortint
 bench_shortint: install_rs_check_toolchain
-	RUSTFLAGS="$(RUSTFLAGS)" __TFHE_RS_BENCH_OP_FLAVOR=$(BENCH_OP_FLAVOR) \
+	RUSTFLAGS="$(RUSTFLAGS)" __TFHE_RS_BENCH_OP_FLAVOR=$(BENCH_OP_FLAVOR) __TFHE_RS_PARAMS_SET=$(BENCH_PARAMS_SET) __TFHE_RS_BENCH_TYPE=$(BENCH_TYPE) \
 	cargo $(CARGO_RS_CHECK_TOOLCHAIN) bench \
 	--bench shortint-bench \
-	--features=shortint,internal-keycache,nightly-avx512 -p $(TFHE_SPEC)
+	--features=shortint,internal-keycache,nightly-avx512 -p tfhe-benchmark
 
 .PHONY: bench_shortint_oprf # Run benchmarks for shortint
 bench_shortint_oprf: install_rs_check_toolchain
-	RUSTFLAGS="$(RUSTFLAGS)" \
+	RUSTFLAGS="$(RUSTFLAGS)" __TFHE_RS_PARAMS_SET=$(BENCH_PARAMS_SET) \
 	cargo $(CARGO_RS_CHECK_TOOLCHAIN) bench \
 	--bench oprf-shortint-bench \
-	--features=shortint,internal-keycache,nightly-avx512 -p $(TFHE_SPEC)
-
-.PHONY: bench_shortint_multi_bit # Run benchmarks for shortint using multi-bit parameters
-bench_shortint_multi_bit: install_rs_check_toolchain
-	RUSTFLAGS="$(RUSTFLAGS)" __TFHE_RS_PARAM_TYPE=MULTI_BIT \
-	__TFHE_RS_BENCH_OP_FLAVOR=$(BENCH_OP_FLAVOR) \
-	cargo $(CARGO_RS_CHECK_TOOLCHAIN) bench \
-	--bench shortint-bench \
-	--features=shortint,internal-keycache,nightly-avx512 -p $(TFHE_SPEC) --
+	--features=shortint,internal-keycache,nightly-avx512 -p tfhe-benchmark
 
 .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=boolean,internal-keycache,nightly-avx512 -p $(TFHE_SPEC)
+	--features=boolean,internal-keycache,nightly-avx512 -p tfhe-benchmark
 
 .PHONY: bench_ks # Run benchmarks for keyswitch
 bench_ks: install_rs_check_toolchain
 	RUSTFLAGS="$(RUSTFLAGS)" __TFHE_RS_PARAM_TYPE=$(BENCH_PARAM_TYPE) __TFHE_RS_PARAMS_SET=$(BENCH_PARAMS_SET) __TFHE_RS_BENCH_TYPE=$(BENCH_TYPE) \
 	cargo $(CARGO_RS_CHECK_TOOLCHAIN) bench \
 	--bench ks-bench \
-	--features=boolean,shortint,internal-keycache,nightly-avx512 -p $(TFHE_SPEC)
+	--features=boolean,shortint,internal-keycache,nightly-avx512 -p tfhe-benchmark
 
 .PHONY: bench_ks_gpu # Run benchmarks for keyswitch on GPU backend
 bench_ks_gpu: install_rs_check_toolchain
 	RUSTFLAGS="$(RUSTFLAGS)" __TFHE_RS_PARAM_TYPE=$(BENCH_PARAM_TYPE) __TFHE_RS_PARAMS_SET=$(BENCH_PARAMS_SET) __TFHE_RS_BENCH_TYPE=$(BENCH_TYPE) \
 	cargo $(CARGO_RS_CHECK_TOOLCHAIN) bench \
 	--bench ks-bench \
-	--features=boolean,shortint,gpu,internal-keycache,nightly-avx512 -p $(TFHE_SPEC)
+	--features=boolean,shortint,gpu,internal-keycache,nightly-avx512 -p tfhe-benchmark
 
 .PHONY: bench_pbs # Run benchmarks for PBS
 bench_pbs: install_rs_check_toolchain
 	RUSTFLAGS="$(RUSTFLAGS)" __TFHE_RS_PARAM_TYPE=$(BENCH_PARAM_TYPE) __TFHE_RS_PARAMS_SET=$(BENCH_PARAMS_SET) __TFHE_RS_BENCH_TYPE=$(BENCH_TYPE) \
 	cargo $(CARGO_RS_CHECK_TOOLCHAIN) bench \
 	--bench pbs-bench \
-	--features=boolean,shortint,internal-keycache,nightly-avx512 -p $(TFHE_SPEC)
+	--features=boolean,shortint,internal-keycache,nightly-avx512 -p tfhe-benchmark
 
 .PHONY: bench_pbs_gpu # Run benchmarks for PBS on GPU backend
 bench_pbs_gpu: install_rs_check_toolchain
 	RUSTFLAGS="$(RUSTFLAGS)" __TFHE_RS_PARAM_TYPE=$(BENCH_PARAM_TYPE) __TFHE_RS_FAST_BENCH=$(FAST_BENCH) __TFHE_RS_PARAMS_SET=$(BENCH_PARAMS_SET) __TFHE_RS_BENCH_TYPE=$(BENCH_TYPE) \
 	cargo $(CARGO_RS_CHECK_TOOLCHAIN) bench \
 	--bench pbs-bench \
-	--features=boolean,shortint,gpu,internal-keycache,nightly-avx512 -p $(TFHE_SPEC)
+	--features=boolean,shortint,gpu,internal-keycache,nightly-avx512 -p tfhe-benchmark
 
 .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=$(BENCH_PARAMS_SET) __TFHE_RS_BENCH_TYPE=$(BENCH_TYPE) \
 	cargo $(CARGO_RS_CHECK_TOOLCHAIN) bench \
 	--bench ks-pbs-bench \
-	--features=boolean,shortint,internal-keycache,nightly-avx512 -p $(TFHE_SPEC)
+	--features=boolean,shortint,internal-keycache,nightly-avx512 -p tfhe-benchmark
 
 .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=$(BENCH_PARAMS_SET) __TFHE_RS_BENCH_TYPE=$(BENCH_TYPE) \
 	cargo $(CARGO_RS_CHECK_TOOLCHAIN) bench \
 	--bench ks-pbs-bench \
-	--features=boolean,shortint,gpu,internal-keycache,nightly-avx512 -p $(TFHE_SPEC)
+	--features=boolean,shortint,gpu,internal-keycache,nightly-avx512 -p tfhe-benchmark
 
 .PHONY: bench_pbs128 # Run benchmarks for PBS using FFT 128 bits
 bench_pbs128: install_rs_check_toolchain
 	RUSTFLAGS="$(RUSTFLAGS)" __TFHE_RS_BENCH_TYPE=$(BENCH_TYPE) \
 	cargo $(CARGO_RS_CHECK_TOOLCHAIN) bench \
 	--bench pbs128-bench \
-	--features=boolean,shortint,internal-keycache,nightly-avx512 -p $(TFHE_SPEC)
+	--features=boolean,shortint,internal-keycache,nightly-avx512 -p tfhe-benchmark
 
 .PHONY: bench_pbs128_gpu # Run benchmarks for PBS using FFT 128 bits on GPU
 bench_pbs128_gpu: install_rs_check_toolchain
 	RUSTFLAGS="$(RUSTFLAGS)" __TFHE_RS_BENCH_TYPE=$(BENCH_TYPE) \
 	cargo $(CARGO_RS_CHECK_TOOLCHAIN) bench \
 	--bench pbs128-bench \
-	--features=boolean,shortint,gpu,internal-keycache,nightly-avx512 -p $(TFHE_SPEC)
+	--features=boolean,shortint,gpu,internal-keycache,nightly-avx512 -p tfhe-benchmark
 
 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"
@@ -1281,17 +1375,37 @@ bench_web_js_api_parallel_firefox_ci: setup_venv
 	nvm use $(NODE_VERSION) && \
 	$(MAKE) bench_web_js_api_parallel_firefox
 
-.PHONY: bench_hlapi_erc20 # Run benchmarks for ECR20 operations
+.PHONY: bench_hlapi_erc20 # Run benchmarks for ERC20 operations
 bench_hlapi_erc20: install_rs_check_toolchain
 	RUSTFLAGS="$(RUSTFLAGS)" cargo $(CARGO_RS_CHECK_TOOLCHAIN) bench \
 	--bench hlapi-erc20 \
-	--features=integer,internal-keycache,pbs-stats,nightly-avx512 -p $(TFHE_SPEC) --
+	--features=integer,internal-keycache,pbs-stats,nightly-avx512 -p tfhe-benchmark --
 
-.PHONY: bench_hlapi_erc20_gpu # Run benchmarks for ECR20 operations on GPU
+.PHONY: bench_hlapi_erc20_gpu # Run benchmarks for ERC20 operations on GPU
 bench_hlapi_erc20_gpu: install_rs_check_toolchain
 	RUSTFLAGS="$(RUSTFLAGS)" cargo $(CARGO_RS_CHECK_TOOLCHAIN) bench \
 	--bench hlapi-erc20 \
-	--features=integer,gpu,internal-keycache,pbs-stats,nightly-avx512 -p $(TFHE_SPEC) --
+	--features=integer,gpu,internal-keycache,pbs-stats,nightly-avx512 -p tfhe-benchmark --
+
+.PHONY: bench_hlapi_dex # Run benchmarks for DEX operations
+bench_hlapi_dex: install_rs_check_toolchain
+	RUSTFLAGS="$(RUSTFLAGS)" cargo $(CARGO_RS_CHECK_TOOLCHAIN) bench \
+	--bench hlapi-dex \
+	--features=integer,internal-keycache,pbs-stats,nightly-avx512 -p tfhe-benchmark --
+
+.PHONY: bench_hlapi_dex_gpu # Run benchmarks for DEX operations on GPU
+bench_hlapi_dex_gpu: install_rs_check_toolchain
+	RUSTFLAGS="$(RUSTFLAGS)" cargo $(CARGO_RS_CHECK_TOOLCHAIN) bench \
+	--bench hlapi-dex \
+	--features=integer,gpu,internal-keycache,pbs-stats,nightly-avx512 -p tfhe-benchmark --
+
+.PHONY: bench_hlapi_erc20_hpu # Run benchmarks for ECR20 operations on HPU
+bench_hlapi_erc20_hpu: install_rs_check_toolchain
+	source ./setup_hpu.sh --config $(HPU_CONFIG) ; \
+	RUSTFLAGS="$(RUSTFLAGS)" \
+	cargo $(CARGO_RS_CHECK_TOOLCHAIN) bench \
+	--bench hlapi-erc20 \
+	--features=integer,internal-keycache,hpu,hpu-v80 -p tfhe-benchmark -- --quick
 
 .PHONY: bench_tfhe_zk_pok # Run benchmarks for the tfhe_zk_pok crate
 bench_tfhe_zk_pok: install_rs_check_toolchain
@@ -1318,20 +1432,23 @@ gen_key_cache_core_crypto: install_rs_build_toolchain
 .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=integer,internal-keycache
+	--bin hlapi_compact_pk_ct_sizes \
+	--features=integer,internal-keycache \
+	-p tfhe-benchmark
 
 .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=shortint,internal-keycache
+	--bin shortint_key_sizes \
+	--features=shortint,internal-keycache \
+	-p tfhe-benchmark
 
 .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=boolean,internal-keycache
+	--bin boolean_key_sizes \
+	--features=boolean,internal-keycache \
+	-p tfhe-benchmark
 
 .PHONY: parse_integer_benches # Run python parser to output a csv containing integer benches data
 parse_integer_benches:
@@ -1342,8 +1459,9 @@ parse_integer_benches:
 .PHONY: parse_wasm_benchmarks # Parse benchmarks performed with WASM web client into a CSV file
 parse_wasm_benchmarks: install_rs_check_toolchain
 	RUSTFLAGS="$(RUSTFLAGS)" cargo $(CARGO_RS_CHECK_TOOLCHAIN) run --profile $(CARGO_PROFILE) \
-	--example wasm_benchmarks_parser \
+	--bin wasm_benchmarks_parser \
 	--features=shortint,internal-keycache \
+	-p tfhe-benchmark \
 	-- wasm_benchmark_results.json
 
 .PHONY: write_params_to_file # Gather all crypto parameters into a file with a Sage readable format.
@@ -1386,7 +1504,10 @@ tfhe_lints
 
 .PHONY: pcc_gpu # pcc stands for pre commit checks for GPU compilation
 pcc_gpu: check_rust_bindings_did_not_change clippy_rustdoc_gpu \
-clippy_gpu clippy_cuda_backend check_compile_tests_benches_gpu
+clippy_gpu clippy_cuda_backend clippy_bench_gpu check_compile_tests_benches_gpu
+
+.PHONY: pcc_hpu # pcc stands for pre commit checks for HPU compilation
+pcc_hpu: clippy_hpu clippy_hpu_backend test_integer_hpu_mockup_ci_fast
 
 .PHONY: fpcc # pcc stands for pre commit checks, the f stands for fast
 fpcc: no_tfhe_typo no_dbg_log check_parameter_export_ok check_fmt check_typos lint_doc \

README.md

@@ -206,6 +206,7 @@ If you want to work within the IND-CPA security model, which is less strict than
 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

_typos.toml

@@ -11,11 +11,13 @@ extend-ignore-identifiers-re = [
     # Example with string replacing "hello" with "herlo"
     "herlo",
     # Example in trivium
-    "C9217BA0D762ACA1"
+    "C9217BA0D762ACA1",
+    "0x[0-9a-fA-F]+"
 ]
 
 [files]
 extend-exclude = [
     "backends/tfhe-cuda-backend/cuda/src/fft128/twiddles.cu",
     "backends/tfhe-cuda-backend/cuda/src/fft/twiddles.cu",
+    "backends/tfhe-hpu-backend/config_store/**/*.link_summary",
 ]

apps/trivium/README.md

@@ -129,7 +129,7 @@ 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 `V1_1_PARAM_MESSAGE_1_CARRY_1_KS_PBS_GAUSSIAN_2M128`).
+The same implementation is also available for generic Ciphertexts representing bits (meant to be used with parameters `V1_2_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.
@@ -137,10 +137,10 @@ its setup a little more intricate.
 Example code:
 ```rust
 use tfhe::shortint::prelude::*;
-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::shortint::parameters::v1_2::{
+    V1_2_PARAM_MESSAGE_1_CARRY_1_KS_PBS_GAUSSIAN_2M128,
+    V1_2_PARAM_MESSAGE_2_CARRY_2_KS_PBS_GAUSSIAN_2M128,
+    V1_2_PARAM_KEYSWITCH_1_1_KS_PBS_TO_2_2_KS_PBS_GAUSSIAN_2M128,
 };
 use tfhe::{ConfigBuilder, generate_keys, FheUint64};
 use tfhe::prelude::*;
@@ -148,17 +148,17 @@ use tfhe_trivium::TriviumStreamShortint;
 
 fn test_shortint() {
     let config = ConfigBuilder::default()
-        .use_custom_parameters(V1_1_PARAM_MESSAGE_2_CARRY_2_KS_PBS_GAUSSIAN_2M128)
+        .use_custom_parameters(V1_2_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(V1_1_PARAM_MESSAGE_1_CARRY_1_KS_PBS_GAUSSIAN_2M128);
+    let (client_key, server_key): (ClientKey, ServerKey) = gen_keys(V1_2_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,
+        V1_2_PARAM_KEYSWITCH_1_1_KS_PBS_TO_2_2_KS_PBS_GAUSSIAN_2M128_2M128,
     );
 
     let key_string = "0053A6F94C9FF24598EB".to_string();

apps/trivium/benches/kreyvium_shortint.rs

@@ -1,9 +1,9 @@
 use criterion::Criterion;
 use tfhe::prelude::*;
-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::parameters::v1_2::{
+    V1_2_PARAM_KEYSWITCH_1_1_KS_PBS_TO_2_2_KS_PBS_GAUSSIAN_2M128,
+    V1_2_PARAM_MESSAGE_1_CARRY_1_KS_PBS_GAUSSIAN_2M128,
+    V1_2_PARAM_MESSAGE_2_CARRY_2_KS_PBS_GAUSSIAN_2M128,
 };
 use tfhe::shortint::prelude::*;
 use tfhe::{generate_keys, ConfigBuilder, FheUint64};
@@ -11,19 +11,19 @@ use tfhe_trivium::{KreyviumStreamShortint, TransCiphering};
 
 pub fn kreyvium_shortint_warmup(c: &mut Criterion) {
     let config = ConfigBuilder::default()
-        .use_custom_parameters(V1_1_PARAM_MESSAGE_2_CARRY_2_KS_PBS_GAUSSIAN_2M128)
+        .use_custom_parameters(V1_2_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(V1_1_PARAM_MESSAGE_1_CARRY_1_KS_PBS_GAUSSIAN_2M128);
+        gen_keys(V1_2_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,
+        V1_2_PARAM_KEYSWITCH_1_1_KS_PBS_TO_2_2_KS_PBS_GAUSSIAN_2M128,
     );
 
     let key_string = "0053A6F94C9FF24598EB000000000000".to_string();
@@ -64,19 +64,19 @@ pub fn kreyvium_shortint_warmup(c: &mut Criterion) {
 
 pub fn kreyvium_shortint_gen(c: &mut Criterion) {
     let config = ConfigBuilder::default()
-        .use_custom_parameters(V1_1_PARAM_MESSAGE_2_CARRY_2_KS_PBS_GAUSSIAN_2M128)
+        .use_custom_parameters(V1_2_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(V1_1_PARAM_MESSAGE_1_CARRY_1_KS_PBS_GAUSSIAN_2M128);
+        gen_keys(V1_2_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,
+        V1_2_PARAM_KEYSWITCH_1_1_KS_PBS_TO_2_2_KS_PBS_GAUSSIAN_2M128,
     );
 
     let key_string = "0053A6F94C9FF24598EB000000000000".to_string();
@@ -112,19 +112,19 @@ pub fn kreyvium_shortint_gen(c: &mut Criterion) {
 
 pub fn kreyvium_shortint_trans(c: &mut Criterion) {
     let config = ConfigBuilder::default()
-        .use_custom_parameters(V1_1_PARAM_MESSAGE_2_CARRY_2_KS_PBS_GAUSSIAN_2M128)
+        .use_custom_parameters(V1_2_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(V1_1_PARAM_MESSAGE_1_CARRY_1_KS_PBS_GAUSSIAN_2M128);
+        gen_keys(V1_2_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,
+        V1_2_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

@@ -1,9 +1,9 @@
 use criterion::Criterion;
 use tfhe::prelude::*;
-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::parameters::v1_2::{
+    V1_2_PARAM_KEYSWITCH_1_1_KS_PBS_TO_2_2_KS_PBS_GAUSSIAN_2M128,
+    V1_2_PARAM_MESSAGE_1_CARRY_1_KS_PBS_GAUSSIAN_2M128,
+    V1_2_PARAM_MESSAGE_2_CARRY_2_KS_PBS_GAUSSIAN_2M128,
 };
 use tfhe::shortint::prelude::*;
 use tfhe::{generate_keys, ConfigBuilder, FheUint64};
@@ -11,19 +11,19 @@ use tfhe_trivium::{TransCiphering, TriviumStreamShortint};
 
 pub fn trivium_shortint_warmup(c: &mut Criterion) {
     let config = ConfigBuilder::default()
-        .use_custom_parameters(V1_1_PARAM_MESSAGE_2_CARRY_2_KS_PBS_GAUSSIAN_2M128)
+        .use_custom_parameters(V1_2_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(V1_1_PARAM_MESSAGE_1_CARRY_1_KS_PBS_GAUSSIAN_2M128);
+        gen_keys(V1_2_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,
+        V1_2_PARAM_KEYSWITCH_1_1_KS_PBS_TO_2_2_KS_PBS_GAUSSIAN_2M128,
     );
 
     let key_string = "0053A6F94C9FF24598EB".to_string();
@@ -64,19 +64,19 @@ pub fn trivium_shortint_warmup(c: &mut Criterion) {
 
 pub fn trivium_shortint_gen(c: &mut Criterion) {
     let config = ConfigBuilder::default()
-        .use_custom_parameters(V1_1_PARAM_MESSAGE_2_CARRY_2_KS_PBS_GAUSSIAN_2M128)
+        .use_custom_parameters(V1_2_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(V1_1_PARAM_MESSAGE_1_CARRY_1_KS_PBS_GAUSSIAN_2M128);
+        gen_keys(V1_2_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,
+        V1_2_PARAM_KEYSWITCH_1_1_KS_PBS_TO_2_2_KS_PBS_GAUSSIAN_2M128,
     );
 
     let key_string = "0053A6F94C9FF24598EB".to_string();
@@ -112,19 +112,19 @@ pub fn trivium_shortint_gen(c: &mut Criterion) {
 
 pub fn trivium_shortint_trans(c: &mut Criterion) {
     let config = ConfigBuilder::default()
-        .use_custom_parameters(V1_1_PARAM_MESSAGE_2_CARRY_2_KS_PBS_GAUSSIAN_2M128)
+        .use_custom_parameters(V1_2_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(V1_1_PARAM_MESSAGE_1_CARRY_1_KS_PBS_GAUSSIAN_2M128);
+        gen_keys(V1_2_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,
+        V1_2_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

@@ -1,9 +1,9 @@
 use crate::{KreyviumStream, KreyviumStreamByte, KreyviumStreamShortint, TransCiphering};
 use tfhe::prelude::*;
-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::parameters::v1_2::{
+    V1_2_PARAM_KEYSWITCH_1_1_KS_PBS_TO_2_2_KS_PBS_GAUSSIAN_2M128,
+    V1_2_PARAM_MESSAGE_1_CARRY_1_KS_PBS_GAUSSIAN_2M128,
+    V1_2_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,
@@ -66,7 +66,7 @@ fn get_hexagonal_string_from_bytes(a: Vec<u8>) -> String {
     assert!(a.len() % 8 == 0);
     let mut hexadecimal: String = "".to_string();
     for test in a {
-        hexadecimal.push_str(&format!("{:02X?}", test));
+        hexadecimal.push_str(&format!("{test:02X?}"));
     }
     hexadecimal
 }
@@ -74,7 +74,7 @@ fn get_hexagonal_string_from_bytes(a: Vec<u8>) -> String {
 fn get_hexagonal_string_from_u64(a: Vec<u64>) -> String {
     let mut hexadecimal: String = "".to_string();
     for test in a {
-        hexadecimal.push_str(&format!("{:016X?}", test));
+        hexadecimal.push_str(&format!("{test:016X?}"));
     }
     hexadecimal
 }
@@ -221,19 +221,19 @@ use tfhe::shortint::prelude::*;
 #[test]
 fn kreyvium_test_shortint_long() {
     let config = ConfigBuilder::default()
-        .use_custom_parameters(V1_1_PARAM_MESSAGE_2_CARRY_2_KS_PBS_GAUSSIAN_2M128)
+        .use_custom_parameters(V1_2_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(V1_1_PARAM_MESSAGE_1_CARRY_1_KS_PBS_GAUSSIAN_2M128);
+        gen_keys(V1_2_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,
+        V1_2_PARAM_KEYSWITCH_1_1_KS_PBS_TO_2_2_KS_PBS_GAUSSIAN_2M128,
     );
 
     let key_string = "0053A6F94C9FF24598EB000000000000".to_string();

apps/trivium/src/static_deque/static_deque.rs

@@ -55,7 +55,7 @@ impl<const N: usize, T> Index<usize> for StaticDeque<N, T> {
     /// 0 is youngest
     fn index(&self, i: usize) -> &T {
         if i >= N {
-            panic!("Index {:?} too high for size {:?}", i, N);
+            panic!("Index {i:?} too high for size {N:?}");
         }
         &self.arr[(N + self.cursor - i - 1) % N]
     }
@@ -66,7 +66,7 @@ impl<const N: usize, T> IndexMut<usize> for StaticDeque<N, T> {
     /// 0 is youngest
     fn index_mut(&mut self, i: usize) -> &mut T {
         if i >= N {
-            panic!("Index {:?} too high for size {:?}", i, N);
+            panic!("Index {i:?} too high for size {N:?}");
         }
         &mut self.arr[(N + self.cursor - i - 1) % N]
     }

apps/trivium/src/trivium/test.rs

@@ -1,9 +1,9 @@
 use crate::{TransCiphering, TriviumStream, TriviumStreamByte, TriviumStreamShortint};
 use tfhe::prelude::*;
-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::parameters::v1_2::{
+    V1_2_PARAM_KEYSWITCH_1_1_KS_PBS_TO_2_2_KS_PBS_GAUSSIAN_2M128,
+    V1_2_PARAM_MESSAGE_1_CARRY_1_KS_PBS_GAUSSIAN_2M128,
+    V1_2_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,
@@ -66,7 +66,7 @@ fn get_hexagonal_string_from_bytes(a: Vec<u8>) -> String {
     assert!(a.len() % 8 == 0);
     let mut hexadecimal: String = "".to_string();
     for test in a {
-        hexadecimal.push_str(&format!("{:02X?}", test));
+        hexadecimal.push_str(&format!("{test:02X?}"));
     }
     hexadecimal
 }
@@ -74,7 +74,7 @@ fn get_hexagonal_string_from_bytes(a: Vec<u8>) -> String {
 fn get_hexagonal_string_from_u64(a: Vec<u64>) -> String {
     let mut hexadecimal: String = "".to_string();
     for test in a {
-        hexadecimal.push_str(&format!("{:016X?}", test));
+        hexadecimal.push_str(&format!("{test:016X?}"));
     }
     hexadecimal
 }
@@ -357,19 +357,19 @@ use tfhe::shortint::prelude::*;
 #[test]
 fn trivium_test_shortint_long() {
     let config = ConfigBuilder::default()
-        .use_custom_parameters(V1_1_PARAM_MESSAGE_2_CARRY_2_KS_PBS_GAUSSIAN_2M128)
+        .use_custom_parameters(V1_2_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(V1_1_PARAM_MESSAGE_1_CARRY_1_KS_PBS_GAUSSIAN_2M128);
+        gen_keys(V1_2_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,
+        V1_2_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 = "0.9.0"
+version = "0.10.0"
 edition = "2021"
 authors = ["Zama team"]
 license = "BSD-3-Clause-Clear"
@@ -18,3 +18,4 @@ bindgen = "0.71"
 
 [features]
 experimental-multi-arch = []
+profile = []

backends/tfhe-cuda-backend/build.rs

@@ -45,6 +45,13 @@ fn main() {
         } else {
             cmake_config.define("MULTI_ARCH", "OFF");
         }
+        // Conditionally pass the "USE_NVTOOLS" variable to CMake if the feature is enabled
+        if cfg!(feature = "profile") {
+            cmake_config.define("USE_NVTOOLS", "ON");
+            println!("cargo:rustc-link-lib=nvToolsExt");
+        } else {
+            cmake_config.define("USE_NVTOOLS", "OFF");
+        }
 
         // Build the CMake project
         let dest = cmake_config.build();
@@ -71,7 +78,7 @@ fn main() {
             "cuda/include/integer/compression/compression.h",
             "cuda/include/integer/integer.h",
             "cuda/include/zk/zk.h",
-            "cuda/include/keyswitch.h",
+            "cuda/include/keyswitch/keyswitch.h",
             "cuda/include/keyswitch/ks_enums.h",
             "cuda/include/linear_algebra.h",
             "cuda/include/fft/fft128.h",
@@ -86,7 +93,7 @@ fn main() {
         };
         let mut headers_modified = bindings_modified;
         for header in headers {
-            println!("cargo:rerun-if-changed={}", header);
+            println!("cargo:rerun-if-changed={header}");
             // Check modification times
             let header_modified = std::fs::metadata(header).unwrap().modified().unwrap();
             if header_modified > headers_modified {

backends/tfhe-cuda-backend/cuda/CMakeLists.txt

@@ -88,11 +88,18 @@ else()
   set(OPTIMIZATION_FLAGS "${OPTIMIZATION_FLAGS} -O3")
 endif()
 
-# in production, should use -arch=sm_70 --ptxas-options=-v to see register spills -lineinfo for better debugging
+# Check if the USE_NVTOOLS environment variable is set
+if(${USE_NVTOOLS})
+  message(STATUS "USE_NVTOOLS is enabled")
+  add_definitions(-DUSE_NVTOOLS)
+endif()
+
+# in production, should use -arch=sm_70 --ptxas-options=-v to see register spills -lineinfo for better debugging to use
+# nvtx when profiling -lnvToolsExt
 set(CMAKE_CUDA_FLAGS
     "${CMAKE_CUDA_FLAGS} -ccbin ${CMAKE_CXX_COMPILER} ${OPTIMIZATION_FLAGS}\
   -std=c++17 --no-exceptions  --expt-relaxed-constexpr -rdc=true \
-  --use_fast_math -Xcompiler -fPIC")
+  --use_fast_math -Xcompiler -fPIC ")
 
 set(INCLUDE_DIR include)
 

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

@@ -28,9 +28,15 @@ void cuda_modulus_switch_inplace_64(void *stream, uint32_t gpu_index,
 
 void cuda_improve_noise_modulus_switch_64(
     void *stream, uint32_t gpu_index, void *lwe_array_out,
-    void const *lwe_array_in, void const *encrypted_zeros, uint32_t lwe_size,
-    uint32_t num_lwes, uint32_t num_zeros, double input_variance,
-    double r_sigma, double bound, uint32_t log_modulus);
+    void const *lwe_array_in, void const *lwe_array_indexes,
+    void const *encrypted_zeros, uint32_t lwe_size, uint32_t num_lwes,
+    uint32_t num_zeros, double input_variance, double r_sigma, double bound,
+    uint32_t log_modulus);
+
+void cuda_glwe_sample_extract_128(
+    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 lwe_per_glwe, uint32_t glwe_dimension, uint32_t polynomial_size);
 }
 
 #endif

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

@@ -8,7 +8,6 @@
 #include <cuda_runtime.h>
 #include <vector>
 
-#define synchronize_threads_in_block() __syncthreads()
 extern "C" {
 
 #define check_cuda_error(ans)                                                  \
@@ -48,13 +47,28 @@ uint32_t cuda_is_available();
 
 void *cuda_malloc(uint64_t size, uint32_t gpu_index);
 
+void *cuda_malloc_with_size_tracking_async(uint64_t size, cudaStream_t stream,
+                                           uint32_t gpu_index,
+                                           uint64_t *size_tracker,
+                                           bool allocate_gpu_memory);
+
 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);
+bool cuda_check_valid_malloc(uint64_t size, uint32_t gpu_index);
+
+void cuda_memcpy_with_size_tracking_async_to_gpu(void *dest, const void *src,
+                                                 uint64_t size,
+                                                 cudaStream_t stream,
+                                                 uint32_t gpu_index,
+                                                 bool gpu_memory_allocated);
 
 void cuda_memcpy_async_to_gpu(void *dest, const void *src, uint64_t size,
                               cudaStream_t stream, uint32_t gpu_index);
 
+void cuda_memcpy_with_size_tracking_async_gpu_to_gpu(
+    void *dest, void const *src, uint64_t size, cudaStream_t stream,
+    uint32_t gpu_index, bool gpu_memory_allocated);
+
 void cuda_memcpy_async_gpu_to_gpu(void *dest, void const *src, uint64_t size,
                                   cudaStream_t stream, uint32_t gpu_index);
 
@@ -64,6 +78,11 @@ void cuda_memcpy_gpu_to_gpu(void *dest, void const *src, uint64_t size,
 void cuda_memcpy_async_to_cpu(void *dest, const void *src, uint64_t size,
                               cudaStream_t stream, uint32_t gpu_index);
 
+void cuda_memset_with_size_tracking_async(void *dest, uint64_t val,
+                                          uint64_t size, cudaStream_t stream,
+                                          uint32_t gpu_index,
+                                          bool gpu_memory_allocated);
+
 void cuda_memset_async(void *dest, uint64_t val, uint64_t size,
                        cudaStream_t stream, uint32_t gpu_index);
 
@@ -73,6 +92,10 @@ void cuda_synchronize_device(uint32_t gpu_index);
 
 void cuda_drop(void *ptr, uint32_t gpu_index);
 
+void cuda_drop_with_size_tracking_async(void *ptr, cudaStream_t stream,
+                                        uint32_t gpu_index,
+                                        bool gpu_memory_allocated);
+
 void cuda_drop_async(void *ptr, cudaStream_t stream, uint32_t gpu_index);
 }
 

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

@@ -8,7 +8,7 @@ extern std::mutex m;
 extern bool p2p_enabled;
 
 extern "C" {
-int32_t cuda_setup_multi_gpu();
+int32_t cuda_setup_multi_gpu(int device_0_id);
 }
 
 // Define a variant type that can be either a vector or a single pointer

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

@@ -4,7 +4,7 @@
 #include "../../pbs/pbs_enums.h"
 
 extern "C" {
-void scratch_cuda_integer_compress_radix_ciphertext_64(
+uint64_t scratch_cuda_integer_compress_radix_ciphertext_64(
     void *const *streams, uint32_t const *gpu_indexes, uint32_t gpu_count,
     int8_t **mem_ptr, uint32_t compression_glwe_dimension,
     uint32_t compression_polynomial_size, uint32_t lwe_dimension,
@@ -13,7 +13,7 @@ void scratch_cuda_integer_compress_radix_ciphertext_64(
     uint32_t lwe_per_glwe, uint32_t storage_log_modulus,
     bool allocate_gpu_memory);
 
-void scratch_cuda_integer_decompress_radix_ciphertext_64(
+uint64_t scratch_cuda_integer_decompress_radix_ciphertext_64(
     void *const *streams, uint32_t const *gpu_indexes, uint32_t gpu_count,
     int8_t **mem_ptr, uint32_t encryption_glwe_dimension,
     uint32_t encryption_polynomial_size, uint32_t compression_glwe_dimension,

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

@@ -14,41 +14,44 @@ template <typename Torus> struct int_compression {
   int8_t *fp_ks_buffer;
   Torus *tmp_lwe;
   Torus *tmp_glwe_array_out;
+  bool gpu_memory_allocated;
 
   int_compression(cudaStream_t const *streams, uint32_t const *gpu_indexes,
                   uint32_t gpu_count, int_radix_params compression_params,
                   uint32_t num_radix_blocks, uint32_t lwe_per_glwe,
-                  uint32_t storage_log_modulus, bool allocate_gpu_memory) {
+                  uint32_t storage_log_modulus, bool allocate_gpu_memory,
+                  uint64_t *size_tracker) {
+    gpu_memory_allocated = allocate_gpu_memory;
     this->compression_params = compression_params;
     this->lwe_per_glwe = lwe_per_glwe;
     this->storage_log_modulus = storage_log_modulus;
     this->body_count = num_radix_blocks;
 
-    if (allocate_gpu_memory) {
-      Torus glwe_accumulator_size = (compression_params.glwe_dimension + 1) *
-                                    compression_params.polynomial_size;
+    Torus glwe_accumulator_size = (compression_params.glwe_dimension + 1) *
+                                  compression_params.polynomial_size;
 
-      tmp_lwe = (Torus *)cuda_malloc_async(
-          num_radix_blocks * (compression_params.small_lwe_dimension + 1) *
-              sizeof(Torus),
-          streams[0], gpu_indexes[0]);
-      tmp_glwe_array_out = (Torus *)cuda_malloc_async(
-          lwe_per_glwe * glwe_accumulator_size * sizeof(Torus), streams[0],
-          gpu_indexes[0]);
+    tmp_lwe = (Torus *)cuda_malloc_with_size_tracking_async(
+        num_radix_blocks * (compression_params.small_lwe_dimension + 1) *
+            sizeof(Torus),
+        streams[0], gpu_indexes[0], size_tracker, allocate_gpu_memory);
+    tmp_glwe_array_out = (Torus *)cuda_malloc_with_size_tracking_async(
+        lwe_per_glwe * glwe_accumulator_size * sizeof(Torus), streams[0],
+        gpu_indexes[0], size_tracker, allocate_gpu_memory);
 
-      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);
-    }
+    *size_tracker += 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, allocate_gpu_memory);
   }
   void release(cudaStream_t const *streams, uint32_t const *gpu_indexes,
                uint32_t gpu_count) {
-    cuda_drop_async(tmp_lwe, streams[0], gpu_indexes[0]);
-    cuda_drop_async(tmp_glwe_array_out, streams[0], gpu_indexes[0]);
-    cleanup_packing_keyswitch_lwe_list_to_glwe(streams[0], gpu_indexes[0],
-                                               &fp_ks_buffer);
+    cuda_drop_with_size_tracking_async(tmp_lwe, streams[0], gpu_indexes[0],
+                                       gpu_memory_allocated);
+    cuda_drop_with_size_tracking_async(tmp_glwe_array_out, streams[0],
+                                       gpu_indexes[0], gpu_memory_allocated);
+    cleanup_packing_keyswitch_lwe_list_to_glwe(
+        streams[0], gpu_indexes[0], &fp_ks_buffer, gpu_memory_allocated);
   }
 };
 
@@ -66,66 +69,71 @@ template <typename Torus> struct int_decompression {
   uint32_t *tmp_indexes_array;
 
   int_radix_lut<Torus> *decompression_rescale_lut;
+  bool gpu_memory_allocated;
 
   int_decompression(cudaStream_t const *streams, uint32_t const *gpu_indexes,
                     uint32_t gpu_count, int_radix_params encryption_params,
                     int_radix_params compression_params,
                     uint32_t num_radix_blocks, uint32_t body_count,
-                    uint32_t storage_log_modulus, bool allocate_gpu_memory) {
+                    uint32_t storage_log_modulus, bool allocate_gpu_memory,
+                    uint64_t *size_tracker) {
+    gpu_memory_allocated = allocate_gpu_memory;
     this->encryption_params = encryption_params;
     this->compression_params = compression_params;
     this->storage_log_modulus = storage_log_modulus;
     this->num_radix_blocks = num_radix_blocks;
     this->body_count = body_count;
 
-    if (allocate_gpu_memory) {
-      Torus glwe_accumulator_size = (compression_params.glwe_dimension + 1) *
-                                    compression_params.polynomial_size;
-      Torus lwe_accumulator_size = (compression_params.glwe_dimension *
-                                        compression_params.polynomial_size +
-                                    1);
-      decompression_rescale_lut = new int_radix_lut<Torus>(
-          streams, gpu_indexes, gpu_count, encryption_params, 1,
-          num_radix_blocks, allocate_gpu_memory);
+    Torus glwe_accumulator_size = (compression_params.glwe_dimension + 1) *
+                                  compression_params.polynomial_size;
+    Torus lwe_accumulator_size = (compression_params.glwe_dimension *
+                                      compression_params.polynomial_size +
+                                  1);
+    decompression_rescale_lut = new int_radix_lut<Torus>(
+        streams, gpu_indexes, gpu_count, encryption_params, 1, num_radix_blocks,
+        allocate_gpu_memory, size_tracker);
 
-      tmp_extracted_glwe = (Torus *)cuda_malloc_async(
-          num_radix_blocks * glwe_accumulator_size * sizeof(Torus), streams[0],
-          gpu_indexes[0]);
-      tmp_indexes_array = (uint32_t *)cuda_malloc_async(
-          num_radix_blocks * sizeof(uint32_t), streams[0], gpu_indexes[0]);
-      tmp_extracted_lwe = (Torus *)cuda_malloc_async(
-          num_radix_blocks * lwe_accumulator_size * sizeof(Torus), streams[0],
-          gpu_indexes[0]);
+    tmp_extracted_glwe = (Torus *)cuda_malloc_with_size_tracking_async(
+        num_radix_blocks * glwe_accumulator_size * sizeof(Torus), streams[0],
+        gpu_indexes[0], size_tracker, allocate_gpu_memory);
+    tmp_indexes_array = (uint32_t *)cuda_malloc_with_size_tracking_async(
+        num_radix_blocks * sizeof(uint32_t), streams[0], gpu_indexes[0],
+        size_tracker, allocate_gpu_memory);
+    tmp_extracted_lwe = (Torus *)cuda_malloc_with_size_tracking_async(
+        num_radix_blocks * lwe_accumulator_size * sizeof(Torus), streams[0],
+        gpu_indexes[0], size_tracker, allocate_gpu_memory);
 
-      // 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; };
+    // 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; };
 
-      auto effective_compression_message_modulus =
-          encryption_params.carry_modulus;
-      auto effective_compression_carry_modulus = 1;
+    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,
-          decompression_rescale_f);
+    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, decompression_rescale_f,
+        gpu_memory_allocated);
 
-      decompression_rescale_lut->broadcast_lut(streams, gpu_indexes, 0);
-    }
+    decompression_rescale_lut->broadcast_lut(streams, gpu_indexes, 0);
   }
   void release(cudaStream_t const *streams, uint32_t const *gpu_indexes,
                uint32_t gpu_count) {
-    cuda_drop_async(tmp_extracted_glwe, streams[0], gpu_indexes[0]);
-    cuda_drop_async(tmp_extracted_lwe, streams[0], gpu_indexes[0]);
-    cuda_drop_async(tmp_indexes_array, streams[0], gpu_indexes[0]);
+    cuda_drop_with_size_tracking_async(tmp_extracted_glwe, streams[0],
+                                       gpu_indexes[0], gpu_memory_allocated);
+    cuda_drop_with_size_tracking_async(tmp_extracted_lwe, streams[0],
+                                       gpu_indexes[0], gpu_memory_allocated);
+    cuda_drop_with_size_tracking_async(tmp_indexes_array, streams[0],
+                                       gpu_indexes[0], gpu_memory_allocated);
 
     decompression_rescale_lut->release(streams, gpu_indexes, gpu_count);
     delete decompression_rescale_lut;

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

@@ -48,7 +48,7 @@ typedef struct {
   uint32_t lwe_dimension;
 } CudaRadixCiphertextFFI;
 
-void scratch_cuda_apply_univariate_lut_kb_64(
+uint64_t 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,
     uint32_t glwe_dimension, uint32_t polynomial_size, uint32_t ks_level,
@@ -56,7 +56,7 @@ void scratch_cuda_apply_univariate_lut_kb_64(
     uint32_t grouping_factor, uint32_t input_lwe_ciphertext_count,
     uint32_t message_modulus, uint32_t carry_modulus, PBS_TYPE pbs_type,
     uint64_t lut_degree, bool allocate_gpu_memory, bool allocate_ms_array);
-void scratch_cuda_apply_many_univariate_lut_kb_64(
+uint64_t 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,
@@ -78,7 +78,7 @@ void cleanup_cuda_apply_univariate_lut_kb_64(void *const *streams,
                                              uint32_t gpu_count,
                                              int8_t **mem_ptr_void);
 
-void scratch_cuda_apply_bivariate_lut_kb_64(
+uint64_t scratch_cuda_apply_bivariate_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,
@@ -109,7 +109,7 @@ void cuda_apply_many_univariate_lut_kb_64(
     CudaModulusSwitchNoiseReductionKeyFFI const *ms_noise_reduction_key,
     void *const *bsks, uint32_t num_luts, uint32_t lut_stride);
 
-void scratch_cuda_full_propagation_64(
+uint64_t scratch_cuda_full_propagation_64(
     void *const *streams, uint32_t const *gpu_indexes, uint32_t gpu_count,
     int8_t **mem_ptr, uint32_t lwe_dimension, uint32_t glwe_dimension,
     uint32_t polynomial_size, uint32_t ks_level, uint32_t ks_base_log,
@@ -127,7 +127,7 @@ void cleanup_cuda_full_propagation(void *const *streams,
                                    uint32_t const *gpu_indexes,
                                    uint32_t gpu_count, int8_t **mem_ptr_void);
 
-void scratch_cuda_integer_mult_radix_ciphertext_kb_64(
+uint64_t scratch_cuda_integer_mult_radix_ciphertext_kb_64(
     void *const *streams, uint32_t const *gpu_indexes, uint32_t gpu_count,
     int8_t **mem_ptr, bool const is_boolean_left, bool const is_boolean_right,
     uint32_t message_modulus, uint32_t carry_modulus, uint32_t glwe_dimension,
@@ -161,7 +161,7 @@ void cuda_scalar_addition_integer_radix_ciphertext_64_inplace(
     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(
+uint64_t scratch_cuda_integer_radix_logical_scalar_shift_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,
@@ -176,7 +176,7 @@ void cuda_integer_radix_logical_scalar_shift_kb_64_inplace(
     void *const *bsks, void *const *ksks,
     CudaModulusSwitchNoiseReductionKeyFFI const *ms_noise_reduction_key);
 
-void scratch_cuda_integer_radix_arithmetic_scalar_shift_kb_64(
+uint64_t scratch_cuda_integer_radix_arithmetic_scalar_shift_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,
@@ -199,7 +199,7 @@ void cleanup_cuda_integer_radix_arithmetic_scalar_shift(
     void *const *streams, uint32_t const *gpu_indexes, uint32_t gpu_count,
     int8_t **mem_ptr_void);
 
-void scratch_cuda_integer_radix_shift_and_rotate_kb_64(
+uint64_t scratch_cuda_integer_radix_shift_and_rotate_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,
@@ -219,7 +219,7 @@ void cleanup_cuda_integer_radix_shift_and_rotate(void *const *streams,
                                                  uint32_t gpu_count,
                                                  int8_t **mem_ptr_void);
 
-void scratch_cuda_integer_radix_comparison_kb_64(
+uint64_t scratch_cuda_integer_radix_comparison_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,
@@ -250,7 +250,7 @@ void cleanup_cuda_integer_comparison(void *const *streams,
                                      uint32_t const *gpu_indexes,
                                      uint32_t gpu_count, int8_t **mem_ptr_void);
 
-void scratch_cuda_integer_radix_bitop_kb_64(
+uint64_t scratch_cuda_integer_radix_bitop_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,
@@ -279,7 +279,7 @@ void cleanup_cuda_integer_bitop(void *const *streams,
                                 uint32_t const *gpu_indexes, uint32_t gpu_count,
                                 int8_t **mem_ptr_void);
 
-void scratch_cuda_integer_radix_cmux_kb_64(
+uint64_t scratch_cuda_integer_radix_cmux_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,
@@ -301,7 +301,7 @@ void cleanup_cuda_integer_radix_cmux(void *const *streams,
                                      uint32_t const *gpu_indexes,
                                      uint32_t gpu_count, int8_t **mem_ptr_void);
 
-void scratch_cuda_integer_radix_scalar_rotate_kb_64(
+uint64_t scratch_cuda_integer_radix_scalar_rotate_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,
@@ -321,7 +321,7 @@ void cleanup_cuda_integer_radix_scalar_rotate(void *const *streams,
                                               uint32_t gpu_count,
                                               int8_t **mem_ptr_void);
 
-void scratch_cuda_propagate_single_carry_kb_64_inplace(
+uint64_t scratch_cuda_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,
@@ -330,7 +330,7 @@ void scratch_cuda_propagate_single_carry_kb_64_inplace(
     uint32_t carry_modulus, PBS_TYPE pbs_type, uint32_t requested_flag,
     uint32_t uses_carry, bool allocate_gpu_memory, bool allocate_ms_array);
 
-void scratch_cuda_add_and_propagate_single_carry_kb_64_inplace(
+uint64_t 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,
@@ -365,7 +365,7 @@ void cleanup_cuda_add_and_propagate_single_carry(void *const *streams,
                                                  uint32_t gpu_count,
                                                  int8_t **mem_ptr_void);
 
-void scratch_cuda_integer_overflowing_sub_kb_64_inplace(
+uint64_t 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,
@@ -388,7 +388,7 @@ void cleanup_cuda_integer_overflowing_sub(void *const *streams,
                                           uint32_t gpu_count,
                                           int8_t **mem_ptr_void);
 
-void scratch_cuda_integer_radix_partial_sum_ciphertexts_vec_kb_64(
+uint64_t 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,
     uint32_t lwe_dimension, uint32_t ks_level, uint32_t ks_base_log,
@@ -408,7 +408,7 @@ 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_scalar_mul_kb_64(
+uint64_t 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,
     uint32_t lwe_dimension, uint32_t ks_level, uint32_t ks_base_log,
@@ -429,7 +429,7 @@ void cleanup_cuda_integer_radix_scalar_mul(void *const *streams,
                                            uint32_t gpu_count,
                                            int8_t **mem_ptr_void);
 
-void scratch_cuda_integer_div_rem_radix_ciphertext_kb_64(
+uint64_t scratch_cuda_integer_div_rem_radix_ciphertext_kb_64(
     void *const *streams, uint32_t const *gpu_indexes, uint32_t gpu_count,
     bool is_signed, int8_t **mem_ptr, uint32_t glwe_dimension,
     uint32_t polynomial_size, uint32_t big_lwe_dimension,
@@ -450,7 +450,7 @@ 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_integer_compute_prefix_sum_hillis_steele_64(
+uint64_t 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,
     uint32_t glwe_dimension, uint32_t polynomial_size, uint32_t ks_level,
@@ -476,7 +476,7 @@ void cuda_integer_reverse_blocks_64_inplace(void *const *streams,
                                             uint32_t gpu_count,
                                             CudaRadixCiphertextFFI *lwe_array);
 
-void scratch_cuda_integer_abs_inplace_radix_ciphertext_kb_64(
+uint64_t scratch_cuda_integer_abs_inplace_radix_ciphertext_kb_64(
     void *const *streams, uint32_t const *gpu_indexes, uint32_t gpu_count,
     int8_t **mem_ptr, bool is_signed, uint32_t glwe_dimension,
     uint32_t polynomial_size, uint32_t big_lwe_dimension,
@@ -496,7 +496,7 @@ void cleanup_cuda_integer_abs_inplace(void *const *streams,
                                       uint32_t gpu_count,
                                       int8_t **mem_ptr_void);
 
-void scratch_cuda_integer_are_all_comparisons_block_true_kb_64(
+uint64_t 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,
@@ -517,7 +517,7 @@ 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(
+uint64_t 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,

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

@@ -3,7 +3,8 @@
 
 void release_radix_ciphertext_async(cudaStream_t const stream,
                                     uint32_t const gpu_index,
-                                    CudaRadixCiphertextFFI *data);
+                                    CudaRadixCiphertextFFI *data,
+                                    const bool gpu_memory_allocated);
 
 void reset_radix_ciphertext_blocks(CudaRadixCiphertextFFI *data,
                                    uint32_t new_num_blocks);

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

@@ -19,7 +19,7 @@ void cuda_keyswitch_lwe_ciphertext_vector_64(
     uint32_t lwe_dimension_out, uint32_t base_log, uint32_t level_count,
     uint32_t num_samples);
 
-void scratch_packing_keyswitch_lwe_list_to_glwe_64(
+uint64_t scratch_packing_keyswitch_lwe_list_to_glwe_64(
     void *stream, uint32_t gpu_index, int8_t **fp_ks_buffer,
     uint32_t lwe_dimension, uint32_t glwe_dimension, uint32_t polynomial_size,
     uint32_t num_lwes, bool allocate_gpu_memory);
@@ -31,9 +31,22 @@ 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);
 
+void scratch_packing_keyswitch_lwe_list_to_glwe_128(
+    void *stream, uint32_t gpu_index, int8_t **fp_ks_buffer,
+    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_128(
+    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,
+    uint32_t input_lwe_dimension, uint32_t output_glwe_dimension,
+    uint32_t output_polynomial_size, uint32_t base_log, uint32_t level_count,
+    uint32_t num_lwes);
+
 void cleanup_packing_keyswitch_lwe_list_to_glwe(void *stream,
                                                 uint32_t gpu_index,
-                                                int8_t **fp_ks_buffer);
+                                                int8_t **fp_ks_buffer,
+                                                bool gpu_memory_allocated);
 }
 
 #endif // CNCRT_KS_H_

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

@@ -42,6 +42,24 @@ void cuda_mult_lwe_ciphertext_vector_cleartext_vector_64(
     void const *lwe_array_in, void const *cleartext_array_in,
     const uint32_t input_lwe_dimension,
     const uint32_t input_lwe_ciphertext_count);
+
+void scratch_wrapping_polynomial_mul_one_to_many_64(void *stream,
+                                                    uint32_t gpu_index,
+                                                    uint32_t polynomial_size,
+                                                    int8_t **circulant_buf);
+
+void cleanup_wrapping_polynomial_mul_one_to_many_64(void *stream,
+                                                    uint32_t gpu_index,
+                                                    int8_t *circulant_buf);
+
+void cuda_wrapping_polynomial_mul_one_to_many_64(
+    void *stream, uint32_t gpu_index, void *result, void const *poly_lhs,
+    int8_t *circulant, void const *poly_rhs, uint32_t polynomial_size,
+    uint32_t n_rhs);
+void cuda_glwe_wrapping_polynomial_mul_one_to_many_64(
+    void *stream, uint32_t gpu_index, void *result, void const *poly_lhs,
+    int8_t *circulant, void const *poly_rhs, uint32_t polynomial_size,
+    uint32_t glwe_dimension, uint32_t n_rhs);
 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,

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

@@ -14,7 +14,7 @@ bool has_support_to_cuda_programmable_bootstrap_tbc_multi_bit(
 
 #if CUDA_ARCH >= 900
 template <typename Torus>
-void scratch_cuda_tbc_multi_bit_programmable_bootstrap(
+uint64_t scratch_cuda_tbc_multi_bit_programmable_bootstrap(
     void *stream, uint32_t gpu_index, pbs_buffer<Torus, MULTI_BIT> **buffer,
     uint32_t glwe_dimension, uint32_t polynomial_size, uint32_t level_count,
     uint32_t input_lwe_ciphertext_count, bool allocate_gpu_memory);
@@ -32,7 +32,7 @@ void cuda_tbc_multi_bit_programmable_bootstrap_lwe_ciphertext_vector(
 #endif
 
 template <typename Torus>
-void scratch_cuda_cg_multi_bit_programmable_bootstrap(
+uint64_t scratch_cuda_cg_multi_bit_programmable_bootstrap(
     void *stream, uint32_t gpu_index, pbs_buffer<Torus, MULTI_BIT> **pbs_buffer,
     uint32_t glwe_dimension, uint32_t polynomial_size, uint32_t level_count,
     uint32_t input_lwe_ciphertext_count, bool allocate_gpu_memory);
@@ -49,7 +49,7 @@ void cuda_cg_multi_bit_programmable_bootstrap_lwe_ciphertext_vector(
     uint32_t num_many_lut, uint32_t lut_stride);
 
 template <typename Torus>
-void scratch_cuda_multi_bit_programmable_bootstrap(
+uint64_t scratch_cuda_multi_bit_programmable_bootstrap(
     void *stream, uint32_t gpu_index, pbs_buffer<Torus, MULTI_BIT> **pbs_buffer,
     uint32_t glwe_dimension, uint32_t polynomial_size, uint32_t level_count,
     uint32_t input_lwe_ciphertext_count, bool allocate_gpu_memory);
@@ -109,11 +109,14 @@ template <typename Torus> struct pbs_buffer<Torus, PBS_TYPE::MULTI_BIT> {
   double2 *global_join_buffer;
 
   PBS_VARIANT pbs_variant;
+  bool gpu_memory_allocated;
 
   pbs_buffer(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, uint32_t lwe_chunk_size,
-             PBS_VARIANT pbs_variant, bool allocate_gpu_memory) {
+             PBS_VARIANT pbs_variant, bool allocate_gpu_memory,
+             uint64_t *size_tracker) {
+    gpu_memory_allocated = allocate_gpu_memory;
     cuda_set_device(gpu_index);
 
     this->pbs_variant = pbs_variant;
@@ -164,107 +167,117 @@ template <typename Torus> struct pbs_buffer<Torus, PBS_TYPE::MULTI_BIT> {
     auto num_blocks_acc_tbc = num_blocks_acc_cg;
 #endif
 
-    if (allocate_gpu_memory) {
-      // Keybundle
-      if (max_shared_memory < full_sm_keybundle)
-        d_mem_keybundle = (int8_t *)cuda_malloc_async(
-            num_blocks_keybundle * full_sm_keybundle, stream, gpu_index);
+    // Keybundle
+    if (max_shared_memory < full_sm_keybundle)
+      d_mem_keybundle = (int8_t *)cuda_malloc_with_size_tracking_async(
+          num_blocks_keybundle * full_sm_keybundle, stream, gpu_index,
+          size_tracker, allocate_gpu_memory);
 
-      switch (pbs_variant) {
-      case PBS_VARIANT::CG:
-        // Accumulator CG
-        if (max_shared_memory < partial_sm_cg_accumulate)
-          d_mem_acc_cg = (int8_t *)cuda_malloc_async(
-              num_blocks_acc_cg * full_sm_cg_accumulate, stream, gpu_index);
-        else if (max_shared_memory < full_sm_cg_accumulate)
-          d_mem_acc_cg = (int8_t *)cuda_malloc_async(
-              num_blocks_acc_cg * partial_sm_cg_accumulate, stream, gpu_index);
-        break;
-      case PBS_VARIANT::DEFAULT:
-        // Accumulator step one
-        if (max_shared_memory < partial_sm_accumulate_step_one)
-          d_mem_acc_step_one = (int8_t *)cuda_malloc_async(
-              num_blocks_acc_step_one * full_sm_accumulate_step_one, stream,
-              gpu_index);
-        else if (max_shared_memory < full_sm_accumulate_step_one)
-          d_mem_acc_step_one = (int8_t *)cuda_malloc_async(
-              num_blocks_acc_step_one * partial_sm_accumulate_step_one, stream,
-              gpu_index);
-
-        // Accumulator step two
-        if (max_shared_memory < full_sm_accumulate_step_two)
-          d_mem_acc_step_two = (int8_t *)cuda_malloc_async(
-              num_blocks_acc_step_two * full_sm_accumulate_step_two, stream,
-              gpu_index);
-        break;
-#if CUDA_ARCH >= 900
-      case TBC:
-        // 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_tbc_accumulate bytes, TBC PBS will run on NOSM. If we have
-        // partial_sm_tbc_accumulate but not full_sm_tbc_accumulate bytes, it
-        // will run on PARTIALSM. Otherwise, FULLSM.
-        //
-        // NOSM mode actually requires minimum_sm_tbc shared memory bytes.
-
-        // Accumulator TBC
-        if (max_shared_memory < partial_sm_tbc_accumulate + minimum_sm_tbc)
-          d_mem_acc_tbc = (int8_t *)cuda_malloc_async(
-              num_blocks_acc_tbc * full_sm_tbc_accumulate, stream, gpu_index);
-        else if (max_shared_memory < full_sm_tbc_accumulate + minimum_sm_tbc)
-          d_mem_acc_tbc = (int8_t *)cuda_malloc_async(
-              num_blocks_acc_tbc * partial_sm_tbc_accumulate, stream,
-              gpu_index);
-        break;
-#endif
-      default:
-        PANIC("Cuda error (PBS): unsupported implementation variant.")
-      }
-
-      keybundle_fft = (double2 *)cuda_malloc_async(
-          num_blocks_keybundle * (polynomial_size / 2) * sizeof(double2),
-          stream, gpu_index);
-      global_accumulator = (Torus *)cuda_malloc_async(
-          input_lwe_ciphertext_count * (glwe_dimension + 1) * polynomial_size *
-              sizeof(Torus),
-          stream, gpu_index);
-      global_join_buffer = (double2 *)cuda_malloc_async(
-          level_count * (glwe_dimension + 1) * input_lwe_ciphertext_count *
-              (polynomial_size / 2) * sizeof(double2),
-          stream, gpu_index);
-    }
-  }
-
-  void release(cudaStream_t stream, uint32_t gpu_index) {
-
-    if (d_mem_keybundle)
-      cuda_drop_async(d_mem_keybundle, stream, gpu_index);
     switch (pbs_variant) {
-    case DEFAULT:
-      if (d_mem_acc_step_one)
-        cuda_drop_async(d_mem_acc_step_one, stream, gpu_index);
-      if (d_mem_acc_step_two)
-        cuda_drop_async(d_mem_acc_step_two, stream, gpu_index);
+    case PBS_VARIANT::CG:
+      // Accumulator CG
+      if (max_shared_memory < partial_sm_cg_accumulate)
+        d_mem_acc_cg = (int8_t *)cuda_malloc_with_size_tracking_async(
+            num_blocks_acc_cg * full_sm_cg_accumulate, stream, gpu_index,
+            size_tracker, allocate_gpu_memory);
+      else if (max_shared_memory < full_sm_cg_accumulate)
+        d_mem_acc_cg = (int8_t *)cuda_malloc_with_size_tracking_async(
+            num_blocks_acc_cg * partial_sm_cg_accumulate, stream, gpu_index,
+            size_tracker, allocate_gpu_memory);
       break;
-    case CG:
-      if (d_mem_acc_cg)
-        cuda_drop_async(d_mem_acc_cg, stream, gpu_index);
+    case PBS_VARIANT::DEFAULT:
+      // Accumulator step one
+      if (max_shared_memory < partial_sm_accumulate_step_one)
+        d_mem_acc_step_one = (int8_t *)cuda_malloc_with_size_tracking_async(
+            num_blocks_acc_step_one * full_sm_accumulate_step_one, stream,
+            gpu_index, size_tracker, allocate_gpu_memory);
+      else if (max_shared_memory < full_sm_accumulate_step_one)
+        d_mem_acc_step_one = (int8_t *)cuda_malloc_with_size_tracking_async(
+            num_blocks_acc_step_one * partial_sm_accumulate_step_one, stream,
+            gpu_index, size_tracker, allocate_gpu_memory);
+
+      // Accumulator step two
+      if (max_shared_memory < full_sm_accumulate_step_two)
+        d_mem_acc_step_two = (int8_t *)cuda_malloc_with_size_tracking_async(
+            num_blocks_acc_step_two * full_sm_accumulate_step_two, stream,
+            gpu_index, size_tracker, allocate_gpu_memory);
       break;
 #if CUDA_ARCH >= 900
     case TBC:
-      if (d_mem_acc_tbc)
-        cuda_drop_async(d_mem_acc_tbc, stream, gpu_index);
+      // 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_tbc_accumulate bytes, TBC PBS will run on NOSM. If we have
+      // partial_sm_tbc_accumulate but not full_sm_tbc_accumulate bytes, it
+      // will run on PARTIALSM. Otherwise, FULLSM.
+      //
+      // NOSM mode actually requires minimum_sm_tbc shared memory bytes.
+
+      // Accumulator TBC
+      if (max_shared_memory < partial_sm_tbc_accumulate + minimum_sm_tbc)
+        d_mem_acc_tbc = (int8_t *)cuda_malloc_with_size_tracking_async(
+            num_blocks_acc_tbc * full_sm_tbc_accumulate, stream, gpu_index,
+            size_tracker, allocate_gpu_memory);
+      else if (max_shared_memory < full_sm_tbc_accumulate + minimum_sm_tbc)
+        d_mem_acc_tbc = (int8_t *)cuda_malloc_with_size_tracking_async(
+            num_blocks_acc_tbc * partial_sm_tbc_accumulate, stream, gpu_index,
+            size_tracker, allocate_gpu_memory);
       break;
 #endif
     default:
       PANIC("Cuda error (PBS): unsupported implementation variant.")
     }
 
-    cuda_drop_async(keybundle_fft, stream, gpu_index);
-    cuda_drop_async(global_accumulator, stream, gpu_index);
-    cuda_drop_async(global_join_buffer, stream, gpu_index);
+    keybundle_fft = (double2 *)cuda_malloc_with_size_tracking_async(
+        num_blocks_keybundle * (polynomial_size / 2) * sizeof(double2), stream,
+        gpu_index, size_tracker, allocate_gpu_memory);
+    global_accumulator = (Torus *)cuda_malloc_with_size_tracking_async(
+        input_lwe_ciphertext_count * (glwe_dimension + 1) * polynomial_size *
+            sizeof(Torus),
+        stream, gpu_index, size_tracker, allocate_gpu_memory);
+    global_join_buffer = (double2 *)cuda_malloc_with_size_tracking_async(
+        level_count * (glwe_dimension + 1) * input_lwe_ciphertext_count *
+            (polynomial_size / 2) * sizeof(double2),
+        stream, gpu_index, size_tracker, allocate_gpu_memory);
+  }
+
+  void release(cudaStream_t stream, uint32_t gpu_index) {
+
+    if (d_mem_keybundle)
+      cuda_drop_with_size_tracking_async(d_mem_keybundle, stream, gpu_index,
+                                         gpu_memory_allocated);
+    switch (pbs_variant) {
+    case DEFAULT:
+      if (d_mem_acc_step_one)
+        cuda_drop_with_size_tracking_async(d_mem_acc_step_one, stream,
+                                           gpu_index, gpu_memory_allocated);
+      if (d_mem_acc_step_two)
+        cuda_drop_with_size_tracking_async(d_mem_acc_step_two, stream,
+                                           gpu_index, gpu_memory_allocated);
+      break;
+    case CG:
+      if (d_mem_acc_cg)
+        cuda_drop_with_size_tracking_async(d_mem_acc_cg, stream, gpu_index,
+                                           gpu_memory_allocated);
+      break;
+#if CUDA_ARCH >= 900
+    case TBC:
+      if (d_mem_acc_tbc)
+        cuda_drop_with_size_tracking_async(d_mem_acc_tbc, stream, gpu_index,
+                                           gpu_memory_allocated);
+      break;
+#endif
+    default:
+      PANIC("Cuda error (PBS): unsupported implementation variant.")
+    }
+
+    cuda_drop_with_size_tracking_async(keybundle_fft, stream, gpu_index,
+                                       gpu_memory_allocated);
+    cuda_drop_with_size_tracking_async(global_accumulator, stream, gpu_index,
+                                       gpu_memory_allocated);
+    cuda_drop_with_size_tracking_async(global_join_buffer, stream, gpu_index,
+                                       gpu_memory_allocated);
   }
 };
 

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

@@ -84,155 +84,159 @@ template <typename Torus> struct pbs_buffer<Torus, PBS_TYPE::CLASSICAL> {
 
   PBS_VARIANT pbs_variant;
   bool uses_noise_reduction;
+  bool gpu_memory_allocated;
 
   pbs_buffer(cudaStream_t stream, uint32_t gpu_index, uint32_t lwe_dimension,
              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,
-             bool allocate_ms_array) {
+             bool allocate_ms_array, uint64_t *size_tracker) {
+    gpu_memory_allocated = allocate_gpu_memory;
     cuda_set_device(gpu_index);
     this->uses_noise_reduction = allocate_ms_array;
     this->pbs_variant = pbs_variant;
 
     auto max_shared_memory = cuda_get_max_shared_memory(gpu_index);
-    if (allocate_ms_array) {
-      this->temp_lwe_array_in = (Torus *)cuda_malloc_async(
-          (lwe_dimension + 1) * input_lwe_ciphertext_count * sizeof(Torus),
-          stream, gpu_index);
-    }
-    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<Torus>(
-                polynomial_size);
-        uint64_t full_sm_step_two =
-            get_buffer_size_full_sm_programmable_bootstrap_step_two<Torus>(
-                polynomial_size);
-        uint64_t partial_sm =
-            get_buffer_size_partial_sm_programmable_bootstrap<Torus>(
-                polynomial_size);
+    this->temp_lwe_array_in = (Torus *)cuda_malloc_with_size_tracking_async(
+        (lwe_dimension + 1) * input_lwe_ciphertext_count * sizeof(Torus),
+        stream, gpu_index, size_tracker, allocate_ms_array);
+    switch (pbs_variant) {
+    case PBS_VARIANT::DEFAULT: {
+      uint64_t full_sm_step_one =
+          get_buffer_size_full_sm_programmable_bootstrap_step_one<Torus>(
+              polynomial_size);
+      uint64_t full_sm_step_two =
+          get_buffer_size_full_sm_programmable_bootstrap_step_two<Torus>(
+              polynomial_size);
+      uint64_t partial_sm =
+          get_buffer_size_partial_sm_programmable_bootstrap<Torus>(
+              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 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 = (double2 *)cuda_malloc_async(
-            (glwe_dimension + 1) * level_count * input_lwe_ciphertext_count *
-                (polynomial_size / 2) * sizeof(double2),
-            stream, gpu_index);
-
-        global_accumulator = (Torus *)cuda_malloc_async(
-            (glwe_dimension + 1) * input_lwe_ciphertext_count *
-                polynomial_size * sizeof(Torus),
-            stream, gpu_index);
-      } break;
-      case PBS_VARIANT::CG: {
-        uint64_t full_sm =
-            get_buffer_size_full_sm_programmable_bootstrap_cg<Torus>(
-                polynomial_size);
-        uint64_t partial_sm =
-            get_buffer_size_partial_sm_programmable_bootstrap_cg<Torus>(
-                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 = (double2 *)cuda_malloc_async(
-            (glwe_dimension + 1) * level_count * input_lwe_ciphertext_count *
-                polynomial_size / 2 * sizeof(double2),
-            stream, gpu_index);
-      } break;
-#if CUDA_ARCH >= 900
-      case PBS_VARIANT::TBC: {
-
-        bool supports_dsm =
-            supports_distributed_shared_memory_on_classic_programmable_bootstrap<
-                Torus>(polynomial_size, max_shared_memory);
-
-        uint64_t full_sm =
-            get_buffer_size_full_sm_programmable_bootstrap_tbc<Torus>(
-                polynomial_size);
-        uint64_t partial_sm =
-            get_buffer_size_partial_sm_programmable_bootstrap_tbc<Torus>(
-                polynomial_size);
-        uint64_t minimum_sm_tbc = 0;
-        if (supports_dsm)
-          minimum_sm_tbc =
-              get_buffer_size_sm_dsm_plus_tbc_classic_programmable_bootstrap<
-                  Torus>(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 = (double2 *)cuda_malloc_async(
-            (glwe_dimension + 1) * level_count * input_lwe_ciphertext_count *
-                polynomial_size / 2 * sizeof(double2),
-            stream, gpu_index);
-      } break;
-#endif
-      default:
-        PANIC("Cuda error (PBS): unsupported implementation variant.")
+      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_with_size_tracking_async(
+          device_mem, stream, gpu_index, size_tracker, allocate_gpu_memory);
+
+      global_join_buffer = (double2 *)cuda_malloc_with_size_tracking_async(
+          (glwe_dimension + 1) * level_count * input_lwe_ciphertext_count *
+              (polynomial_size / 2) * sizeof(double2),
+          stream, gpu_index, size_tracker, allocate_gpu_memory);
+
+      global_accumulator = (Torus *)cuda_malloc_with_size_tracking_async(
+          (glwe_dimension + 1) * input_lwe_ciphertext_count * polynomial_size *
+              sizeof(Torus),
+          stream, gpu_index, size_tracker, allocate_gpu_memory);
+    } break;
+    case PBS_VARIANT::CG: {
+      uint64_t full_sm =
+          get_buffer_size_full_sm_programmable_bootstrap_cg<Torus>(
+              polynomial_size);
+      uint64_t partial_sm =
+          get_buffer_size_partial_sm_programmable_bootstrap_cg<Torus>(
+              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_with_size_tracking_async(
+          device_mem, stream, gpu_index, size_tracker, allocate_gpu_memory);
+
+      global_join_buffer = (double2 *)cuda_malloc_with_size_tracking_async(
+          (glwe_dimension + 1) * level_count * input_lwe_ciphertext_count *
+              polynomial_size / 2 * sizeof(double2),
+          stream, gpu_index, size_tracker, allocate_gpu_memory);
+    } break;
+#if CUDA_ARCH >= 900
+    case PBS_VARIANT::TBC: {
+
+      bool supports_dsm =
+          supports_distributed_shared_memory_on_classic_programmable_bootstrap<
+              Torus>(polynomial_size, max_shared_memory);
+
+      uint64_t full_sm =
+          get_buffer_size_full_sm_programmable_bootstrap_tbc<Torus>(
+              polynomial_size);
+      uint64_t partial_sm =
+          get_buffer_size_partial_sm_programmable_bootstrap_tbc<Torus>(
+              polynomial_size);
+      uint64_t minimum_sm_tbc = 0;
+      if (supports_dsm)
+        minimum_sm_tbc =
+            get_buffer_size_sm_dsm_plus_tbc_classic_programmable_bootstrap<
+                Torus>(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_with_size_tracking_async(
+          device_mem, stream, gpu_index, size_tracker, allocate_gpu_memory);
+
+      global_join_buffer = (double2 *)cuda_malloc_with_size_tracking_async(
+          (glwe_dimension + 1) * level_count * input_lwe_ciphertext_count *
+              polynomial_size / 2 * sizeof(double2),
+          stream, gpu_index, size_tracker, allocate_gpu_memory);
+    } 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);
+    cuda_drop_with_size_tracking_async(d_mem, stream, gpu_index,
+                                       gpu_memory_allocated);
+    cuda_drop_with_size_tracking_async(global_join_buffer, stream, gpu_index,
+                                       gpu_memory_allocated);
 
     if (pbs_variant == DEFAULT)
-      cuda_drop_async(global_accumulator, stream, gpu_index);
+      cuda_drop_with_size_tracking_async(global_accumulator, stream, gpu_index,
+                                         gpu_memory_allocated);
 
     if (uses_noise_reduction)
-      cuda_drop_async(temp_lwe_array_in, stream, gpu_index);
+      cuda_drop_with_size_tracking_async(temp_lwe_array_in, stream, gpu_index,
+                                         gpu_memory_allocated);
   }
 };
 
@@ -244,156 +248,185 @@ template <> struct pbs_buffer_128<PBS_TYPE::CLASSICAL> {
   __uint128_t *global_accumulator;
   double *global_join_buffer;
   __uint128_t *temp_lwe_array_in;
+  uint64_t *trivial_indexes;
 
   PBS_VARIANT pbs_variant;
   bool uses_noise_reduction;
+  bool gpu_memory_allocated;
 
   pbs_buffer_128(cudaStream_t stream, uint32_t gpu_index,
                  uint32_t lwe_dimension, 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, bool allocate_ms_array) {
+                 bool allocate_gpu_memory, bool allocate_ms_array,
+                 uint64_t *size_tracker) {
+    gpu_memory_allocated = allocate_gpu_memory;
     cuda_set_device(gpu_index);
     this->pbs_variant = pbs_variant;
     this->uses_noise_reduction = allocate_ms_array;
     if (allocate_ms_array) {
-      this->temp_lwe_array_in = (__uint128_t *)cuda_malloc_async(
-          (lwe_dimension + 1) * input_lwe_ciphertext_count *
-              sizeof(__uint128_t),
-          stream, gpu_index);
+      this->temp_lwe_array_in =
+          (__uint128_t *)cuda_malloc_with_size_tracking_async(
+              (lwe_dimension + 1) * input_lwe_ciphertext_count *
+                  sizeof(__uint128_t),
+              stream, gpu_index, size_tracker, allocate_ms_array);
+      this->trivial_indexes = (uint64_t *)cuda_malloc_with_size_tracking_async(
+          input_lwe_ciphertext_count * sizeof(uint64_t), stream, gpu_index,
+          size_tracker, allocate_ms_array);
+      uint64_t *h_trivial_indexes = new uint64_t[input_lwe_ciphertext_count];
+      for (uint32_t i = 0; i < input_lwe_ciphertext_count; i++)
+        h_trivial_indexes[i] = i;
+
+      cuda_memcpy_with_size_tracking_async_to_gpu(
+          trivial_indexes, h_trivial_indexes,
+          input_lwe_ciphertext_count * sizeof(uint64_t), stream, gpu_index,
+          allocate_gpu_memory);
+
+      cuda_synchronize_stream(stream, gpu_index);
+      delete[] h_trivial_indexes;
     }
     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);
+    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 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.")
+      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_with_size_tracking_async(
+          device_mem, stream, gpu_index, size_tracker, allocate_gpu_memory);
+
+      global_join_buffer = (double *)cuda_malloc_with_size_tracking_async(
+          global_join_buffer_size, stream, gpu_index, size_tracker,
+          allocate_gpu_memory);
+
+      global_accumulator = (__uint128_t *)cuda_malloc_with_size_tracking_async(
+          (glwe_dimension + 1) * input_lwe_ciphertext_count * polynomial_size *
+              sizeof(__uint128_t),
+          stream, gpu_index, size_tracker, allocate_gpu_memory);
+    } 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_with_size_tracking_async(
+          device_mem, stream, gpu_index, size_tracker, allocate_gpu_memory);
+
+      global_join_buffer = (double *)cuda_malloc_with_size_tracking_async(
+          global_join_buffer_size, stream, gpu_index, size_tracker,
+          allocate_gpu_memory);
+    } 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_with_size_tracking_async(
+          device_mem, stream, gpu_index, size_tracker, allocate_gpu_memory);
+
+      global_join_buffer = (double *)cuda_malloc_with_size_tracking_async(
+          global_join_buffer_size, stream, gpu_index, size_tracker,
+          allocate_gpu_memory);
+    } 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);
+    cuda_drop_with_size_tracking_async(d_mem, stream, gpu_index,
+                                       gpu_memory_allocated);
+    cuda_drop_with_size_tracking_async(global_join_buffer, stream, gpu_index,
+                                       gpu_memory_allocated);
 
     if (pbs_variant == DEFAULT)
-      cuda_drop_async(global_accumulator, stream, gpu_index);
+      cuda_drop_with_size_tracking_async(global_accumulator, stream, gpu_index,
+                                         gpu_memory_allocated);
 
-    if (uses_noise_reduction)
-      cuda_drop_async(temp_lwe_array_in, stream, gpu_index);
+    if (uses_noise_reduction) {
+      cuda_drop_with_size_tracking_async(temp_lwe_array_in, stream, gpu_index,
+                                         gpu_memory_allocated);
+      cuda_drop_with_size_tracking_async(trivial_indexes, stream, gpu_index,
+                                         gpu_memory_allocated);
+    }
   }
 };
 
@@ -464,7 +497,7 @@ void cuda_programmable_bootstrap_tbc_lwe_ciphertext_vector(
     uint32_t lut_stride);
 
 template <typename Torus>
-void scratch_cuda_programmable_bootstrap_tbc(
+uint64_t scratch_cuda_programmable_bootstrap_tbc(
     void *stream, uint32_t gpu_index, pbs_buffer<Torus, CLASSICAL> **pbs_buffer,
     uint32_t lwe_dimension, uint32_t glwe_dimension, uint32_t polynomial_size,
     uint32_t level_count, uint32_t input_lwe_ciphertext_count,
@@ -472,14 +505,14 @@ void scratch_cuda_programmable_bootstrap_tbc(
 #endif
 
 template <typename Torus>
-void scratch_cuda_programmable_bootstrap_cg(
+uint64_t scratch_cuda_programmable_bootstrap_cg(
     void *stream, uint32_t gpu_index, pbs_buffer<Torus, CLASSICAL> **pbs_buffer,
     uint32_t lwe_dimension, uint32_t glwe_dimension, uint32_t polynomial_size,
     uint32_t level_count, uint32_t input_lwe_ciphertext_count,
     bool allocate_gpu_memory, bool allocate_ms_array);
 
 template <typename Torus>
-void scratch_cuda_programmable_bootstrap(
+uint64_t scratch_cuda_programmable_bootstrap(
     void *stream, uint32_t gpu_index, pbs_buffer<Torus, CLASSICAL> **buffer,
     uint32_t lwe_dimension, uint32_t glwe_dimension, uint32_t polynomial_size,
     uint32_t level_count, uint32_t input_lwe_ciphertext_count,
@@ -489,7 +522,8 @@ template <typename Torus>
 bool has_support_to_cuda_programmable_bootstrap_tbc(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);
 
 #ifdef __CUDACC__
 __device__ inline int get_start_ith_ggsw(int i, uint32_t polynomial_size,

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

@@ -25,12 +25,12 @@ void cuda_convert_lwe_programmable_bootstrap_key_128(
     uint32_t input_lwe_dim, uint32_t glwe_dim, uint32_t level_count,
     uint32_t polynomial_size);
 
-void scratch_cuda_programmable_bootstrap_amortized_32(
+uint64_t scratch_cuda_programmable_bootstrap_amortized_32(
     void *stream, uint32_t gpu_index, int8_t **pbs_buffer,
     uint32_t glwe_dimension, uint32_t polynomial_size,
     uint32_t input_lwe_ciphertext_count, bool allocate_gpu_memory);
 
-void scratch_cuda_programmable_bootstrap_amortized_64(
+uint64_t scratch_cuda_programmable_bootstrap_amortized_64(
     void *stream, uint32_t gpu_index, int8_t **pbs_buffer,
     uint32_t glwe_dimension, uint32_t polynomial_size,
     uint32_t input_lwe_ciphertext_count, bool allocate_gpu_memory);
@@ -57,19 +57,19 @@ void cleanup_cuda_programmable_bootstrap_amortized(void *stream,
                                                    uint32_t gpu_index,
                                                    int8_t **pbs_buffer);
 
-void scratch_cuda_programmable_bootstrap_32(
+uint64_t scratch_cuda_programmable_bootstrap_32(
     void *stream, uint32_t gpu_index, int8_t **buffer, uint32_t lwe_dimension,
     uint32_t glwe_dimension, uint32_t polynomial_size, uint32_t level_count,
     uint32_t input_lwe_ciphertext_count, bool allocate_gpu_memory,
     bool allocate_ms_array);
 
-void scratch_cuda_programmable_bootstrap_64(
+uint64_t scratch_cuda_programmable_bootstrap_64(
     void *stream, uint32_t gpu_index, int8_t **buffer, uint32_t lwe_dimension,
     uint32_t glwe_dimension, uint32_t polynomial_size, uint32_t level_count,
     uint32_t input_lwe_ciphertext_count, bool allocate_gpu_memory,
     bool allocate_ms_array);
 
-void scratch_cuda_programmable_bootstrap_128(
+uint64_t scratch_cuda_programmable_bootstrap_128(
     void *stream, uint32_t gpu_index, int8_t **buffer, uint32_t lwe_dimension,
     uint32_t glwe_dimension, uint32_t polynomial_size, uint32_t level_count,
     uint32_t input_lwe_ciphertext_count, bool allocate_gpu_memory,
@@ -90,18 +90,19 @@ void cuda_programmable_bootstrap_lwe_ciphertext_vector_64(
     void const *lut_vector_indexes, void const *lwe_array_in,
     void const *lwe_input_indexes, void const *bootstrapping_key,
     CudaModulusSwitchNoiseReductionKeyFFI const *ms_noise_reduction_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 num_many_lut, uint32_t lut_stride);
+    void *ms_noise_reduction_ptr, 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 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,
     CudaModulusSwitchNoiseReductionKeyFFI const *ms_noise_reduction_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 *ms_noise_reduction_ptr, 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);

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

@@ -15,7 +15,7 @@ void cuda_convert_lwe_multi_bit_programmable_bootstrap_key_64(
     uint32_t input_lwe_dim, uint32_t glwe_dim, uint32_t level_count,
     uint32_t polynomial_size, uint32_t grouping_factor);
 
-void scratch_cuda_multi_bit_programmable_bootstrap_64(
+uint64_t scratch_cuda_multi_bit_programmable_bootstrap_64(
     void *stream, uint32_t gpu_index, int8_t **pbs_buffer,
     uint32_t glwe_dimension, uint32_t polynomial_size, uint32_t level_count,
     uint32_t input_lwe_ciphertext_count, bool allocate_gpu_memory);