chore(ci): remove wasm test that is not relevant

- allows to use ZK parameters directly

.github/actionlint.yaml

@@ -3,6 +3,8 @@ self-hosted-runner:
   labels:
     - m1mac
     - 4090-desktop
+    - large_windows_16_latest
+    - large_ubuntu_16
 # 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/workflows/aws_tfhe_backward_compat_tests.yml

@@ -0,0 +1,121 @@
+# Run backward compatibility tests
+name: Backward compatibility Tests on CPU
+
+env:
+  CARGO_TERM_COLOR: always
+  ACTION_RUN_URL: ${{ github.server_url }}/${{ github.repository }}/actions/runs/${{ github.run_id }}
+  RUSTFLAGS: "-C target-cpu=native"
+  RUST_BACKTRACE: "full"
+  RUST_MIN_STACK: "8388608"
+  SLACK_CHANNEL: ${{ secrets.SLACK_CHANNEL }}
+  SLACK_ICON: https://pbs.twimg.com/profile_images/1274014582265298945/OjBKP9kn_400x400.png
+  SLACK_USERNAME: ${{ secrets.BOT_USERNAME }}
+  SLACK_WEBHOOK: ${{ secrets.SLACK_WEBHOOK }}
+
+on:
+  # Allows you to run this workflow manually from the Actions tab as an alternative.
+  workflow_dispatch:
+  pull_request:
+
+jobs:
+  setup-instance:
+    name: Setup instance (backward-compat-tests)
+    runs-on: ubuntu-latest
+    outputs:
+      runner-name: ${{ steps.start-instance.outputs.label }}
+    steps:
+      - name: Start instance
+        id: start-instance
+        uses: zama-ai/slab-github-runner@c0e7168795bd78f61f61146951ed9d0c73c9b701
+        with:
+          mode: start
+          github-token: ${{ secrets.SLAB_ACTION_TOKEN }}
+          slab-url: ${{ secrets.SLAB_BASE_URL }}
+          job-secret: ${{ secrets.JOB_SECRET }}
+          backend: aws
+          profile: cpu-small
+
+  backward-compat-tests:
+    name: Backward compatibility tests
+    needs: [ setup-instance ]
+    concurrency:
+      group: ${{ github.workflow }}_${{ github.ref }}
+      cancel-in-progress: true
+    runs-on: ${{ needs.setup-instance.outputs.runner-name }}
+    steps:
+      - name: Checkout tfhe-rs
+        uses: actions/checkout@d632683dd7b4114ad314bca15554477dd762a938
+        with:
+          persist-credentials: 'false'
+          token: ${{ secrets.FHE_ACTIONS_TOKEN }}
+
+      - name: Set up home
+        run: |
+          echo "HOME=/home/ubuntu" >> "${GITHUB_ENV}"
+
+      - name: Install latest stable
+        uses: dtolnay/rust-toolchain@7b1c307e0dcbda6122208f10795a713336a9b35a
+        with:
+          toolchain: stable
+
+      - name: Install git-lfs
+        run: |
+          sudo apt update && sudo apt -y install git-lfs
+
+      - name: Use specific data branch
+        if: ${{ contains(github.event.pull_request.labels.*.name, 'data_PR') }}
+        env:
+          PR_BRANCH: ${{ github.head_ref || github.ref_name }}
+        run: |
+          echo "BACKWARD_COMPAT_DATA_BRANCH=${PR_BRANCH}" >> "${GITHUB_ENV}"
+
+      - name: Get backward compat branch
+        id: backward_compat_branch
+        run: |
+          BRANCH="$(make backward_compat_branch)"
+          echo "branch=${BRANCH}" >> "${GITHUB_OUTPUT}"
+
+      - name: Clone test data
+        uses: actions/checkout@d632683dd7b4114ad314bca15554477dd762a938
+        with:
+          persist-credentials: 'false'
+          repository: zama-ai/tfhe-backward-compat-data
+          path: tfhe/tfhe-backward-compat-data
+          lfs: 'true'
+          ref: ${{ steps.backward_compat_branch.outputs.branch }}
+
+      - name: Run backward compatibility tests
+        run: |
+          make test_backward_compatibility_ci
+
+      - name: Slack Notification
+        if: ${{ failure() }}
+        continue-on-error: true
+        uses: rtCamp/action-slack-notify@4e5fb42d249be6a45a298f3c9543b111b02f7907
+        env:
+          SLACK_COLOR: ${{ job.status }}
+          SLACK_MESSAGE: "Backward compatibility tests finished with status: ${{ job.status }}. (${{ env.ACTION_RUN_URL }})"
+
+  teardown-instance:
+    name: Teardown instance (backward-compat-tests)
+    if: ${{ always() && needs.setup-instance.result != 'skipped' }}
+    needs: [ setup-instance, backward-compat-tests ]
+    runs-on: ubuntu-latest
+    steps:
+      - name: Stop instance
+        id: stop-instance
+        uses: zama-ai/slab-github-runner@c0e7168795bd78f61f61146951ed9d0c73c9b701
+        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@4e5fb42d249be6a45a298f3c9543b111b02f7907
+        env:
+          SLACK_COLOR: ${{ job.status }}
+          SLACK_MESSAGE: "Instance teardown (backward-compat-tests) finished with status: ${{ job.status }}. (${{ env.ACTION_RUN_URL }})"

.github/workflows/aws_tfhe_fast_tests.yml

@@ -1,4 +1,4 @@
-# Run a small subset of shortint and integer tests to ensure quick feedback.
+# Run a small subset of tests to ensure quick feedback.
 name: Fast AWS Tests on CPU
 
 env:
@@ -11,6 +11,7 @@ env:
   SLACK_ICON: https://pbs.twimg.com/profile_images/1274014582265298945/OjBKP9kn_400x400.png
   SLACK_USERNAME: ${{ secrets.BOT_USERNAME }}
   SLACK_WEBHOOK: ${{ secrets.SLACK_WEBHOOK }}
+  IS_PULL_REQUEST: ${{ github.event_name == 'pull_request' }}
 
 on:
   # Allows you to run this workflow manually from the Actions tab as an alternative.
@@ -18,15 +19,119 @@ on:
   pull_request:
 
 jobs:
+  should-run:
+    runs-on: ubuntu-latest
+    permissions:
+      pull-requests: write
+    outputs:
+      csprng_test: ${{ env.IS_PULL_REQUEST == 'false' || steps.changed-files.outputs.csprng_any_changed }}
+      zk_pok_test: ${{ env.IS_PULL_REQUEST == 'false' || steps.changed-files.outputs.zk_pok_any_changed }}
+      versionable_test: ${{ env.IS_PULL_REQUEST == 'false' || steps.changed-files.outputs.versionable_any_changed }}
+      core_crypto_test: ${{ env.IS_PULL_REQUEST == 'false' ||
+        steps.changed-files.outputs.core_crypto_any_changed ||
+        steps.changed-files.outputs.dependencies_any_changed }}
+      boolean_test: ${{ env.IS_PULL_REQUEST == 'false' ||
+        steps.changed-files.outputs.boolean_any_changed ||
+        steps.changed-files.outputs.dependencies_any_changed }}
+      shortint_test: ${{ env.IS_PULL_REQUEST == 'false' ||
+        steps.changed-files.outputs.shortint_any_changed ||
+        steps.changed-files.outputs.dependencies_any_changed }}
+      integer_test: ${{ env.IS_PULL_REQUEST == 'false' ||
+        steps.changed-files.outputs.integer_any_changed ||
+        steps.changed-files.outputs.dependencies_any_changed }}
+      wasm_test: ${{ env.IS_PULL_REQUEST == 'false' ||
+        steps.changed-files.outputs.wasm_any_changed ||
+        steps.changed-files.outputs.dependencies_any_changed }}
+      high_level_api_test: ${{ env.IS_PULL_REQUEST == 'false' ||
+        steps.changed-files.outputs.high_level_api_any_changed ||
+        steps.changed-files.outputs.dependencies_any_changed }}
+      user_docs_test: ${{ env.IS_PULL_REQUEST == 'false' ||
+        steps.changed-files.outputs.user_docs_any_changed ||
+        steps.changed-files.outputs.dependencies_any_changed }}
+      any_file_changed: ${{ env.IS_PULL_REQUEST == 'false' || steps.aggregated-changes.outputs.any_changed }}
+    steps:
+      - name: Checkout tfhe-rs
+        uses: actions/checkout@d632683dd7b4114ad314bca15554477dd762a938
+        with:
+          fetch-depth: 0
+
+      - name: Check for file changes
+        id: changed-files
+        uses: tj-actions/changed-files@c3a1bb2c992d77180ae65be6ae6c166cf40f857c
+        with:
+          since_last_remote_commit: true
+          files_yaml: |
+            dependencies:
+              - tfhe/Cargo.toml
+              - concrete-csprng/**
+              - tfhe-zk-pok/**
+              - utils/tfhe-versionable/**
+              - utils/tfhe-versionable-derive/**
+            csprng:
+              - concrete-csprng/**
+            zk_pok:
+              - tfhe-zk-pok/**
+            versionable:
+              - utils/tfhe-versionable/**
+              - utils/tfhe-versionable-derive/**
+            core_crypto:
+              - tfhe/src/core_crypto/**
+            boolean:
+              - tfhe/src/core_crypto/**
+              - tfhe/src/boolean/**
+            shortint:
+              - tfhe/src/core_crypto/**
+              - tfhe/src/shortint/**
+            integer:
+              - tfhe/src/core_crypto/**
+              - tfhe/src/shortint/**
+              - tfhe/src/integer/**
+            wasm:
+              - tfhe/src/**
+              - tfhe/js_on_wasm_tests/**
+              - tfhe/web_wasm_parallel_tests/**
+              - '!tfhe/src/c_api/**'
+              - '!tfhe/src/boolean/**'
+            high_level_api:
+              - tfhe/src/**
+              - '!tfhe/src/c_api/**'
+              - '!tfhe/src/boolean/**'
+              - '!tfhe/src/c_api/**'
+              - '!tfhe/src/js_on_wasm_api/**'
+            user_docs:
+              - tfhe/src/**
+              - '!tfhe/src/c_api/**'
+              - 'tfhe/docs/**.md'
+              - README.md
+
+      - name: Aggregate file changes
+        id: aggregated-changes
+        if: ( steps.changed-files.outputs.dependencies_any_changed == 'true' ||
+          steps.changed-files.outputs.csprng_any_changed == 'true' ||
+          steps.changed-files.outputs.zk_pok_any_changed == 'true' ||
+          steps.changed-files.outputs.versionable_any_changed == 'true' ||
+          steps.changed-files.outputs.core_crypto_any_changed == 'true' ||
+          steps.changed-files.outputs.boolean_any_changed == 'true' ||
+          steps.changed-files.outputs.shortint_any_changed == 'true' ||
+          steps.changed-files.outputs.integer_any_changed == 'true' ||
+          steps.changed-files.outputs.wasm_any_changed == 'true' ||
+          steps.changed-files.outputs.high_level_api_any_changed == 'true' ||
+          steps.changed-files.outputs.user_docs_any_changed == 'true')
+        run: |
+          echo "any_changed=true" >> "$GITHUB_OUTPUT"
+
   setup-instance:
     name: Setup instance (fast-tests)
+    if: github.event_name != 'pull_request' ||
+      needs.should-run.outputs.any_file_changed == 'true'
+    needs: should-run
     runs-on: ubuntu-latest
     outputs:
       runner-name: ${{ steps.start-instance.outputs.label }}
     steps:
       - name: Start instance
         id: start-instance
-        uses: zama-ai/slab-github-runner@58f2cae4bf2c0b6728083f5f009b6dc0eb6dc3ac
+        uses: zama-ai/slab-github-runner@c0e7168795bd78f61f61146951ed9d0c73c9b701
         with:
           mode: start
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}
@@ -37,75 +142,82 @@ jobs:
 
   fast-tests:
     name: Fast CPU tests
-    needs: setup-instance
+    if: github.event_name != 'pull_request' ||
+      (github.event_name == 'pull_request' && needs.setup-instance.result != 'skipped')
+    needs: [ should-run, setup-instance ]
     concurrency:
       group: ${{ github.workflow }}_${{ github.ref }}
       cancel-in-progress: true
     runs-on: ${{ needs.setup-instance.outputs.runner-name }}
     steps:
       - name: Checkout tfhe-rs
-        uses: actions/checkout@692973e3d937129bcbf40652eb9f2f61becf3332
+        uses: actions/checkout@d632683dd7b4114ad314bca15554477dd762a938
         with:
           persist-credentials: 'false'
+          token: ${{ secrets.FHE_ACTIONS_TOKEN }}
 
       - name: Set up home
         run: |
           echo "HOME=/home/ubuntu" >> "${GITHUB_ENV}"
 
       - name: Install latest stable
-        uses: dtolnay/rust-toolchain@21dc36fb71dd22e3317045c0c31a3f4249868b17
+        uses: dtolnay/rust-toolchain@7b1c307e0dcbda6122208f10795a713336a9b35a
         with:
           toolchain: stable
 
-      - name: Install git-lfs
-        run: |
-          sudo apt update && sudo apt -y install git-lfs
-
       - name: Run concrete-csprng tests
+        if: needs.should-run.outputs.csprng_test == 'true'
         run: |
           make test_concrete_csprng
 
       - name: Run tfhe-zk-pok tests
+        if: needs.should-run.outputs.zk_pok_test == 'true'
         run: |
           make test_zk_pok
 
+      - name: Run tfhe-versionable tests
+        if: needs.should-run.outputs.versionable_test == 'true'
+        run: |
+          make test_versionable
+
       - name: Run core tests
+        if: needs.should-run.outputs.core_crypto_test == 'true'
         run: |
           AVX512_SUPPORT=ON make test_core_crypto
 
       - name: Run boolean tests
+        if: needs.should-run.outputs.boolean_test == 'true'
         run: |
           make test_boolean
 
       - name: Run user docs tests
+        if: needs.should-run.outputs.user_docs_test == 'true'
         run: |
           make test_user_doc
 
       - name: Run js on wasm API tests
+        if: needs.should-run.outputs.wasm_test == 'true'
         run: |
           make test_nodejs_wasm_api_in_docker
 
       - name: Gen Keys if required
+        if: needs.should-run.outputs.shortint_test == 'true' ||
+          needs.should-run.outputs.integer_test == 'true'
         run: |
           make gen_key_cache
 
       - name: Run shortint tests
+        if: needs.should-run.outputs.shortint_test == 'true'
         run: |
           BIG_TESTS_INSTANCE=TRUE FAST_TESTS=TRUE make test_shortint_ci
 
       - name: Run integer tests
+        if: needs.should-run.outputs.integer_test == 'true'
         run: |
           BIG_TESTS_INSTANCE=TRUE FAST_TESTS=TRUE make test_integer_ci
 
-      - name: Run shortint multi-bit tests
-        run: |
-          BIG_TESTS_INSTANCE=TRUE FAST_TESTS=TRUE make test_shortint_multi_bit_ci
-
-      - name: Run integer multi-bit tests
-        run: |
-          BIG_TESTS_INSTANCE=TRUE FAST_TESTS=TRUE make test_integer_multi_bit_ci
-
       - name: Run high-level API tests
+        if: needs.should-run.outputs.high_level_api_test == 'true'
         run: |
           make test_high_level_api
 
@@ -113,19 +225,8 @@ jobs:
         run: |
           make test_safe_deserialization
 
-      - name: Clone test data
-        uses: actions/checkout@692973e3d937129bcbf40652eb9f2f61becf3332
-        with:
-          repository: zama-ai/tfhe-backward-compat-data
-          path: tfhe/tfhe-backward-compat-data
-          lfs: 'true'
-
-      - name: Run backward compatibility tests
-        run: |
-          make test_backward_compatibility_ci
-
       - name: Slack Notification
-        if: ${{ always() }}
+        if: ${{ failure() }}
         continue-on-error: true
         uses: rtCamp/action-slack-notify@4e5fb42d249be6a45a298f3c9543b111b02f7907
         env:
@@ -140,7 +241,7 @@ jobs:
     steps:
       - name: Stop instance
         id: stop-instance
-        uses: zama-ai/slab-github-runner@58f2cae4bf2c0b6728083f5f009b6dc0eb6dc3ac
+        uses: zama-ai/slab-github-runner@c0e7168795bd78f61f61146951ed9d0c73c9b701
         with:
           mode: stop
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}

.github/workflows/aws_tfhe_integer_tests.yml

@@ -13,24 +13,66 @@ env:
   # We clear the cache to reduce memory pressure because of the numerous processes of cargo
   # nextest
   TFHE_RS_CLEAR_IN_MEMORY_KEY_CACHE: "1"
+  NO_BIG_PARAMS: FALSE
 
 on:
   # Allows you to run this workflow manually from the Actions tab as an alternative.
   workflow_dispatch:
   pull_request:
-    types: [ labeled ]
+    types: [labeled]
+  push:
+    branches:
+      - main
 
 jobs:
+  should-run:
+    if:
+      (github.event_name == 'push' && github.repository == 'zama-ai/tfhe-rs') ||
+      (github.event_name == 'schedule' && github.repository == 'zama-ai/tfhe-rs') ||
+      (github.event_name == 'pull_request' && contains(github.event.label.name, 'approved')) ||
+      github.event_name == 'workflow_dispatch'
+    runs-on: ubuntu-latest
+    permissions:
+      pull-requests: write
+    outputs:
+      integer_test: ${{ github.event_name == 'workflow_dispatch' ||
+        steps.changed-files.outputs.integer_any_changed }}
+    steps:
+      - name: Checkout tfhe-rs
+        uses: actions/checkout@d632683dd7b4114ad314bca15554477dd762a938
+        with:
+          fetch-depth: 0
+          persist-credentials: "false"
+
+      - name: Check for file changes
+        id: changed-files
+        uses: tj-actions/changed-files@c3a1bb2c992d77180ae65be6ae6c166cf40f857c
+        with:
+          since_last_remote_commit: true
+          files_yaml: |
+            integer:
+              - tfhe/Cargo.toml
+              - concrete-csprng/**
+              - tfhe-zk-pok/**
+              - tfhe/src/core_crypto/**
+              - tfhe/src/shortint/**
+              - tfhe/src/integer/**
+
   setup-instance:
     name: Setup instance (unsigned-integer-tests)
-    if: ${{ github.event_name == 'workflow_dispatch' || contains(github.event.label.name, 'approved') }}
+    needs: should-run
+    if:
+      (github.event_name == 'push' && github.repository == 'zama-ai/tfhe-rs' && needs.should-run.outputs.integer_test == 'true') ||
+      (github.event_name == 'schedule' && github.repository == 'zama-ai/tfhe-rs') ||
+      (github.event_name == 'pull_request' && contains(github.event.label.name, 'approved')) ||
+      github.event_name == 'workflow_dispatch'
     runs-on: ubuntu-latest
     outputs:
       runner-name: ${{ steps.start-instance.outputs.label }}
     steps:
       - name: Start instance
         id: start-instance
-        uses: zama-ai/slab-github-runner@58f2cae4bf2c0b6728083f5f009b6dc0eb6dc3ac
+        uses: zama-ai/slab-github-runner@c0e7168795bd78f61f61146951ed9d0c73c9b701
         with:
           mode: start
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}
@@ -43,24 +85,30 @@ jobs:
     name: Unsigned integer tests
     needs: setup-instance
     concurrency:
-      group: ${{ github.workflow }}_${{ github.ref }}
-      cancel-in-progress: true
+      group: ${{ github.workflow }}_${{ github.ref }}${{ github.ref == 'refs/heads/main' && github.sha || '' }}
+      cancel-in-progress: ${{ github.ref != 'refs/heads/main' }}
     runs-on: ${{ needs.setup-instance.outputs.runner-name }}
     steps:
       - name: Checkout tfhe-rs
-        uses: actions/checkout@692973e3d937129bcbf40652eb9f2f61becf3332
+        uses: actions/checkout@d632683dd7b4114ad314bca15554477dd762a938
         with:
-          persist-credentials: 'false'
+          persist-credentials: "false"
+          token: ${{ secrets.FHE_ACTIONS_TOKEN }}
 
       - name: Set up home
         run: |
           echo "HOME=/home/ubuntu" >> "${GITHUB_ENV}"
 
       - name: Install latest stable
-        uses: dtolnay/rust-toolchain@21dc36fb71dd22e3317045c0c31a3f4249868b17
+        uses: dtolnay/rust-toolchain@7b1c307e0dcbda6122208f10795a713336a9b35a
         with:
           toolchain: stable
 
+      - name: Should skip big parameters set
+        if: github.event_name == 'pull_request'
+        run: |
+          echo "NO_BIG_PARAMS=TRUE" >> "${GITHUB_ENV}"
+
       - name: Gen Keys if required
         run: |
           make GEN_KEY_CACHE_MULTI_BIT_ONLY=TRUE gen_key_cache
@@ -75,10 +123,10 @@ jobs:
 
       - name: Run unsigned integer tests
         run: |
-          AVX512_SUPPORT=ON BIG_TESTS_INSTANCE=TRUE make test_unsigned_integer_ci
+          AVX512_SUPPORT=ON NO_BIG_PARAMS=${{ env.NO_BIG_PARAMS }} BIG_TESTS_INSTANCE=TRUE make test_unsigned_integer_ci
 
       - name: Slack Notification
-        if: ${{ always() }}
+        if: ${{ failure() }}
         continue-on-error: true
         uses: rtCamp/action-slack-notify@4e5fb42d249be6a45a298f3c9543b111b02f7907
         env:
@@ -88,12 +136,12 @@ jobs:
   teardown-instance:
     name: Teardown instance (unsigned-integer-tests)
     if: ${{ always() && needs.setup-instance.result != 'skipped' }}
-    needs: [ setup-instance, unsigned-integer-tests ]
+    needs: [setup-instance, unsigned-integer-tests]
     runs-on: ubuntu-latest
     steps:
       - name: Stop instance
         id: stop-instance
-        uses: zama-ai/slab-github-runner@58f2cae4bf2c0b6728083f5f009b6dc0eb6dc3ac
+        uses: zama-ai/slab-github-runner@c0e7168795bd78f61f61146951ed9d0c73c9b701
         with:
           mode: stop
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}

.github/workflows/aws_tfhe_signed_integer_tests.yml

@@ -13,24 +13,66 @@ env:
   # We clear the cache to reduce memory pressure because of the numerous processes of cargo
   # nextest
   TFHE_RS_CLEAR_IN_MEMORY_KEY_CACHE: "1"
+  NO_BIG_PARAMS: FALSE
 
 on:
   # Allows you to run this workflow manually from the Actions tab as an alternative.
   workflow_dispatch:
   pull_request:
-    types: [ labeled ]
+    types: [labeled]
+  push:
+    branches:
+      - main
 
 jobs:
+  should-run:
+    if:
+      (github.event_name == 'push' && github.repository == 'zama-ai/tfhe-rs') ||
+      (github.event_name == 'schedule' && github.repository == 'zama-ai/tfhe-rs') ||
+      (github.event_name == 'pull_request' && contains(github.event.label.name, 'approved')) ||
+      github.event_name == 'workflow_dispatch'
+    runs-on: ubuntu-latest
+    permissions:
+      pull-requests: write
+    outputs:
+      integer_test: ${{ github.event_name == 'workflow_dispatch' ||
+        steps.changed-files.outputs.integer_any_changed }}
+    steps:
+      - name: Checkout tfhe-rs
+        uses: actions/checkout@d632683dd7b4114ad314bca15554477dd762a938
+        with:
+          fetch-depth: 0
+          persist-credentials: "false"
+
+      - name: Check for file changes
+        id: changed-files
+        uses: tj-actions/changed-files@c3a1bb2c992d77180ae65be6ae6c166cf40f857c
+        with:
+          since_last_remote_commit: true
+          files_yaml: |
+            integer:
+              - tfhe/Cargo.toml
+              - concrete-csprng/**
+              - tfhe-zk-pok/**
+              - tfhe/src/core_crypto/**
+              - tfhe/src/shortint/**
+              - tfhe/src/integer/**
+
   setup-instance:
-    name: Setup instance (signed-integer-tests)
-    if: ${{ github.event_name == 'workflow_dispatch' || contains(github.event.label.name, 'approved') }}
+    name: Setup instance (unsigned-integer-tests)
+    needs: should-run
+    if:
+      (github.event_name == 'push' && github.repository == 'zama-ai/tfhe-rs' && needs.should-run.outputs.integer_test == 'true') ||
+      (github.event_name == 'schedule' && github.repository == 'zama-ai/tfhe-rs') ||
+      (github.event_name == 'pull_request' && contains(github.event.label.name, 'approved')) ||
+      github.event_name == 'workflow_dispatch'
     runs-on: ubuntu-latest
     outputs:
       runner-name: ${{ steps.start-instance.outputs.label }}
     steps:
       - name: Start instance
         id: start-instance
-        uses: zama-ai/slab-github-runner@58f2cae4bf2c0b6728083f5f009b6dc0eb6dc3ac
+        uses: zama-ai/slab-github-runner@c0e7168795bd78f61f61146951ed9d0c73c9b701
         with:
           mode: start
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}
@@ -43,24 +85,30 @@ jobs:
     name: Signed integer tests
     needs: setup-instance
     concurrency:
-      group: ${{ github.workflow }}_${{ github.ref }}
-      cancel-in-progress: true
+      group: ${{ github.workflow }}_${{ github.ref }}${{ github.ref == 'refs/heads/main' && github.sha || '' }}
+      cancel-in-progress: ${{ github.ref != 'refs/heads/main' }}
     runs-on: ${{ needs.setup-instance.outputs.runner-name }}
     steps:
       - name: Checkout tfhe-rs
-        uses: actions/checkout@692973e3d937129bcbf40652eb9f2f61becf3332
+        uses: actions/checkout@d632683dd7b4114ad314bca15554477dd762a938
         with:
-          persist-credentials: 'false'
+          persist-credentials: "false"
+          token: ${{ secrets.FHE_ACTIONS_TOKEN }}
 
       - name: Set up home
         run: |
           echo "HOME=/home/ubuntu" >> "${GITHUB_ENV}"
 
       - name: Install latest stable
-        uses: dtolnay/rust-toolchain@21dc36fb71dd22e3317045c0c31a3f4249868b17
+        uses: dtolnay/rust-toolchain@7b1c307e0dcbda6122208f10795a713336a9b35a
         with:
           toolchain: stable
 
+      - name: Should skip big parameters set
+        if: github.event_name == 'pull_request'
+        run: |
+          echo "NO_BIG_PARAMS=TRUE" >> "${GITHUB_ENV}"
+
       - name: Gen Keys if required
         run: |
           make GEN_KEY_CACHE_MULTI_BIT_ONLY=TRUE gen_key_cache
@@ -79,10 +127,10 @@ jobs:
 
       - name: Run signed integer tests
         run: |
-          AVX512_SUPPORT=ON BIG_TESTS_INSTANCE=TRUE make test_signed_integer_ci
+          AVX512_SUPPORT=ON NO_BIG_PARAMS=${{ env.NO_BIG_PARAMS }} BIG_TESTS_INSTANCE=TRUE make test_signed_integer_ci
 
       - name: Slack Notification
-        if: ${{ always() }}
+        if: ${{ failure() }}
         continue-on-error: true
         uses: rtCamp/action-slack-notify@4e5fb42d249be6a45a298f3c9543b111b02f7907
         env:
@@ -92,12 +140,12 @@ jobs:
   teardown-instance:
     name: Teardown instance (signed-integer-tests)
     if: ${{ always() && needs.setup-instance.result != 'skipped' }}
-    needs: [ setup-instance, signed-integer-tests ]
+    needs: [setup-instance, signed-integer-tests]
     runs-on: ubuntu-latest
     steps:
       - name: Stop instance
         id: stop-instance
-        uses: zama-ai/slab-github-runner@58f2cae4bf2c0b6728083f5f009b6dc0eb6dc3ac
+        uses: zama-ai/slab-github-runner@c0e7168795bd78f61f61146951ed9d0c73c9b701
         with:
           mode: stop
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}

.github/workflows/aws_tfhe_tests.yml

@@ -57,13 +57,13 @@ jobs:
       any_file_changed: ${{ env.IS_PULL_REQUEST == 'false' || steps.aggregated-changes.outputs.any_changed }}
     steps:
       - name: Checkout tfhe-rs
-        uses: actions/checkout@692973e3d937129bcbf40652eb9f2f61becf3332
+        uses: actions/checkout@d632683dd7b4114ad314bca15554477dd762a938
         with:
           fetch-depth: 0
 
       - name: Check for file changes
         id: changed-files
-        uses: tj-actions/changed-files@eaf854ef0c266753e1abec356dcf17d92695b251
+        uses: tj-actions/changed-files@c3a1bb2c992d77180ae65be6ae6c166cf40f857c
         with:
           since_last_remote_commit: true
           files_yaml: |
@@ -86,6 +86,8 @@ jobs:
             high_level_api:
               - tfhe/src/**
               - '!tfhe/src/c_api/**'
+              - '!tfhe/src/boolean/**'
+              - '!tfhe/src/js_on_wasm_api/**'
             c_api:
               - tfhe/src/**
             examples:
@@ -121,7 +123,7 @@ jobs:
   setup-instance:
     name: Setup instance (cpu-tests)
     if: github.event_name != 'pull_request' ||
-      (github.event_name == 'pull_request' && needs.should-run.outputs.any_file_changed == 'true')
+      (github.event.action == 'labeled' && github.event.label.name == 'approved' && needs.should-run.outputs.any_file_changed == 'true')
     needs: should-run
     runs-on: ubuntu-latest
     outputs:
@@ -129,7 +131,7 @@ jobs:
     steps:
       - name: Start instance
         id: start-instance
-        uses: zama-ai/slab-github-runner@58f2cae4bf2c0b6728083f5f009b6dc0eb6dc3ac
+        uses: zama-ai/slab-github-runner@c0e7168795bd78f61f61146951ed9d0c73c9b701
         with:
           mode: start
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}
@@ -149,16 +151,17 @@ jobs:
     runs-on: ${{ needs.setup-instance.outputs.runner-name }}
     steps:
       - name: Checkout tfhe-rs
-        uses: actions/checkout@692973e3d937129bcbf40652eb9f2f61becf3332
+        uses: actions/checkout@d632683dd7b4114ad314bca15554477dd762a938
         with:
           persist-credentials: 'false'
+          token: ${{ secrets.FHE_ACTIONS_TOKEN }}
 
       - name: Set up home
         run: |
           echo "HOME=/home/ubuntu" >> "${GITHUB_ENV}"
 
       - name: Install latest stable
-        uses: dtolnay/rust-toolchain@21dc36fb71dd22e3317045c0c31a3f4249868b17
+        uses: dtolnay/rust-toolchain@7b1c307e0dcbda6122208f10795a713336a9b35a
         with:
           toolchain: stable
 
@@ -220,7 +223,7 @@ jobs:
           make test_kreyvium
 
       - name: Slack Notification
-        if: ${{ always() }}
+        if: ${{ failure() }}
         continue-on-error: true
         uses: rtCamp/action-slack-notify@4e5fb42d249be6a45a298f3c9543b111b02f7907
         env:
@@ -235,7 +238,7 @@ jobs:
     steps:
       - name: Stop instance
         id: stop-instance
-        uses: zama-ai/slab-github-runner@58f2cae4bf2c0b6728083f5f009b6dc0eb6dc3ac
+        uses: zama-ai/slab-github-runner@c0e7168795bd78f61f61146951ed9d0c73c9b701
         with:
           mode: stop
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}

.github/workflows/aws_tfhe_wasm_tests.yml

@@ -27,7 +27,7 @@ jobs:
     steps:
       - name: Start instance
         id: start-instance
-        uses: zama-ai/slab-github-runner@58f2cae4bf2c0b6728083f5f009b6dc0eb6dc3ac
+        uses: zama-ai/slab-github-runner@c0e7168795bd78f61f61146951ed9d0c73c9b701
         with:
           mode: start
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}
@@ -45,22 +45,25 @@ jobs:
     runs-on: ${{ needs.setup-instance.outputs.runner-name }}
     steps:
       - name: Checkout tfhe-rs
-        uses: actions/checkout@692973e3d937129bcbf40652eb9f2f61becf3332
+        uses: actions/checkout@d632683dd7b4114ad314bca15554477dd762a938
         with:
           persist-credentials: 'false'
+          token: ${{ secrets.FHE_ACTIONS_TOKEN }}
 
       - name: Set up home
         run: |
           echo "HOME=/home/ubuntu" >> "${GITHUB_ENV}"
 
       - name: Install latest stable
-        uses: dtolnay/rust-toolchain@21dc36fb71dd22e3317045c0c31a3f4249868b17
+        uses: dtolnay/rust-toolchain@7b1c307e0dcbda6122208f10795a713336a9b35a
         with:
           toolchain: stable
 
-      - name: Install Node
+      - name: Install web resources
         run: |
           make install_node
+          make install_chrome_browser
+          make install_chrome_web_driver
 
       - name: Run fmt checks
         run: |
@@ -72,10 +75,10 @@ jobs:
 
       - name: Run parallel wasm tests
         run: |
-          make test_web_js_api_parallel_ci
+          make test_web_js_api_parallel_chrome_ci
 
       - name: Slack Notification
-        if: ${{ always() }}
+        if: ${{ failure() }}
         continue-on-error: true
         uses: rtCamp/action-slack-notify@4e5fb42d249be6a45a298f3c9543b111b02f7907
         env:
@@ -90,7 +93,7 @@ jobs:
     steps:
       - name: Stop instance
         id: stop-instance
-        uses: zama-ai/slab-github-runner@58f2cae4bf2c0b6728083f5f009b6dc0eb6dc3ac
+        uses: zama-ai/slab-github-runner@c0e7168795bd78f61f61146951ed9d0c73c9b701
         with:
           mode: stop
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}

.github/workflows/benchmark_boolean.yml

@@ -0,0 +1,150 @@
+# Run boolean benchmarks on an AWS instance and return parsed results to Slab CI bot.
+name: Boolean benchmarks
+
+on:
+  workflow_dispatch:
+  schedule:
+    # Weekly benchmarks will be triggered each Saturday at 1a.m.
+    - cron: '0 1 * * 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 }}
+
+jobs:
+  setup-instance:
+    name: Setup instance (boolean-benchmarks)
+    runs-on: ubuntu-latest
+    if: github.event_name != 'schedule' ||
+      (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@c0e7168795bd78f61f61146951ed9d0c73c9b701
+        with:
+          mode: start
+          github-token: ${{ secrets.SLAB_ACTION_TOKEN }}
+          slab-url: ${{ secrets.SLAB_BASE_URL }}
+          job-secret: ${{ secrets.JOB_SECRET }}
+          backend: aws
+          profile: bench
+
+  boolean-benchmarks:
+    name: Execute boolean benchmarks in EC2
+    needs: setup-instance
+    runs-on: ${{ needs.setup-instance.outputs.runner-name }}
+    concurrency:
+      group: ${{ github.workflow }}_${{ github.ref }}
+      cancel-in-progress: ${{ github.ref != 'refs/heads/main' }}
+    continue-on-error: true
+    steps:
+      - name: Checkout tfhe-rs repo with tags
+        uses: actions/checkout@d632683dd7b4114ad314bca15554477dd762a938
+        with:
+          fetch-depth: 0
+          token: ${{ secrets.FHE_ACTIONS_TOKEN }}
+
+      - name: Get benchmark details
+        run: |
+          {
+            echo "BENCH_DATE=$(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)";
+          } >> "${GITHUB_ENV}"
+
+      - name: Set up home
+        # "Install rust" step require root user to have a HOME directory which is not set.
+        run: |
+          echo "HOME=/home/ubuntu" >> "${GITHUB_ENV}"
+
+      - name: Install rust
+        uses: dtolnay/rust-toolchain@7b1c307e0dcbda6122208f10795a713336a9b35a
+        with:
+          toolchain: nightly
+
+      - name: Run benchmarks with AVX512
+        run: |
+          make bench_boolean
+
+      - name: Parse results
+        run: |
+          python3 ./ci/benchmark_parser.py target/criterion ${{ env.RESULTS_FILENAME }} \
+          --database tfhe_rs \
+          --hardware "hpc7a.96xlarge" \
+          --project-version "${{ env.COMMIT_HASH }}" \
+          --branch ${{ github.ref_name }} \
+          --commit-date "${{ env.COMMIT_DATE }}" \
+          --bench-date "${{ env.BENCH_DATE }}" \
+          --walk-subdirs \
+          --name-suffix avx512 \
+          --throughput
+
+      - name: Measure key sizes
+        run: |
+          make measure_boolean_key_sizes
+
+      - name: Parse key sizes results
+        run: |
+          python3 ./ci/benchmark_parser.py tfhe/boolean_key_sizes.csv ${{ env.RESULTS_FILENAME }} \
+          --key-sizes \
+          --append-results
+
+      - name: Upload parsed results artifact
+        uses: actions/upload-artifact@50769540e7f4bd5e21e526ee35c689e35e0d6874
+        with:
+          name: ${{ github.sha }}_boolean
+          path: ${{ env.RESULTS_FILENAME }}
+
+      - name: Checkout Slab repo
+        uses: actions/checkout@d632683dd7b4114ad314bca15554477dd762a938
+        with:
+          repository: zama-ai/slab
+          path: slab
+          token: ${{ secrets.FHE_ACTIONS_TOKEN }}
+
+      - name: Send data to Slab
+        shell: bash
+        run: |
+          python3 slab/scripts/data_sender.py ${{ env.RESULTS_FILENAME }} "${{ secrets.JOB_SECRET }}" \
+          --slab-url "${{ secrets.SLAB_URL }}"
+
+      - name: Slack Notification
+        if: ${{ failure() }}
+        continue-on-error: true
+        uses: rtCamp/action-slack-notify@4e5fb42d249be6a45a298f3c9543b111b02f7907
+        env:
+          SLACK_COLOR: ${{ job.status }}
+          SLACK_MESSAGE: "Boolean benchmarks finished with status: ${{ job.status }}. (${{ env.ACTION_RUN_URL }})"
+
+  teardown-instance:
+    name: Teardown instance (boolean-benchmarks)
+    if: ${{ always() && needs.setup-instance.result != 'skipped' }}
+    needs: [ setup-instance, boolean-benchmarks ]
+    runs-on: ubuntu-latest
+    steps:
+      - name: Stop instance
+        id: stop-instance
+        uses: zama-ai/slab-github-runner@c0e7168795bd78f61f61146951ed9d0c73c9b701
+        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@4e5fb42d249be6a45a298f3c9543b111b02f7907
+        env:
+          SLACK_COLOR: ${{ job.status }}
+          SLACK_MESSAGE: "Instance teardown (boolean-benchmarks) finished with status: ${{ job.status }}. (${{ env.ACTION_RUN_URL }})"

.github/workflows/benchmark_core_crypto.yml

@@ -0,0 +1,138 @@
+# Run core crypto benchmarks on an AWS instance and return parsed results to Slab CI bot.
+name: Core crypto 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"
+  SLACK_CHANNEL: ${{ secrets.SLACK_CHANNEL }}
+  SLACK_ICON: https://pbs.twimg.com/profile_images/1274014582265298945/OjBKP9kn_400x400.png
+  SLACK_USERNAME: ${{ secrets.BOT_USERNAME }}
+  SLACK_WEBHOOK: ${{ secrets.SLACK_WEBHOOK }}
+
+jobs:
+  setup-instance:
+    name: Setup instance (core-crypto-benchmarks)
+    runs-on: ubuntu-latest
+    if: github.event_name != 'schedule' ||
+      (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@c0e7168795bd78f61f61146951ed9d0c73c9b701
+        with:
+          mode: start
+          github-token: ${{ secrets.SLAB_ACTION_TOKEN }}
+          slab-url: ${{ secrets.SLAB_BASE_URL }}
+          job-secret: ${{ secrets.JOB_SECRET }}
+          backend: aws
+          profile: bench
+
+  core-crypto-benchmarks:
+    name: Execute core crypto benchmarks in EC2
+    needs: setup-instance
+    runs-on: ${{ needs.setup-instance.outputs.runner-name }}
+    concurrency:
+      group: ${{ github.workflow }}_${{ github.ref }}
+      cancel-in-progress: ${{ github.ref != 'refs/heads/main' }}
+    steps:
+      - name: Checkout tfhe-rs repo with tags
+        uses: actions/checkout@d632683dd7b4114ad314bca15554477dd762a938
+        with:
+          fetch-depth: 0
+          token: ${{ secrets.FHE_ACTIONS_TOKEN }}
+
+      - name: Get benchmark details
+        run: |
+          {
+            echo "BENCH_DATE=$(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)";
+          } >> "${GITHUB_ENV}"
+
+      - name: Set up home
+        # "Install rust" step require root user to have a HOME directory which is not set.
+        run: |
+          echo "HOME=/home/ubuntu" >> "${GITHUB_ENV}"
+
+      - name: Install rust
+        uses: dtolnay/rust-toolchain@7b1c307e0dcbda6122208f10795a713336a9b35a
+        with:
+          toolchain: nightly
+
+      - name: Run benchmarks with AVX512
+        run: |
+          make bench_pbs
+          make bench_pbs128
+          make bench_ks
+
+      - name: Parse results
+        run: |
+          python3 ./ci/benchmark_parser.py target/criterion ${{ env.RESULTS_FILENAME }} \
+          --database tfhe_rs \
+          --hardware "hpc7a.96xlarge" \
+          --project-version "${{ env.COMMIT_HASH }}" \
+          --branch ${{ github.ref_name }} \
+          --commit-date "${{ env.COMMIT_DATE }}" \
+          --bench-date "${{ env.BENCH_DATE }}" \
+          --name-suffix avx512 \
+          --walk-subdirs \
+          --throughput
+
+      - name: Upload parsed results artifact
+        uses: actions/upload-artifact@50769540e7f4bd5e21e526ee35c689e35e0d6874
+        with:
+          name: ${{ github.sha }}_core_crypto
+          path: ${{ env.RESULTS_FILENAME }}
+
+      - name: Checkout Slab repo
+        uses: actions/checkout@d632683dd7b4114ad314bca15554477dd762a938
+        with:
+          repository: zama-ai/slab
+          path: slab
+          token: ${{ secrets.FHE_ACTIONS_TOKEN }}
+
+      - name: Send data to Slab
+        shell: bash
+        run: |
+          python3 slab/scripts/data_sender.py ${{ env.RESULTS_FILENAME }} "${{ secrets.JOB_SECRET }}" \
+          --slab-url "${{ secrets.SLAB_URL }}"
+
+      - name: Slack Notification
+        if: ${{ failure() }}
+        continue-on-error: true
+        uses: rtCamp/action-slack-notify@4e5fb42d249be6a45a298f3c9543b111b02f7907
+        env:
+          SLACK_COLOR: ${{ job.status }}
+          SLACK_MESSAGE: "PBS benchmarks finished with status: ${{ job.status }}. (${{ env.ACTION_RUN_URL }})"
+
+  teardown-instance:
+    name: Teardown instance (core-crypto-benchmarks)
+    if: ${{ always() && needs.setup-instance.result != 'skipped' }}
+    needs: [ setup-instance, core-crypto-benchmarks ]
+    runs-on: ubuntu-latest
+    steps:
+      - name: Stop instance
+        id: stop-instance
+        uses: zama-ai/slab-github-runner@c0e7168795bd78f61f61146951ed9d0c73c9b701
+        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@4e5fb42d249be6a45a298f3c9543b111b02f7907
+        env:
+          SLACK_COLOR: ${{ job.status }}
+          SLACK_MESSAGE: "Instance teardown (core-crypto-benchmarks) finished with status: ${{ job.status }}. (${{ env.ACTION_RUN_URL }})"

.github/workflows/benchmark_gpu_4090.yml

@@ -1,5 +1,5 @@
-# Run all benchmarks on an RTX 4090 machine and return parsed results to Slab CI bot.
-name: TFHE Cuda Backend - 4090 full benchmarks
+# Run benchmarks on an RTX 4090 machine and return parsed results to Slab CI bot.
+name: TFHE Cuda Backend - 4090 benchmarks
 
 env:
   CARGO_TERM_COLOR: always
@@ -11,6 +11,7 @@ env:
   SLACK_ICON: https://pbs.twimg.com/profile_images/1274014582265298945/OjBKP9kn_400x400.png
   SLACK_USERNAME: ${{ secrets.BOT_USERNAME }}
   SLACK_WEBHOOK: ${{ secrets.SLACK_WEBHOOK }}
+  FAST_BENCH: TRUE
 
 on:
   # Allows you to run this workflow manually from the Actions tab as an alternative.
@@ -23,25 +24,25 @@ on:
 
 jobs:
   cuda-integer-benchmarks:
-    name: Cuda integer benchmarks for all operations flavor  (RTX 4090)
-    if: ${{ github.event_name == 'workflow_dispatch' || github.event_name == 'schedule' || contains(github.event.label.name, '4090_bench') }}
+    name: Cuda integer benchmarks (RTX 4090)
+    if: ${{ github.event_name == 'workflow_dispatch' ||
+      github.event_name == 'schedule' && github.repository == 'zama-ai/tfhe-rs' ||
+      contains(github.event.label.name, '4090_bench') }}
     concurrency:
       group: ${{ github.workflow }}_${{ github.ref }}_cuda_integer_bench
-      cancel-in-progress: true
+      cancel-in-progress: ${{ github.ref != 'refs/heads/main' }}
     runs-on: ["self-hosted", "4090-desktop"]
     timeout-minutes: 1440 # 24 hours
     strategy:
       fail-fast: false
       max-parallel: 1
-      matrix:
-        command: [integer, integer_multi_bit]
-        op_flavor: [default, unchecked]
 
     steps:
       - name: Checkout tfhe-rs
-        uses: actions/checkout@692973e3d937129bcbf40652eb9f2f61becf3332
+        uses: actions/checkout@d632683dd7b4114ad314bca15554477dd762a938
         with:
           fetch-depth: 0
+          token: ${{ secrets.FHE_ACTIONS_TOKEN }}
 
       - name: Get benchmark details
         run: |
@@ -50,14 +51,15 @@ jobs:
             echo "COMMIT_DATE=$(git --no-pager show -s --format=%cd --date=iso8601-strict ${{ github.sha }})";
             echo "COMMIT_HASH=$(git describe --tags --dirty)";
           } >> "${GITHUB_ENV}"
+          echo "FAST_BENCH=TRUE" >> "${GITHUB_ENV}"
 
       - name: Install rust
-        uses: dtolnay/rust-toolchain@21dc36fb71dd22e3317045c0c31a3f4249868b17
+        uses: dtolnay/rust-toolchain@7b1c307e0dcbda6122208f10795a713336a9b35a
         with:
           toolchain: nightly
 
       - name: Checkout Slab repo
-        uses: actions/checkout@692973e3d937129bcbf40652eb9f2f61becf3332
+        uses: actions/checkout@d632683dd7b4114ad314bca15554477dd762a938
         with:
           repository: zama-ai/slab
           path: slab
@@ -65,7 +67,7 @@ jobs:
 
       - name: Run integer benchmarks
         run: |
-          make BENCH_OP_FLAVOR=${{ matrix.op_flavor }} bench_${{ matrix.command }}_gpu
+          make BENCH_OP_FLAVOR=default bench_integer_multi_bit_gpu
 
       - name: Parse results
         run: |
@@ -81,27 +83,19 @@ jobs:
           --throughput
 
       - name: Upload parsed results artifact
-        uses: actions/upload-artifact@65462800fd760344b1a7b4382951275a0abb4808
+        uses: actions/upload-artifact@50769540e7f4bd5e21e526ee35c689e35e0d6874
         with:
-          name: ${{ github.sha }}_${{ matrix.command }}_${{ matrix.op_flavor }}
+          name: ${{ github.sha }}_integer_multi_bit_gpu_default
           path: ${{ env.RESULTS_FILENAME }}
 
       - 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 ${{ env.RESULTS_FILENAME }} "${{ secrets.JOB_SECRET }}" \
+          --slab-url "${{ secrets.SLAB_URL }}"
 
       - name: Slack Notification
-        if: ${{ always() }}
+        if: ${{ failure() }}
         continue-on-error: true
         uses: rtCamp/action-slack-notify@4e5fb42d249be6a45a298f3c9543b111b02f7907
         env:
@@ -114,13 +108,13 @@ jobs:
     needs: cuda-integer-benchmarks
     concurrency:
       group: ${{ github.workflow }}_${{ github.ref }}_cuda_core_crypto_bench
-      cancel-in-progress: true
+      cancel-in-progress: ${{ github.ref != 'refs/heads/main' }}
     runs-on: ["self-hosted", "4090-desktop"]
     timeout-minutes: 1440 # 24 hours
 
     steps:
       - name: Checkout tfhe-rs
-        uses: actions/checkout@692973e3d937129bcbf40652eb9f2f61becf3332
+        uses: actions/checkout@d632683dd7b4114ad314bca15554477dd762a938
         with:
           fetch-depth: 0
 
@@ -133,18 +127,18 @@ jobs:
           } >> "${GITHUB_ENV}"
 
       - name: Install rust
-        uses: dtolnay/rust-toolchain@21dc36fb71dd22e3317045c0c31a3f4249868b17
+        uses: dtolnay/rust-toolchain@7b1c307e0dcbda6122208f10795a713336a9b35a
         with:
           toolchain: nightly
 
       - name: Checkout Slab repo
-        uses: actions/checkout@692973e3d937129bcbf40652eb9f2f61becf3332
+        uses: actions/checkout@d632683dd7b4114ad314bca15554477dd762a938
         with:
           repository: zama-ai/slab
           path: slab
           token: ${{ secrets.FHE_ACTIONS_TOKEN }}
 
-      - name: Run integer benchmarks
+      - name: Run core crypto benchmarks
         run: |
           make bench_pbs_gpu
           make bench_ks_gpu
@@ -163,7 +157,7 @@ jobs:
           --throughput
 
       - name: Upload parsed results artifact
-        uses: actions/upload-artifact@65462800fd760344b1a7b4382951275a0abb4808
+        uses: actions/upload-artifact@50769540e7f4bd5e21e526ee35c689e35e0d6874
         with:
           name: ${{ github.sha }}_core_crypto
           path: ${{ env.RESULTS_FILENAME }}
@@ -183,7 +177,7 @@ jobs:
           ${{ secrets.SLAB_URL }}
 
       - name: Slack Notification
-        if: ${{ !success() && !cancelled() }}
+        if: ${{ failure() }}
         continue-on-error: true
         uses: rtCamp/action-slack-notify@4e5fb42d249be6a45a298f3c9543b111b02f7907
         env:

.github/workflows/benchmark_gpu_core_crypto.yml

@@ -27,7 +27,7 @@ jobs:
     steps:
       - name: Start instance
         id: start-instance
-        uses: zama-ai/slab-github-runner@58f2cae4bf2c0b6728083f5f009b6dc0eb6dc3ac
+        uses: zama-ai/slab-github-runner@c0e7168795bd78f61f61146951ed9d0c73c9b701
         with:
           mode: start
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}
@@ -50,7 +50,7 @@ jobs:
             gcc: 11
     env:
       CUDA_PATH: /usr/local/cuda-${{ matrix.cuda }}
-      CMAKE_VERSION: 3.29.1
+      CMAKE_VERSION: 3.29.6
     steps:
       # Mandatory on hyperstack since a bootable volume is not re-usable yet.
       - name: Install dependencies
@@ -65,9 +65,10 @@ jobs:
           sudo make install
 
       - name: Checkout tfhe-rs repo with tags
-        uses: actions/checkout@692973e3d937129bcbf40652eb9f2f61becf3332
+        uses: actions/checkout@d632683dd7b4114ad314bca15554477dd762a938
         with:
           fetch-depth: 0
+          token: ${{ secrets.FHE_ACTIONS_TOKEN }}
 
       - name: Get benchmark details
         run: |
@@ -83,7 +84,7 @@ jobs:
           echo "HOME=/home/ubuntu" >> "${GITHUB_ENV}"
 
       - name: Install rust
-        uses: dtolnay/rust-toolchain@21dc36fb71dd22e3317045c0c31a3f4249868b17
+        uses: dtolnay/rust-toolchain@7b1c307e0dcbda6122208f10795a713336a9b35a
         with:
           toolchain: nightly
 
@@ -128,13 +129,13 @@ jobs:
           --throughput
 
       - name: Upload parsed results artifact
-        uses: actions/upload-artifact@65462800fd760344b1a7b4382951275a0abb4808
+        uses: actions/upload-artifact@50769540e7f4bd5e21e526ee35c689e35e0d6874
         with:
           name: ${{ github.sha }}_core_crypto
           path: ${{ env.RESULTS_FILENAME }}
 
       - name: Checkout Slab repo
-        uses: actions/checkout@692973e3d937129bcbf40652eb9f2f61becf3332
+        uses: actions/checkout@d632683dd7b4114ad314bca15554477dd762a938
         with:
           repository: zama-ai/slab
           path: slab
@@ -143,16 +144,8 @@ 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 ${{ env.RESULTS_FILENAME }} "${{ secrets.JOB_SECRET }}" \
+          --slab-url "${{ secrets.SLAB_URL }}"
 
   slack-notify:
     name: Slack Notification
@@ -175,7 +168,7 @@ jobs:
     steps:
       - name: Stop instance
         id: stop-instance
-        uses: zama-ai/slab-github-runner@58f2cae4bf2c0b6728083f5f009b6dc0eb6dc3ac
+        uses: zama-ai/slab-github-runner@c0e7168795bd78f61f61146951ed9d0c73c9b701
         with:
           mode: stop
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}

.github/workflows/benchmark_gpu_integer.yml

@@ -23,14 +23,14 @@ jobs:
   setup-instance:
     name: Setup instance (cuda-integer-benchmarks)
     runs-on: ubuntu-latest
-    if:  github.event_name != 'push' ||
+    if:  github.event_name == 'workflow_dispatch' ||
       (github.event_name == 'push' && 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@58f2cae4bf2c0b6728083f5f009b6dc0eb6dc3ac
+        uses: zama-ai/slab-github-runner@c0e7168795bd78f61f61146951ed9d0c73c9b701
         with:
           mode: start
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}
@@ -53,7 +53,7 @@ jobs:
             gcc: 11
     env:
       CUDA_PATH: /usr/local/cuda-${{ matrix.cuda }}
-      CMAKE_VERSION: 3.29.1
+      CMAKE_VERSION: 3.29.6
     steps:
       # Mandatory on hyperstack since a bootable volume is not re-usable yet.
       - name: Install dependencies
@@ -68,9 +68,10 @@ jobs:
           sudo make install
 
       - name: Checkout tfhe-rs repo with tags
-        uses: actions/checkout@692973e3d937129bcbf40652eb9f2f61becf3332
+        uses: actions/checkout@d632683dd7b4114ad314bca15554477dd762a938
         with:
           fetch-depth: 0
+          token: ${{ secrets.FHE_ACTIONS_TOKEN }}
 
       - name: Get benchmark details
         run: |
@@ -86,7 +87,7 @@ jobs:
           echo "HOME=/home/ubuntu" >> "${GITHUB_ENV}"
 
       - name: Install rust
-        uses: dtolnay/rust-toolchain@21dc36fb71dd22e3317045c0c31a3f4249868b17
+        uses: dtolnay/rust-toolchain@7b1c307e0dcbda6122208f10795a713336a9b35a
         with:
           toolchain: nightly
 
@@ -110,6 +111,10 @@ jobs:
             echo "CUDAHOSTCXX=/usr/bin/g++-${{ matrix.gcc }}";
           } >> "${GITHUB_ENV}"
 
+      - name: Check device is detected
+        if: ${{ !cancelled() }}
+        run: nvidia-smi
+
       - name: Run benchmarks with AVX512
         run: |
           make FAST_BENCH=TRUE BENCH_OP_FLAVOR=default bench_integer_gpu
@@ -120,7 +125,7 @@ jobs:
             parse_integer_benches
 
       - name: Upload csv results artifact
-        uses: actions/upload-artifact@65462800fd760344b1a7b4382951275a0abb4808
+        uses: actions/upload-artifact@50769540e7f4bd5e21e526ee35c689e35e0d6874
         with:
           name: ${{ github.sha }}_csv_integer
           path: ${{ env.PARSE_INTEGER_BENCH_CSV_FILE }}
@@ -140,13 +145,13 @@ jobs:
           --throughput
 
       - name: Upload parsed results artifact
-        uses: actions/upload-artifact@65462800fd760344b1a7b4382951275a0abb4808
+        uses: actions/upload-artifact@50769540e7f4bd5e21e526ee35c689e35e0d6874
         with:
           name: ${{ github.sha }}_integer
           path: ${{ env.RESULTS_FILENAME }}
 
       - name: Checkout Slab repo
-        uses: actions/checkout@692973e3d937129bcbf40652eb9f2f61becf3332
+        uses: actions/checkout@d632683dd7b4114ad314bca15554477dd762a938
         with:
           repository: zama-ai/slab
           path: slab
@@ -155,22 +160,14 @@ 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 ${{ env.RESULTS_FILENAME }} "${{ secrets.JOB_SECRET }}" \
+          --slab-url "${{ secrets.SLAB_URL }}"
 
   slack-notify:
     name: Slack Notification
     needs: [ setup-instance, cuda-integer-benchmarks ]
     runs-on: ubuntu-latest
-    if: ${{ !success() && !cancelled() }}
+    if: ${{ always() && needs.cuda-integer-benchmarks.result != 'skipped' && failure() }}
     continue-on-error: true
     steps:
       - name: Send message
@@ -187,7 +184,7 @@ jobs:
     steps:
       - name: Stop instance
         id: stop-instance
-        uses: zama-ai/slab-github-runner@58f2cae4bf2c0b6728083f5f009b6dc0eb6dc3ac
+        uses: zama-ai/slab-github-runner@c0e7168795bd78f61f61146951ed9d0c73c9b701
         with:
           mode: stop
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}

.github/workflows/benchmark_gpu_integer_2H100_full.yml

@@ -0,0 +1,194 @@
+# Run integer benchmarks on an instance with CUDA and return parsed results to Slab CI bot.
+name: Integer 2xH100 benchmarks
+
+on:
+  workflow_dispatch:
+  schedule:
+    # Weekly benchmarks will be triggered each Saturday at 1a.m.
+    - cron: '0 1 * * 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 }}
+
+jobs:
+  setup-instance:
+    name: Setup instance (cuda-integer-full-2-gpu-benchmarks)
+    runs-on: ubuntu-latest
+    if: github.event_name != 'schedule' ||
+      (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@c0e7168795bd78f61f61146951ed9d0c73c9b701
+        with:
+          mode: start
+          github-token: ${{ secrets.SLAB_ACTION_TOKEN }}
+          slab-url: ${{ secrets.SLAB_BASE_URL }}
+          job-secret: ${{ secrets.JOB_SECRET }}
+          backend: hyperstack
+          profile: 2-h100
+
+  cuda-integer-full-2-gpu-benchmarks:
+    name: Execute 2xH100 integer benchmarks
+    needs: setup-instance
+    runs-on: ${{ needs.setup-instance.outputs.runner-name }}
+    timeout-minutes: 1440 # 24 hours
+    continue-on-error: true
+    strategy:
+      fail-fast: false
+      max-parallel: 1
+      matrix:
+        command: [integer_multi_bit]
+        op_flavor: [default]
+        # explicit include-based build matrix, of known valid options
+        include:
+          - os: ubuntu-22.04
+            cuda: "12.2"
+            gcc: 11
+    env:
+      CUDA_PATH: /usr/local/cuda-${{ matrix.cuda }}
+      CMAKE_VERSION: 3.29.6
+    steps:
+      # Mandatory on hyperstack since a bootable volume is not re-usable yet.
+      - name: Install dependencies
+        run: |
+          sudo apt update
+          sudo apt install -y checkinstall zlib1g-dev libssl-dev
+          wget https://github.com/Kitware/CMake/releases/download/v${{ env.CMAKE_VERSION }}/cmake-${{ env.CMAKE_VERSION }}.tar.gz
+          tar -zxvf cmake-${{ env.CMAKE_VERSION }}.tar.gz
+          cd cmake-${{ env.CMAKE_VERSION }}
+          ./bootstrap
+          make -j"$(nproc)"
+          sudo make install
+
+      - name: Checkout tfhe-rs repo with tags
+        uses: actions/checkout@d632683dd7b4114ad314bca15554477dd762a938
+        with:
+          fetch-depth: 0
+          token: ${{ secrets.FHE_ACTIONS_TOKEN }}
+
+      - name: Get benchmark details
+        run: |
+          {
+            echo "BENCH_DATE=$(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)";
+          } >> "${GITHUB_ENV}"
+
+      - name: Set up home
+        # "Install rust" step require root user to have a HOME directory which is not set.
+        run: |
+          echo "HOME=/home/ubuntu" >> "${GITHUB_ENV}"
+
+      - name: Install rust
+        uses: dtolnay/rust-toolchain@7b1c307e0dcbda6122208f10795a713336a9b35a
+        with:
+          toolchain: nightly
+
+      - name: Export CUDA variables
+        if: ${{ !cancelled() }}
+        run: |
+          {
+            echo "CUDA_PATH=$CUDA_PATH";
+            echo "LD_LIBRARY_PATH=$CUDA_PATH/lib:$LD_LIBRARY_PATH";
+            echo "CUDACXX=/usr/local/cuda-${{ matrix.cuda }}/bin/nvcc";
+          } >> "${GITHUB_ENV}"
+          echo "$CUDA_PATH/bin" >> "${GITHUB_PATH}"
+
+      # Specify the correct host compilers
+      - name: Export gcc and g++ variables
+        if: ${{ !cancelled() }}
+        run: |
+          {
+            echo "CC=/usr/bin/gcc-${{ matrix.gcc }}";
+            echo "CXX=/usr/bin/g++-${{ matrix.gcc }}";
+            echo "CUDAHOSTCXX=/usr/bin/g++-${{ matrix.gcc }}";
+          } >> "${GITHUB_ENV}"
+
+      - name: Checkout Slab repo
+        uses: actions/checkout@d632683dd7b4114ad314bca15554477dd762a938
+        with:
+          repository: zama-ai/slab
+          path: slab
+          token: ${{ secrets.FHE_ACTIONS_TOKEN }}
+
+      - name: Check device is detected
+        if: ${{ !cancelled() }}
+        run: nvidia-smi
+
+      - name: Run benchmarks with AVX512
+        run: |
+          make BENCH_OP_FLAVOR=${{ matrix.op_flavor }} bench_${{ matrix.command }}_gpu
+
+      - name: Parse results
+        run: |
+          python3 ./ci/benchmark_parser.py target/criterion ${{ env.RESULTS_FILENAME }} \
+          --database tfhe_rs \
+          --hardware "n3-H100x2" \
+          --backend gpu \
+          --project-version "${{ env.COMMIT_HASH }}" \
+          --branch ${{ github.ref_name }} \
+          --commit-date "${{ env.COMMIT_DATE }}" \
+          --bench-date "${{ env.BENCH_DATE }}" \
+          --walk-subdirs \
+          --name-suffix avx512 \
+          --throughput
+
+      - name: Upload parsed results artifact
+        uses: actions/upload-artifact@50769540e7f4bd5e21e526ee35c689e35e0d6874
+        with:
+          name: ${{ github.sha }}_${{ matrix.command }}_${{ matrix.op_flavor }}
+          path: ${{ env.RESULTS_FILENAME }}
+
+      - name: Send data to Slab
+        shell: bash
+        run: |
+          python3 slab/scripts/data_sender.py ${{ env.RESULTS_FILENAME }} "${{ secrets.JOB_SECRET }}" \
+          --slab-url "${{ secrets.SLAB_URL }}"
+
+  slack-notify:
+    name: Slack Notification
+    needs: [ setup-instance, cuda-integer-full-2-gpu-benchmarks ]
+    runs-on: ubuntu-latest
+    if: ${{ always() && needs.cuda-integer-full-2-gpu-benchmarks.result != 'skipped' && failure() }}
+    continue-on-error: true
+    steps:
+      - name: Send message
+        uses: rtCamp/action-slack-notify@4e5fb42d249be6a45a298f3c9543b111b02f7907
+        env:
+          SLACK_COLOR: ${{ needs.cuda-integer-full-2-gpu-benchmarks.result }}
+          SLACK_MESSAGE: "Integer GPU 2xH100 benchmarks finished with status: ${{ needs.cuda-integer-full-2-gpu-benchmarks.result }}. (${{ env.ACTION_RUN_URL }})"
+
+  teardown-instance:
+    name: Teardown instance (cuda-integer-full-2-gpu-benchmarks)
+    if: ${{ always() && needs.setup-instance.result != 'skipped' }}
+    needs: [ setup-instance, cuda-integer-full-2-gpu-benchmarks ]
+    runs-on: ubuntu-latest
+    steps:
+      - name: Stop instance
+        id: stop-instance
+        uses: zama-ai/slab-github-runner@c0e7168795bd78f61f61146951ed9d0c73c9b701
+        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@4e5fb42d249be6a45a298f3c9543b111b02f7907
+        env:
+          SLACK_COLOR: ${{ job.status }}
+          SLACK_MESSAGE: "Instance teardown (cuda-integer-full-2-gpu-benchmarks) finished with status: ${{ job.status }}. (${{ env.ACTION_RUN_URL }})"

.github/workflows/benchmark_gpu_integer_full.yml

@@ -29,7 +29,7 @@ jobs:
     steps:
       - name: Start instance
         id: start-instance
-        uses: zama-ai/slab-github-runner@58f2cae4bf2c0b6728083f5f009b6dc0eb6dc3ac
+        uses: zama-ai/slab-github-runner@c0e7168795bd78f61f61146951ed9d0c73c9b701
         with:
           mode: start
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}
@@ -49,7 +49,7 @@ jobs:
       max-parallel: 1
       matrix:
         command: [integer, integer_multi_bit]
-        op_flavor: [default, unchecked]
+        op_flavor: [default]
         # explicit include-based build matrix, of known valid options
         include:
           - os: ubuntu-22.04
@@ -57,7 +57,7 @@ jobs:
             gcc: 11
     env:
       CUDA_PATH: /usr/local/cuda-${{ matrix.cuda }}
-      CMAKE_VERSION: 3.29.1
+      CMAKE_VERSION: 3.29.6
     steps:
       # Mandatory on hyperstack since a bootable volume is not re-usable yet.
       - name: Install dependencies
@@ -72,9 +72,10 @@ jobs:
           sudo make install
 
       - name: Checkout tfhe-rs repo with tags
-        uses: actions/checkout@692973e3d937129bcbf40652eb9f2f61becf3332
+        uses: actions/checkout@d632683dd7b4114ad314bca15554477dd762a938
         with:
           fetch-depth: 0
+          token: ${{ secrets.FHE_ACTIONS_TOKEN }}
 
       - name: Get benchmark details
         run: |
@@ -90,7 +91,7 @@ jobs:
           echo "HOME=/home/ubuntu" >> "${GITHUB_ENV}"
 
       - name: Install rust
-        uses: dtolnay/rust-toolchain@21dc36fb71dd22e3317045c0c31a3f4249868b17
+        uses: dtolnay/rust-toolchain@7b1c307e0dcbda6122208f10795a713336a9b35a
         with:
           toolchain: nightly
 
@@ -115,16 +116,26 @@ jobs:
           } >> "${GITHUB_ENV}"
 
       - name: Checkout Slab repo
-        uses: actions/checkout@692973e3d937129bcbf40652eb9f2f61becf3332
+        uses: actions/checkout@d632683dd7b4114ad314bca15554477dd762a938
         with:
           repository: zama-ai/slab
           path: slab
           token: ${{ secrets.FHE_ACTIONS_TOKEN }}
 
+      - name: Check device is detected
+        if: ${{ !cancelled() }}
+        run: nvidia-smi
+
       - name: Run benchmarks with AVX512
         run: |
           make BENCH_OP_FLAVOR=${{ matrix.op_flavor }} bench_${{ matrix.command }}_gpu
 
+      # Run these benchmarks only once
+      - name: Run compression benchmarks with AVX512
+        if: matrix.op_flavor == 'default' && matrix.command == 'integer'
+        run: |
+          make bench_integer_compression_gpu
+
       - name: Parse results
         run: |
           python3 ./ci/benchmark_parser.py target/criterion ${{ env.RESULTS_FILENAME }} \
@@ -140,7 +151,7 @@ jobs:
           --throughput
 
       - name: Upload parsed results artifact
-        uses: actions/upload-artifact@65462800fd760344b1a7b4382951275a0abb4808
+        uses: actions/upload-artifact@50769540e7f4bd5e21e526ee35c689e35e0d6874
         with:
           name: ${{ github.sha }}_${{ matrix.command }}_${{ matrix.op_flavor }}
           path: ${{ env.RESULTS_FILENAME }}
@@ -148,22 +159,14 @@ jobs:
       - name: Send data to Slab
         shell: bash
         run: |
-          echo "Computing HMac on results file"
-          SIGNATURE="$(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 ${{ env.RESULTS_FILENAME }} "${{ secrets.JOB_SECRET }}" \
+          --slab-url "${{ secrets.SLAB_URL }}"
 
   slack-notify:
     name: Slack Notification
     needs: [ setup-instance, cuda-integer-full-benchmarks ]
     runs-on: ubuntu-latest
-    if: ${{ !success() && !cancelled() }}
+    if: ${{ always() && needs.cuda-integer-full-benchmarks.result != 'skipped' && failure() }}
     continue-on-error: true
     steps:
       - name: Send message
@@ -180,7 +183,7 @@ jobs:
     steps:
       - name: Stop instance
         id: stop-instance
-        uses: zama-ai/slab-github-runner@58f2cae4bf2c0b6728083f5f009b6dc0eb6dc3ac
+        uses: zama-ai/slab-github-runner@c0e7168795bd78f61f61146951ed9d0c73c9b701
         with:
           mode: stop
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}

.github/workflows/benchmark_gpu_integer_multi_bit.yml

@@ -3,6 +3,16 @@ name: Integer GPU Multi-bit benchmarks
 
 on:
   workflow_dispatch:
+    inputs:
+      all_precisions:
+        description: "Run all precisions"
+        type: boolean
+        default: false
+      fast_default:
+        description: "Run only deduplicated default operations without scalar variants"
+        type: boolean
+        default: false
+
   schedule:
     # Weekly benchmarks will be triggered each Saturday at 1a.m.
     - cron: '0 1 * * 6'
@@ -18,6 +28,8 @@ env:
   SLACK_ICON: https://pbs.twimg.com/profile_images/1274014582265298945/OjBKP9kn_400x400.png
   SLACK_USERNAME: ${{ secrets.BOT_USERNAME }}
   SLACK_WEBHOOK: ${{ secrets.SLACK_WEBHOOK }}
+  FAST_BENCH: TRUE
+  BENCH_OP_FLAVOR: default
 
 jobs:
   setup-instance:
@@ -30,7 +42,7 @@ jobs:
     steps:
       - name: Start instance
         id: start-instance
-        uses: zama-ai/slab-github-runner@58f2cae4bf2c0b6728083f5f009b6dc0eb6dc3ac
+        uses: zama-ai/slab-github-runner@c0e7168795bd78f61f61146951ed9d0c73c9b701
         with:
           mode: start
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}
@@ -54,7 +66,7 @@ jobs:
             gcc: 11
     env:
       CUDA_PATH: /usr/local/cuda-${{ matrix.cuda }}
-      CMAKE_VERSION: 3.29.1
+      CMAKE_VERSION: 3.29.6
     steps:
       # Mandatory on hyperstack since a bootable volume is not re-usable yet.
       - name: Install dependencies
@@ -69,9 +81,10 @@ jobs:
           sudo make install
 
       - name: Checkout tfhe-rs repo with tags
-        uses: actions/checkout@692973e3d937129bcbf40652eb9f2f61becf3332
+        uses: actions/checkout@d632683dd7b4114ad314bca15554477dd762a938
         with:
           fetch-depth: 0
+          token: ${{ secrets.FHE_ACTIONS_TOKEN }}
 
       - name: Get benchmark details
         run: |
@@ -87,7 +100,7 @@ jobs:
           echo "HOME=/home/ubuntu" >> "${GITHUB_ENV}"
 
       - name: Install rust
-        uses: dtolnay/rust-toolchain@21dc36fb71dd22e3317045c0c31a3f4249868b17
+        uses: dtolnay/rust-toolchain@7b1c307e0dcbda6122208f10795a713336a9b35a
         with:
           toolchain: nightly
 
@@ -111,9 +124,23 @@ jobs:
             echo "CUDAHOSTCXX=/usr/bin/g++-${{ matrix.gcc }}";
           } >> "${GITHUB_ENV}"
 
+      - name: Should run benchmarks with all precisions
+        if: inputs.all_precisions
+        run: |
+          echo "FAST_BENCH=FALSE" >> "${GITHUB_ENV}"
+
+      - name: Should run fast subset benchmarks
+        if: inputs.fast_default
+        run: |
+          echo "BENCH_OP_FLAVOR=fast_default" >> "${GITHUB_ENV}"
+
+      - name: Check device is detected
+        if: ${{ !cancelled() }}
+        run: nvidia-smi
+
       - name: Run multi-bit benchmarks with AVX512
         run: |
-          make FAST_BENCH=TRUE BENCH_OP_FLAVOR=default bench_integer_multi_bit_gpu
+          make bench_unsigned_integer_multi_bit_gpu
 
       - name: Parse benchmarks to csv
         run: |
@@ -121,7 +148,7 @@ jobs:
             parse_integer_benches
 
       - name: Upload csv results artifact
-        uses: actions/upload-artifact@65462800fd760344b1a7b4382951275a0abb4808
+        uses: actions/upload-artifact@50769540e7f4bd5e21e526ee35c689e35e0d6874
         with:
           name: ${{ github.sha }}_csv_integer
           path: ${{ env.PARSE_INTEGER_BENCH_CSV_FILE }}
@@ -141,13 +168,13 @@ jobs:
           --throughput
 
       - name: Upload parsed results artifact
-        uses: actions/upload-artifact@65462800fd760344b1a7b4382951275a0abb4808
+        uses: actions/upload-artifact@50769540e7f4bd5e21e526ee35c689e35e0d6874
         with:
           name: ${{ github.sha }}_integer
           path: ${{ env.RESULTS_FILENAME }}
 
       - name: Checkout Slab repo
-        uses: actions/checkout@692973e3d937129bcbf40652eb9f2f61becf3332
+        uses: actions/checkout@d632683dd7b4114ad314bca15554477dd762a938
         with:
           repository: zama-ai/slab
           path: slab
@@ -156,23 +183,14 @@ 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 ${{ env.RESULTS_FILENAME }} "${{ secrets.JOB_SECRET }}" \
+          --slab-url "${{ secrets.SLAB_URL }}"
 
   slack-notify:
     name: Slack Notification
     needs: [ setup-instance, cuda-integer-multi-bit-benchmarks ]
     runs-on: ubuntu-latest
-    if: ${{ !success() && !cancelled() }}
+    if: ${{ always() && needs.cuda-integer-multi-bit-benchmarks.result != 'skipped' && failure() }}
     continue-on-error: true
     steps:
       - name: Send message
@@ -189,7 +207,7 @@ jobs:
     steps:
       - name: Stop instance
         id: stop-instance
-        uses: zama-ai/slab-github-runner@58f2cae4bf2c0b6728083f5f009b6dc0eb6dc3ac
+        uses: zama-ai/slab-github-runner@c0e7168795bd78f61f61146951ed9d0c73c9b701
         with:
           mode: stop
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}

.github/workflows/benchmark_gpu_integer_multi_bit_multi_gpu.yml

@@ -3,6 +3,16 @@ name: Integer multi GPU Multi-bit benchmarks
 
 on:
   workflow_dispatch:
+    inputs:
+      all_precisions:
+        description: "Run all precisions"
+        type: boolean
+        default: false
+      fast_default:
+        description: "Run only deduplicated default operations without scalar variants"
+        type: boolean
+        default: false
+
   schedule:
     # Weekly benchmarks will be triggered each Saturday at 1a.m.
     - cron: '0 1 * * 6'
@@ -17,25 +27,29 @@ env:
   SLACK_ICON: https://pbs.twimg.com/profile_images/1274014582265298945/OjBKP9kn_400x400.png
   SLACK_USERNAME: ${{ secrets.BOT_USERNAME }}
   SLACK_WEBHOOK: ${{ secrets.SLACK_WEBHOOK }}
+  FAST_BENCH: TRUE
+  BENCH_OP_FLAVOR: default
 
 jobs:
   setup-instance:
     name: Setup instance (cuda-integer-multi-bit-multi-gpu-benchmarks)
     runs-on: ubuntu-latest
-    if: ${{ (github.event_name == 'push' && github.repository == 'zama-ai/tfhe-rs') || github.event_name == 'workflow_dispatch' }}
+    if: ${{ (github.event_name == 'push' && github.repository == 'zama-ai/tfhe-rs') ||
+      (github.event_name == 'schedule' && github.repository == 'zama-ai/tfhe-rs') ||
+      github.event_name == 'workflow_dispatch' }}
     outputs:
       runner-name: ${{ steps.start-instance.outputs.label }}
     steps:
       - name: Start instance
         id: start-instance
-        uses: zama-ai/slab-github-runner@58f2cae4bf2c0b6728083f5f009b6dc0eb6dc3ac
+        uses: zama-ai/slab-github-runner@c0e7168795bd78f61f61146951ed9d0c73c9b701
         with:
           mode: start
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}
           slab-url: ${{ secrets.SLAB_BASE_URL }}
           job-secret: ${{ secrets.JOB_SECRET }}
-          backend: aws
-          profile: multi-gpu-test
+          backend: hyperstack
+          profile: multi-h100
 
   cuda-integer-multi-bit-multi-gpu-benchmarks:
     name: Execute multi GPU integer multi-bit benchmarks
@@ -50,15 +64,28 @@ jobs:
         include:
           - os: ubuntu-22.04
             cuda: "12.2"
-            gcc: 9
+            gcc: 11
     env:
       CUDA_PATH: /usr/local/cuda-${{ matrix.cuda }}
-    
+      CMAKE_VERSION: 3.29.6
     steps:
+      # Mandatory on hyperstack since a bootable volume is not re-usable yet.
+      - name: Install dependencies
+        run: |
+          sudo apt update
+          sudo apt install -y checkinstall zlib1g-dev libssl-dev
+          wget https://github.com/Kitware/CMake/releases/download/v${{ env.CMAKE_VERSION }}/cmake-${{ env.CMAKE_VERSION }}.tar.gz
+          tar -zxvf cmake-${{ env.CMAKE_VERSION }}.tar.gz
+          cd cmake-${{ env.CMAKE_VERSION }}
+          ./bootstrap
+          make -j"$(nproc)"
+          sudo make install
+
       - name: Checkout tfhe-rs repo with tags
-        uses: actions/checkout@692973e3d937129bcbf40652eb9f2f61becf3332
+        uses: actions/checkout@d632683dd7b4114ad314bca15554477dd762a938
         with:
           fetch-depth: 0
+          token: ${{ secrets.FHE_ACTIONS_TOKEN }}
 
       - name: Get benchmark details
         run: |
@@ -74,7 +101,7 @@ jobs:
           echo "HOME=/home/ubuntu" >> "${GITHUB_ENV}"
 
       - name: Install rust
-        uses: dtolnay/rust-toolchain@21dc36fb71dd22e3317045c0c31a3f4249868b17
+        uses: dtolnay/rust-toolchain@7b1c307e0dcbda6122208f10795a713336a9b35a
         with:
           toolchain: nightly
 
@@ -99,21 +126,35 @@ jobs:
           } >> "${GITHUB_ENV}"
 
       - name: Checkout Slab repo
-        uses: actions/checkout@692973e3d937129bcbf40652eb9f2f61becf3332
+        uses: actions/checkout@d632683dd7b4114ad314bca15554477dd762a938
         with:
           repository: zama-ai/slab
           path: slab
           token: ${{ secrets.FHE_ACTIONS_TOKEN }}
 
+      - name: Should run benchmarks with all precisions
+        if: inputs.all_precisions
+        run: |
+          echo "FAST_BENCH=FALSE" >> "${GITHUB_ENV}"
+
+      - name: Should run fast subset benchmarks
+        if: inputs.fast_default
+        run: |
+          echo "BENCH_OP_FLAVOR=fast_default" >> "${GITHUB_ENV}"
+
+      - name: Check device is detected
+        if: ${{ !cancelled() }}
+        run: nvidia-smi
+
       - name: Run multi-bit benchmarks with AVX512
         run: |
-          make FAST_BENCH=TRUE BENCH_OP_FLAVOR=default bench_integer_multi_bit_gpu
+          make bench_unsigned_integer_multi_bit_gpu
 
       - name: Parse results
         run: |
           python3 ./ci/benchmark_parser.py target/criterion ${{ env.RESULTS_FILENAME }} \
           --database tfhe_rs \
-          --hardware "p3.8xlarge" \
+          --hardware "n3-H100x8" \
           --backend gpu \
           --project-version "${{ env.COMMIT_HASH }}" \
           --branch ${{ github.ref_name }} \
@@ -124,7 +165,7 @@ jobs:
           --throughput
 
       - name: Upload parsed results artifact
-        uses: actions/upload-artifact@65462800fd760344b1a7b4382951275a0abb4808
+        uses: actions/upload-artifact@50769540e7f4bd5e21e526ee35c689e35e0d6874
         with:
           name: ${{ github.sha }}_integer
           path: ${{ env.RESULTS_FILENAME }}
@@ -132,22 +173,14 @@ jobs:
       - name: Send data to Slab
         shell: bash
         run: |
-          echo "Computing HMac on results file"
-          SIGNATURE="$(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 ${{ env.RESULTS_FILENAME }} "${{ secrets.JOB_SECRET }}" \
+          --slab-url "${{ secrets.SLAB_URL }}"
 
   slack-notify:
     name: Slack Notification
     needs: [ setup-instance, cuda-integer-multi-bit-multi-gpu-benchmarks ]
     runs-on: ubuntu-latest
-    if: ${{ !success() && !cancelled() }}
+    if: ${{ always() && needs.cuda-integer-multi-bit-multi-gpu-benchmarks.result != 'skipped' && failure() }}
     continue-on-error: true
     steps:
       - name: Send message
@@ -164,7 +197,7 @@ jobs:
     steps:
       - name: Stop instance
         id: stop-instance
-        uses: zama-ai/slab-github-runner@58f2cae4bf2c0b6728083f5f009b6dc0eb6dc3ac
+        uses: zama-ai/slab-github-runner@c0e7168795bd78f61f61146951ed9d0c73c9b701
         with:
           mode: stop
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}

.github/workflows/benchmark_gpu_integer_multi_gpu_full.yml

@@ -29,17 +29,17 @@ jobs:
     steps:
       - name: Start instance
         id: start-instance
-        uses: zama-ai/slab-github-runner@58f2cae4bf2c0b6728083f5f009b6dc0eb6dc3ac
+        uses: zama-ai/slab-github-runner@c0e7168795bd78f61f61146951ed9d0c73c9b701
         with:
           mode: start
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}
           slab-url: ${{ secrets.SLAB_BASE_URL }}
           job-secret: ${{ secrets.JOB_SECRET }}
-          backend: aws
-          profile: multi-gpu-test
+          backend: hyperstack
+          profile: multi-h100
 
   cuda-integer-full-multi-gpu-benchmarks:
-    name: Execute multi GPU integer benchmarks for all operations flavor
+    name: Execute multi GPU integer benchmarks
     needs: setup-instance
     runs-on: ${{ needs.setup-instance.outputs.runner-name }}
     timeout-minutes: 1440 # 24 hours
@@ -48,21 +48,34 @@ jobs:
       fail-fast: false
       max-parallel: 1
       matrix:
-        command: [integer, integer_multi_bit]
-        op_flavor: [default, unchecked]
+        command: [integer_multi_bit]
+        op_flavor: [default]
         # explicit include-based build matrix, of known valid options
         include:
           - os: ubuntu-22.04
             cuda: "12.2"
-            gcc: 9
+            gcc: 11
     env:
       CUDA_PATH: /usr/local/cuda-${{ matrix.cuda }}
-    
+      CMAKE_VERSION: 3.29.6
     steps:
+      # Mandatory on hyperstack since a bootable volume is not re-usable yet.
+      - name: Install dependencies
+        run: |
+          sudo apt update
+          sudo apt install -y checkinstall zlib1g-dev libssl-dev
+          wget https://github.com/Kitware/CMake/releases/download/v${{ env.CMAKE_VERSION }}/cmake-${{ env.CMAKE_VERSION }}.tar.gz
+          tar -zxvf cmake-${{ env.CMAKE_VERSION }}.tar.gz
+          cd cmake-${{ env.CMAKE_VERSION }}
+          ./bootstrap
+          make -j"$(nproc)"
+          sudo make install
+
       - name: Checkout tfhe-rs repo with tags
-        uses: actions/checkout@692973e3d937129bcbf40652eb9f2f61becf3332
+        uses: actions/checkout@d632683dd7b4114ad314bca15554477dd762a938
         with:
           fetch-depth: 0
+          token: ${{ secrets.FHE_ACTIONS_TOKEN }}
 
       - name: Get benchmark details
         run: |
@@ -78,7 +91,7 @@ jobs:
           echo "HOME=/home/ubuntu" >> "${GITHUB_ENV}"
 
       - name: Install rust
-        uses: dtolnay/rust-toolchain@21dc36fb71dd22e3317045c0c31a3f4249868b17
+        uses: dtolnay/rust-toolchain@7b1c307e0dcbda6122208f10795a713336a9b35a
         with:
           toolchain: nightly
 
@@ -103,12 +116,16 @@ jobs:
           } >> "${GITHUB_ENV}"
 
       - name: Checkout Slab repo
-        uses: actions/checkout@692973e3d937129bcbf40652eb9f2f61becf3332
+        uses: actions/checkout@d632683dd7b4114ad314bca15554477dd762a938
         with:
           repository: zama-ai/slab
           path: slab
           token: ${{ secrets.FHE_ACTIONS_TOKEN }}
 
+      - name: Check device is detected
+        if: ${{ !cancelled() }}
+        run: nvidia-smi
+
       - name: Run benchmarks with AVX512
         run: |
           make BENCH_OP_FLAVOR=${{ matrix.op_flavor }} bench_${{ matrix.command }}_gpu
@@ -117,7 +134,7 @@ jobs:
         run: |
           python3 ./ci/benchmark_parser.py target/criterion ${{ env.RESULTS_FILENAME }} \
           --database tfhe_rs \
-          --hardware "p3.8xlarge" \
+          --hardware "n3-H100x8" \
           --backend gpu \
           --project-version "${{ env.COMMIT_HASH }}" \
           --branch ${{ github.ref_name }} \
@@ -128,7 +145,7 @@ jobs:
           --throughput
 
       - name: Upload parsed results artifact
-        uses: actions/upload-artifact@65462800fd760344b1a7b4382951275a0abb4808
+        uses: actions/upload-artifact@50769540e7f4bd5e21e526ee35c689e35e0d6874
         with:
           name: ${{ github.sha }}_${{ matrix.command }}_${{ matrix.op_flavor }}
           path: ${{ env.RESULTS_FILENAME }}
@@ -136,22 +153,14 @@ jobs:
       - name: Send data to Slab
         shell: bash
         run: |
-          echo "Computing HMac on results file"
-          SIGNATURE="$(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 ${{ env.RESULTS_FILENAME }} "${{ secrets.JOB_SECRET }}" \
+          --slab-url "${{ secrets.SLAB_URL }}"
 
   slack-notify:
     name: Slack Notification
     needs: [ setup-instance, cuda-integer-full-multi-gpu-benchmarks ]
     runs-on: ubuntu-latest
-    if: ${{ !success() && !cancelled() }}
+    if: ${{ always() && needs.cuda-integer-full-multi-gpu-benchmarks.result != 'skipped' && failure() }}
     continue-on-error: true
     steps:
       - name: Send message
@@ -168,7 +177,7 @@ jobs:
     steps:
       - name: Stop instance
         id: stop-instance
-        uses: zama-ai/slab-github-runner@58f2cae4bf2c0b6728083f5f009b6dc0eb6dc3ac
+        uses: zama-ai/slab-github-runner@c0e7168795bd78f61f61146951ed9d0c73c9b701
         with:
           mode: stop
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}

.github/workflows/benchmark_gpu_l40.yml

@@ -0,0 +1,206 @@
+# Run benchmarks on an L40 VM and return parsed results to Slab CI bot.
+name: Cuda benchmarks (L40)
+
+on:
+  workflow_dispatch:
+  schedule:
+    # Weekly benchmarks will be triggered each Saturday at 1a.m.
+    - cron: '0 1 * * 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 }}
+
+jobs:
+  setup-instance:
+    name: Setup instance (cuda-l40-benchmarks)
+    runs-on: ubuntu-latest
+    if: github.event_name != 'schedule' ||
+      (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@c0e7168795bd78f61f61146951ed9d0c73c9b701
+        with:
+          mode: start
+          github-token: ${{ secrets.SLAB_ACTION_TOKEN }}
+          slab-url: ${{ secrets.SLAB_BASE_URL }}
+          job-secret: ${{ secrets.JOB_SECRET }}
+          backend: hyperstack
+          profile: l40 
+
+  cuda-l40-benchmarks:
+    name: Cuda benchmarks (L40)
+    needs: setup-instance
+    runs-on: ${{ needs.setup-instance.outputs.runner-name }}
+    timeout-minutes: 1440 # 24 hours
+    continue-on-error: true
+    strategy:
+      fail-fast: false
+      max-parallel: 1
+      matrix:
+        command: [integer_multi_bit]
+        op_flavor: [default]
+        # explicit include-based build matrix, of known valid options
+        include:
+          - os: ubuntu-22.04
+            cuda: "12.2"
+            gcc: 11
+    env:
+      CUDA_PATH: /usr/local/cuda-${{ matrix.cuda }}
+      CMAKE_VERSION: 3.29.6
+    steps:
+      # Mandatory on hyperstack since a bootable volume is not re-usable yet.
+      - name: Install dependencies
+        run: |
+          sudo apt update
+          sudo apt install -y checkinstall zlib1g-dev libssl-dev
+          wget https://github.com/Kitware/CMake/releases/download/v${{ env.CMAKE_VERSION }}/cmake-${{ env.CMAKE_VERSION }}.tar.gz
+          tar -zxvf cmake-${{ env.CMAKE_VERSION }}.tar.gz
+          cd cmake-${{ env.CMAKE_VERSION }}
+          ./bootstrap
+          make -j"$(nproc)"
+          sudo make install
+
+      - name: Checkout tfhe-rs repo with tags
+        uses: actions/checkout@d632683dd7b4114ad314bca15554477dd762a938
+        with:
+          fetch-depth: 0
+          token: ${{ secrets.FHE_ACTIONS_TOKEN }}
+
+      - name: Get benchmark details
+        run: |
+          {
+            echo "BENCH_DATE=$(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)";
+          } >> "${GITHUB_ENV}"
+
+      - name: Set up home
+        # "Install rust" step require root user to have a HOME directory which is not set.
+        run: |
+          echo "HOME=/home/ubuntu" >> "${GITHUB_ENV}"
+
+      - name: Install rust
+        uses: dtolnay/rust-toolchain@7b1c307e0dcbda6122208f10795a713336a9b35a
+        with:
+          toolchain: nightly
+
+      - name: Export CUDA variables
+        if: ${{ !cancelled() }}
+        run: |
+          {
+            echo "CUDA_PATH=$CUDA_PATH";
+            echo "LD_LIBRARY_PATH=$CUDA_PATH/lib:$LD_LIBRARY_PATH";
+            echo "CUDACXX=/usr/local/cuda-${{ matrix.cuda }}/bin/nvcc";
+          } >> "${GITHUB_ENV}"
+          echo "$CUDA_PATH/bin" >> "${GITHUB_PATH}"
+
+      # Specify the correct host compilers
+      - name: Export gcc and g++ variables
+        if: ${{ !cancelled() }}
+        run: |
+          {
+            echo "CC=/usr/bin/gcc-${{ matrix.gcc }}";
+            echo "CXX=/usr/bin/g++-${{ matrix.gcc }}";
+            echo "CUDAHOSTCXX=/usr/bin/g++-${{ matrix.gcc }}";
+          } >> "${GITHUB_ENV}"
+
+      - name: Checkout Slab repo
+        uses: actions/checkout@d632683dd7b4114ad314bca15554477dd762a938
+        with:
+          repository: zama-ai/slab
+          path: slab
+          token: ${{ secrets.FHE_ACTIONS_TOKEN }}
+
+      - name: Check device is detected
+        if: ${{ !cancelled() }}
+        run: nvidia-smi
+
+      - name: Run benchmarks with AVX512
+        run: |
+          make BENCH_OP_FLAVOR=${{ matrix.op_flavor }} bench_${{ matrix.command }}_gpu
+
+      - name: Run compression benchmarks with AVX512
+        run: |
+          make bench_integer_compression_gpu
+
+      - name: Run PBS benchmarks 
+        run: |
+          make bench_pbs_gpu
+
+      - name: Run KS benchmarks 
+        run: |
+          make bench_ks_gpu
+
+      - name: Parse results
+        run: |
+          python3 ./ci/benchmark_parser.py target/criterion ${{ env.RESULTS_FILENAME }} \
+          --database tfhe_rs \
+          --hardware "n3-L40x1" \
+          --backend gpu \
+          --project-version "${{ env.COMMIT_HASH }}" \
+          --branch ${{ github.ref_name }} \
+          --commit-date "${{ env.COMMIT_DATE }}" \
+          --bench-date "${{ env.BENCH_DATE }}" \
+          --walk-subdirs \
+          --name-suffix avx512 \
+          --throughput
+
+      - name: Upload parsed results artifact
+        uses: actions/upload-artifact@50769540e7f4bd5e21e526ee35c689e35e0d6874
+        with:
+          name: ${{ github.sha }}_${{ matrix.command }}_${{ matrix.op_flavor }}
+          path: ${{ env.RESULTS_FILENAME }}
+
+      - name: Send data to Slab
+        shell: bash
+        run: |
+          python3 slab/scripts/data_sender.py ${{ env.RESULTS_FILENAME }} "${{ secrets.JOB_SECRET }}" \
+          --slab-url "${{ secrets.SLAB_URL }}"
+
+  slack-notify:
+    name: Slack Notification
+    needs: [ setup-instance, cuda-l40-benchmarks ]
+    runs-on: ubuntu-latest
+    if: ${{ always() && needs.cuda-l40-benchmarks.result != 'skipped' && failure() }}
+    continue-on-error: true
+    steps:
+      - name: Send message
+        uses: rtCamp/action-slack-notify@4e5fb42d249be6a45a298f3c9543b111b02f7907
+        env:
+          SLACK_COLOR: ${{ needs.cuda-l40-benchmarks.result }}
+          SLACK_MESSAGE: "Cuda benchmarks (L40) finished with status: ${{ needs.cuda-l40-benchmarks.result }}. (${{ env.ACTION_RUN_URL }})"
+
+  teardown-instance:
+    name: Teardown instance (cuda-l40-benchmarks)
+    if: ${{ always() && needs.setup-instance.result != 'skipped' }}
+    needs: [ setup-instance, cuda-l40-benchmarks, slack-notify ]
+    runs-on: ubuntu-latest
+    steps:
+      - name: Stop instance
+        id: stop-instance
+        uses: zama-ai/slab-github-runner@c0e7168795bd78f61f61146951ed9d0c73c9b701
+        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@4e5fb42d249be6a45a298f3c9543b111b02f7907
+        env:
+          SLACK_COLOR: ${{ job.status }}
+          SLACK_MESSAGE: "Instance teardown (cuda-l40-benchmarks) finished with status: ${{ job.status }}. (${{ env.ACTION_RUN_URL }})"

.github/workflows/benchmark_integer.yml

@@ -0,0 +1,190 @@
+# Run all integer benchmarks on an AWS instance and return parsed results to Slab CI bot.
+name: Integer benchmarks
+
+on:
+  workflow_dispatch:
+    inputs:
+      all_precisions:
+        description: "Run all precisions"
+        type: boolean
+        default: false
+
+  schedule:
+    # Weekly benchmarks will be triggered each Saturday at 1a.m.
+    - cron: '0 1 * * 6'
+    # Quarterly benchmarks will be triggered right before end of quarter, the 25th of the current month at 4a.m.
+    # These benchmarks are far longer to execute hence the reason to run them only four time a year.
+    - cron: '0 4 25 MAR,JUN,SEP,DEC *'
+
+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 }}
+  FAST_BENCH: TRUE
+
+jobs:
+  prepare-matrix:
+    name: Prepare operations matrix
+    runs-on: ubuntu-latest
+    if: github.event_name != 'schedule' ||
+      (github.event_name == 'schedule' && github.repository == 'zama-ai/tfhe-rs')
+    outputs:
+      op_flavor: ${{ steps.set_op_flavor.outputs.op_flavor }}
+    steps:
+      - name: Weekly benchmarks
+        if: github.event_name == 'workflow_dispatch' ||
+          github.event.schedule == '0 1 * * 6'
+        run: |
+          echo "OP_FLAVOR=[\"default\"]" >> "${GITHUB_ENV}"
+
+      - name: Quarterly benchmarks
+        if: github.event.schedule == '0 4 25 MAR,JUN,SEP,DEC *'
+        run: |
+          echo "OP_FLAVOR=[\"default\", \"smart\", \"unchecked\", \"misc\"]" >> "${GITHUB_ENV}"
+
+      -  name: Set operation flavor output
+         id: set_op_flavor
+         run: |
+          echo "op_flavor=${{ toJSON(env.OP_FLAVOR) }}" >> "${GITHUB_OUTPUT}"
+
+  setup-instance:
+    name: Setup instance (integer-benchmarks)
+    needs: prepare-matrix
+    runs-on: ubuntu-latest
+    outputs:
+      runner-name: ${{ steps.start-instance.outputs.label }}
+    steps:
+      - name: Start instance
+        id: start-instance
+        uses: zama-ai/slab-github-runner@c0e7168795bd78f61f61146951ed9d0c73c9b701
+        with:
+          mode: start
+          github-token: ${{ secrets.SLAB_ACTION_TOKEN }}
+          slab-url: ${{ secrets.SLAB_BASE_URL }}
+          job-secret: ${{ secrets.JOB_SECRET }}
+          backend: aws
+          profile: bench
+
+  integer-benchmarks:
+    name: Execute integer benchmarks for all operations flavor
+    needs: [ prepare-matrix, setup-instance ]
+    runs-on: ${{ needs.setup-instance.outputs.runner-name }}
+    concurrency:
+      group: ${{ github.workflow }}_${{ github.ref }}
+      cancel-in-progress: ${{ github.ref != 'refs/heads/main' }}
+    continue-on-error: true
+    timeout-minutes: 1440  # 24 hours
+    strategy:
+      max-parallel: 1
+      matrix:
+        command: [ integer, integer_multi_bit]
+        op_flavor: ${{ fromJson(needs.prepare-matrix.outputs.op_flavor) }}
+    steps:
+      - name: Checkout tfhe-rs repo with tags
+        uses: actions/checkout@d632683dd7b4114ad314bca15554477dd762a938
+        with:
+          fetch-depth: 0
+          token: ${{ secrets.FHE_ACTIONS_TOKEN }}
+
+      - name: Get benchmark details
+        run: |
+          {
+            echo "BENCH_DATE=$(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)";
+          } >> "${GITHUB_ENV}"
+
+      - name: Set up home
+        # "Install rust" step require root user to have a HOME directory which is not set.
+        run: |
+          echo "HOME=/home/ubuntu" >> "${GITHUB_ENV}"
+
+      - name: Install rust
+        uses: dtolnay/rust-toolchain@7b1c307e0dcbda6122208f10795a713336a9b35a
+        with:
+          toolchain: nightly
+
+      - name: Checkout Slab repo
+        uses: actions/checkout@d632683dd7b4114ad314bca15554477dd762a938
+        with:
+          repository: zama-ai/slab
+          path: slab
+          token: ${{ secrets.FHE_ACTIONS_TOKEN }}
+
+      - name: Should run benchmarks with all precisions
+        if: inputs.all_precisions
+        run: |
+          echo "FAST_BENCH=FALSE" >> "${GITHUB_ENV}"
+
+      - name: Run benchmarks with AVX512
+        run: |
+          make BENCH_OP_FLAVOR=${{ matrix.op_flavor }} bench_${{ matrix.command }}
+
+      # Run these benchmarks only once
+      - name: Run compression benchmarks with AVX512
+        if: matrix.op_flavor == 'default' && matrix.command == 'integer'
+        run: |
+          make bench_integer_compression
+
+      - name: Parse results
+        run: |
+          python3 ./ci/benchmark_parser.py target/criterion ${{ env.RESULTS_FILENAME }} \
+          --database tfhe_rs \
+          --hardware "hpc7a.96xlarge" \
+          --project-version "${{ env.COMMIT_HASH }}" \
+          --branch ${{ github.ref_name }} \
+          --commit-date "${{ env.COMMIT_DATE }}" \
+          --bench-date "${{ env.BENCH_DATE }}" \
+          --walk-subdirs \
+          --name-suffix avx512 \
+          --throughput
+
+      - name: Upload parsed results artifact
+        uses: actions/upload-artifact@50769540e7f4bd5e21e526ee35c689e35e0d6874
+        with:
+          name: ${{ github.sha }}_${{ matrix.command }}_${{ matrix.op_flavor }}
+          path: ${{ env.RESULTS_FILENAME }}
+
+      - name: Send data to Slab
+        shell: bash
+        run: |
+          python3 slab/scripts/data_sender.py ${{ env.RESULTS_FILENAME }} "${{ secrets.JOB_SECRET }}" \
+          --slab-url "${{ secrets.SLAB_URL }}"
+
+      - name: Slack Notification
+        if: ${{ failure() }}
+        continue-on-error: true
+        uses: rtCamp/action-slack-notify@4e5fb42d249be6a45a298f3c9543b111b02f7907
+        env:
+          SLACK_COLOR: ${{ job.status }}
+          SLACK_MESSAGE: "Integer full benchmarks finished with status: ${{ job.status }}. (${{ env.ACTION_RUN_URL }})"
+
+  teardown-instance:
+    name: Teardown instance (integer-benchmarks)
+    if: ${{ always() && needs.setup-instance.result != 'skipped' }}
+    needs: [ setup-instance, integer-benchmarks ]
+    runs-on: ubuntu-latest
+    steps:
+      - name: Stop instance
+        id: stop-instance
+        uses: zama-ai/slab-github-runner@c0e7168795bd78f61f61146951ed9d0c73c9b701
+        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@4e5fb42d249be6a45a298f3c9543b111b02f7907
+        env:
+          SLACK_COLOR: ${{ job.status }}
+          SLACK_MESSAGE: "Instance teardown (integer-benchmarks) finished with status: ${{ job.status }}. (${{ env.ACTION_RUN_URL }})"

.github/workflows/benchmark_shortint.yml

@@ -0,0 +1,186 @@
+# Run all shortint benchmarks on an AWS instance and return parsed results to Slab CI bot.
+name: Shortint full benchmarks
+
+on:
+  workflow_dispatch:
+  schedule:
+    # Weekly benchmarks will be triggered each Saturday at 1a.m.
+    - cron: '0 1 * * 6'
+    # Quarterly benchmarks will be triggered right before end of quarter, the 25th of the current month at 4a.m.
+    # These benchmarks are far longer to execute hence the reason to run them only four time a year.
+    - cron: '0 4 25 MAR,JUN,SEP,DEC *'
+
+
+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 }}
+
+jobs:
+  prepare-matrix:
+    name: Prepare operations matrix
+    runs-on: ubuntu-latest
+    if: github.event_name != 'schedule' ||
+      (github.event_name == 'schedule' && github.repository == 'zama-ai/tfhe-rs')
+    outputs:
+      op_flavor: ${{ steps.set_op_flavor.outputs.op_flavor }}
+    steps:
+      - name: Weekly benchmarks
+        if: github.event_name == 'workflow_dispatch' ||
+          github.event.schedule == '0 1 * * 6'
+        run: |
+          echo "OP_FLAVOR=[\"default\"]" >> "${GITHUB_ENV}"
+
+      - name: Quarterly benchmarks
+        if: github.event.schedule == '0 4 25 MAR,JUN,SEP,DEC *'
+        run: |
+          echo "OP_FLAVOR=[\"default\", \"smart\", \"unchecked\"]" >> "${GITHUB_ENV}"
+
+      - name: Set operation flavor output
+        id: set_op_flavor
+        run: |
+          echo "op_flavor=${{ toJSON(env.OP_FLAVOR) }}" >> "${GITHUB_OUTPUT}"
+
+  setup-instance:
+    name: Setup instance (shortint-benchmarks)
+    needs: prepare-matrix
+    runs-on: ubuntu-latest
+    outputs:
+      runner-name: ${{ steps.start-instance.outputs.label }}
+    steps:
+      - name: Start instance
+        id: start-instance
+        uses: zama-ai/slab-github-runner@c0e7168795bd78f61f61146951ed9d0c73c9b701
+        with:
+          mode: start
+          github-token: ${{ secrets.SLAB_ACTION_TOKEN }}
+          slab-url: ${{ secrets.SLAB_BASE_URL }}
+          job-secret: ${{ secrets.JOB_SECRET }}
+          backend: aws
+          profile: bench
+
+  shortint-benchmarks:
+    name: Execute shortint benchmarks for all operations flavor
+    needs: [ prepare-matrix, setup-instance ]
+    runs-on: ${{ needs.setup-instance.outputs.runner-name }}
+    concurrency:
+      group: ${{ github.workflow }}_${{ github.ref }}
+      cancel-in-progress: ${{ github.ref != 'refs/heads/main' }}
+    continue-on-error: true
+    strategy:
+      max-parallel: 1
+      matrix:
+        op_flavor: ${{ fromJson(needs.prepare-matrix.outputs.op_flavor) }}
+    steps:
+      - name: Checkout tfhe-rs repo with tags
+        uses: actions/checkout@d632683dd7b4114ad314bca15554477dd762a938
+        with:
+          fetch-depth: 0
+          token: ${{ secrets.FHE_ACTIONS_TOKEN }}
+
+      - name: Get benchmark details
+        run: |
+          {
+            echo "BENCH_DATE=$(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)";
+          } >> "${GITHUB_ENV}"
+
+      - name: Set up home
+        # "Install rust" step require root user to have a HOME directory which is not set.
+        run: |
+          echo "HOME=/home/ubuntu" >> "${GITHUB_ENV}"
+
+      - name: Install rust
+        uses: dtolnay/rust-toolchain@7b1c307e0dcbda6122208f10795a713336a9b35a
+        with:
+          toolchain: nightly
+
+      - name: Checkout Slab repo
+        uses: actions/checkout@d632683dd7b4114ad314bca15554477dd762a938
+        with:
+          repository: zama-ai/slab
+          path: slab
+          token: ${{ secrets.FHE_ACTIONS_TOKEN }}
+
+      - name: Run benchmarks with AVX512
+        run: |
+          make BENCH_OP_FLAVOR=${{ matrix.op_flavor }} bench_shortint
+
+      - 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 ${{ env.RESULTS_FILENAME }} \
+          --database tfhe_rs \
+          --hardware "hpc7a.96xlarge" \
+          --project-version "${COMMIT_HASH}" \
+          --branch ${{ github.ref_name }} \
+          --commit-date "${COMMIT_DATE}" \
+          --bench-date "${{ env.BENCH_DATE }}" \
+          --walk-subdirs \
+          --name-suffix avx512 \
+          --throughput
+
+      # This small benchmark needs to be executed only once.
+      - name: Measure key sizes
+        if: matrix.op_flavor == 'default'
+        run: |
+          make measure_shortint_key_sizes
+
+      - name: Parse key sizes results
+        if: matrix.op_flavor == 'default'
+        run: |
+          python3 ./ci/benchmark_parser.py tfhe/shortint_key_sizes.csv ${{ env.RESULTS_FILENAME }} \
+          --key-sizes \
+          --append-results
+
+      - name: Upload parsed results artifact
+        uses: actions/upload-artifact@50769540e7f4bd5e21e526ee35c689e35e0d6874
+        with:
+          name: ${{ github.sha }}_shortint_${{ matrix.op_flavor }}
+          path: ${{ env.RESULTS_FILENAME }}
+
+      - name: Send data to Slab
+        shell: bash
+        run: |
+          python3 slab/scripts/data_sender.py ${{ env.RESULTS_FILENAME }} "${{ secrets.JOB_SECRET }}" \
+          --slab-url "${{ secrets.SLAB_URL }}"
+
+      - name: Slack Notification
+        if: ${{ failure() }}
+        continue-on-error: true
+        uses: rtCamp/action-slack-notify@4e5fb42d249be6a45a298f3c9543b111b02f7907
+        env:
+          SLACK_COLOR: ${{ job.status }}
+          SLACK_MESSAGE: "Shortint full benchmarks finished with status: ${{ job.status }}. (${{ env.ACTION_RUN_URL }})"
+
+  teardown-instance:
+    name: Teardown instance (shortint-benchmarks)
+    if: ${{ always() && needs.setup-instance.result != 'skipped' }}
+    needs: [ setup-instance, shortint-benchmarks ]
+    runs-on: ubuntu-latest
+    steps:
+      - name: Stop instance
+        id: stop-instance
+        uses: zama-ai/slab-github-runner@c0e7168795bd78f61f61146951ed9d0c73c9b701
+        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@4e5fb42d249be6a45a298f3c9543b111b02f7907
+        env:
+          SLACK_COLOR: ${{ job.status }}
+          SLACK_MESSAGE: "Instance teardown (shortint-benchmarks) finished with status: ${{ job.status }}. (${{ env.ACTION_RUN_URL }})"

.github/workflows/benchmark_signed_integer.yml

@@ -0,0 +1,184 @@
+# Run all signed integer benchmarks on an AWS instance and return parsed results to Slab CI bot.
+name: Signed Integer full benchmarks
+
+on:
+  workflow_dispatch:
+    inputs:
+      all_precisions:
+        description: "Run all precisions"
+        type: boolean
+        default: false
+
+  schedule:
+    # Weekly benchmarks will be triggered each Saturday at 1a.m.
+    - cron: '0 1 * * 6'
+    # Quarterly benchmarks will be triggered right before end of quarter, the 25th of the current month at 4a.m.
+    # These benchmarks are far longer to execute hence the reason to run them only four time a year.
+    - cron: '0 4 25 MAR,JUN,SEP,DEC *'
+
+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 }}
+  FAST_BENCH: TRUE
+
+jobs:
+  prepare-matrix:
+    name: Prepare operations matrix
+    runs-on: ubuntu-latest
+    if: github.event_name != 'schedule' ||
+      (github.event_name == 'schedule' && github.repository == 'zama-ai/tfhe-rs')
+    outputs:
+      op_flavor: ${{ steps.set_op_flavor.outputs.op_flavor }}
+    steps:
+      - name: Weekly benchmarks
+        if: github.event_name == 'workflow_dispatch' ||
+          github.event.schedule == '0 1 * * 6'
+        run: |
+          echo "OP_FLAVOR=[\"default\"]" >> "${GITHUB_ENV}"
+
+      - name: Quarterly benchmarks
+        if: github.event.schedule == '0 4 25 MAR,JUN,SEP,DEC *'
+        run: |
+          echo "OP_FLAVOR=[\"default\", \"unchecked\"]" >> "${GITHUB_ENV}"
+
+      - name: Set operation flavor output
+        id: set_op_flavor
+        run: |
+          echo "op_flavor=${{ toJSON(env.OP_FLAVOR) }}" >> "${GITHUB_OUTPUT}"
+
+  setup-instance:
+    name: Setup instance (signed-integer-benchmarks)
+    needs: prepare-matrix
+    runs-on: ubuntu-latest
+    outputs:
+      runner-name: ${{ steps.start-instance.outputs.label }}
+    steps:
+      - name: Start instance
+        id: start-instance
+        uses: zama-ai/slab-github-runner@c0e7168795bd78f61f61146951ed9d0c73c9b701
+        with:
+          mode: start
+          github-token: ${{ secrets.SLAB_ACTION_TOKEN }}
+          slab-url: ${{ secrets.SLAB_BASE_URL }}
+          job-secret: ${{ secrets.JOB_SECRET }}
+          backend: aws
+          profile: bench
+
+  signed-integer-benchmarks:
+    name: Execute signed integer benchmarks for all operations flavor
+    needs: [ prepare-matrix, setup-instance ]
+    runs-on: ${{ needs.setup-instance.outputs.runner-name }}
+    concurrency:
+      group: ${{ github.workflow }}_${{ github.ref }}
+      cancel-in-progress: ${{ github.ref != 'refs/heads/main' }}
+    continue-on-error: true
+    timeout-minutes: 1440  # 24 hours
+    strategy:
+      max-parallel: 1
+      matrix:
+        command: [ integer, integer_multi_bit ]
+        op_flavor: [ default, unchecked ]
+    steps:
+      - name: Checkout tfhe-rs repo with tags
+        uses: actions/checkout@d632683dd7b4114ad314bca15554477dd762a938
+        with:
+          fetch-depth: 0
+          token: ${{ secrets.FHE_ACTIONS_TOKEN }}
+
+      - name: Get benchmark details
+        run: |
+          {
+            echo "BENCH_DATE=$(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)";
+          } >> "${GITHUB_ENV}"
+
+      - name: Set up home
+        # "Install rust" step require root user to have a HOME directory which is not set.
+        run: |
+          echo "HOME=/home/ubuntu" >> "${GITHUB_ENV}"
+
+      - name: Install rust
+        uses: dtolnay/rust-toolchain@7b1c307e0dcbda6122208f10795a713336a9b35a
+        with:
+          toolchain: nightly
+
+      - name: Checkout Slab repo
+        uses: actions/checkout@d632683dd7b4114ad314bca15554477dd762a938
+        with:
+          repository: zama-ai/slab
+          path: slab
+          token: ${{ secrets.FHE_ACTIONS_TOKEN }}
+
+      - name: Should run benchmarks with all precisions
+        if: inputs.all_precisions
+        run: |
+          echo "FAST_BENCH=FALSE" >> "${GITHUB_ENV}"
+
+      - name: Run benchmarks with AVX512
+        run: |
+          make BENCH_OP_FLAVOR=${{ matrix.op_flavor }} bench_signed_${{ matrix.command }}
+
+      - name: Parse results
+        run: |
+          python3 ./ci/benchmark_parser.py target/criterion ${{ env.RESULTS_FILENAME }} \
+          --database tfhe_rs \
+          --hardware "hpc7a.96xlarge" \
+          --project-version "${{ env.COMMIT_HASH }}" \
+          --branch ${{ github.ref_name }} \
+          --commit-date "${{ env.COMMIT_DATE }}" \
+          --bench-date "${{ env.BENCH_DATE }}" \
+          --walk-subdirs \
+          --name-suffix avx512 \
+          --throughput
+
+      - name: Upload parsed results artifact
+        uses: actions/upload-artifact@50769540e7f4bd5e21e526ee35c689e35e0d6874
+        with:
+          name: ${{ github.sha }}_${{ matrix.command }}_${{ matrix.op_flavor }}
+          path: ${{ env.RESULTS_FILENAME }}
+
+      - name: Send data to Slab
+        shell: bash
+        run: |
+          python3 slab/scripts/data_sender.py ${{ env.RESULTS_FILENAME }} "${{ secrets.JOB_SECRET }}" \
+          --slab-url "${{ secrets.SLAB_URL }}"
+
+      - name: Slack Notification
+        if: ${{ failure() }}
+        continue-on-error: true
+        uses: rtCamp/action-slack-notify@4e5fb42d249be6a45a298f3c9543b111b02f7907
+        env:
+          SLACK_COLOR: ${{ job.status }}
+          SLACK_MESSAGE: "Signed integer full benchmarks finished with status: ${{ job.status }}. (${{ env.ACTION_RUN_URL }})"
+
+  teardown-instance:
+    name: Teardown instance (integer-benchmarks)
+    if: ${{ always() && needs.setup-instance.result != 'skipped' }}
+    needs: [ setup-instance, signed-integer-benchmarks ]
+    runs-on: ubuntu-latest
+    steps:
+      - name: Stop instance
+        id: stop-instance
+        uses: zama-ai/slab-github-runner@c0e7168795bd78f61f61146951ed9d0c73c9b701
+        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@4e5fb42d249be6a45a298f3c9543b111b02f7907
+        env:
+          SLACK_COLOR: ${{ job.status }}
+          SLACK_MESSAGE: "Instance teardown (signed-integer-benchmarks) finished with status: ${{ job.status }}. (${{ env.ACTION_RUN_URL }})"

.github/workflows/benchmark_wasm_client.yml

@@ -25,20 +25,21 @@ jobs:
   should-run:
     runs-on: ubuntu-latest
     if: github.event_name == 'workflow_dispatch' ||
-      ((github.event_name == 'push' || github.event_name == 'schedule') && github.repository == 'zama-ai/tfhe-rs')
+      (github.event_name == 'schedule' && github.repository == 'zama-ai/tfhe-rs') ||
+      (github.event_name == 'push' && github.repository == 'zama-ai/tfhe-rs')
     permissions:
       pull-requests: write
     outputs:
       wasm_bench: ${{ steps.changed-files.outputs.wasm_bench_any_changed }}
     steps:
       - name: Checkout tfhe-rs
-        uses: actions/checkout@692973e3d937129bcbf40652eb9f2f61becf3332
+        uses: actions/checkout@d632683dd7b4114ad314bca15554477dd762a938
         with:
           fetch-depth: 0
 
       - name: Check for file changes
         id: changed-files
-        uses: tj-actions/changed-files@eaf854ef0c266753e1abec356dcf17d92695b251
+        uses: tj-actions/changed-files@c3a1bb2c992d77180ae65be6ae6c166cf40f857c
         with:
           since_last_remote_commit: true
           files_yaml: |
@@ -53,7 +54,8 @@ jobs:
 
   setup-instance:
     name: Setup instance (wasm-client-benchmarks)
-    if: github.event_name != 'push' ||
+    if: github.event_name == 'workflow_dispatch' ||
+      (github.event_name == 'schedule' && github.repository == 'zama-ai/tfhe-rs') ||
       (github.event_name == 'push' && github.repository == 'zama-ai/tfhe-rs' && needs.should-run.outputs.wasm_bench)
     needs: should-run
     runs-on: ubuntu-latest
@@ -62,7 +64,7 @@ jobs:
     steps:
       - name: Start instance
         id: start-instance
-        uses: zama-ai/slab-github-runner@58f2cae4bf2c0b6728083f5f009b6dc0eb6dc3ac
+        uses: zama-ai/slab-github-runner@c0e7168795bd78f61f61146951ed9d0c73c9b701
         with:
           mode: start
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}
@@ -73,15 +75,19 @@ jobs:
 
   wasm-client-benchmarks:
     name: Execute WASM client benchmarks
-    needs: [ should-run, setup-instance ]
-    if: github.event_name != 'push' ||
-      (github.event_name == 'push' && github.repository == 'zama-ai/tfhe-rs' && needs.should-run.outputs.wasm_bench)
+    needs: setup-instance
+    if: needs.setup-instance.result != 'skipped'
     runs-on: ${{ needs.setup-instance.outputs.runner-name }}
+    strategy:
+      max-parallel: 1
+      matrix:
+        browser: [ chrome, firefox ]
     steps:
       - name: Checkout tfhe-rs repo with tags
-        uses: actions/checkout@692973e3d937129bcbf40652eb9f2f61becf3332
+        uses: actions/checkout@d632683dd7b4114ad314bca15554477dd762a938
         with:
           fetch-depth: 0
+          token: ${{ secrets.FHE_ACTIONS_TOKEN }}
 
       - name: Get benchmark details
         run: |
@@ -97,14 +103,19 @@ jobs:
           echo "HOME=/home/ubuntu" >> "${GITHUB_ENV}"
 
       - name: Install rust
-        uses: dtolnay/rust-toolchain@21dc36fb71dd22e3317045c0c31a3f4249868b17
+        uses: dtolnay/rust-toolchain@7b1c307e0dcbda6122208f10795a713336a9b35a
         with:
           toolchain: nightly
 
-      - name: Run benchmarks
+      - name: Install web resources
         run: |
           make install_node
-          make bench_web_js_api_parallel_ci
+          make install_${{ matrix.browser }}_browser
+          make install_${{ matrix.browser }}_driver
+
+      - name: Run benchmarks
+        run: |
+          make bench_web_js_api_parallel_${{ matrix.browser }}_ci
 
       - name: Parse results
         run: |
@@ -117,25 +128,29 @@ jobs:
           --commit-date "${{ env.COMMIT_DATE }}" \
           --bench-date "${{ env.BENCH_DATE }}" \
           --key-gen
+          rm tfhe/wasm_pk_gen.csv
 
+      # Run these benchmarks only once
       - name: Measure public key and ciphertext sizes in HL Api
+        if:  matrix.browser == 'chrome'
         run: |
           make measure_hlapi_compact_pk_ct_sizes
 
       - 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 }} \
           --key-gen \
           --append-results
 
       - name: Upload parsed results artifact
-        uses: actions/upload-artifact@65462800fd760344b1a7b4382951275a0abb4808
+        uses: actions/upload-artifact@50769540e7f4bd5e21e526ee35c689e35e0d6874
         with:
-          name: ${{ github.sha }}_wasm
+          name: ${{ github.sha }}_wasm_${{ matrix.browser }}
           path: ${{ env.RESULTS_FILENAME }}
 
       - name: Checkout Slab repo
-        uses: actions/checkout@692973e3d937129bcbf40652eb9f2f61becf3332
+        uses: actions/checkout@d632683dd7b4114ad314bca15554477dd762a938
         with:
           repository: zama-ai/slab
           path: slab
@@ -144,16 +159,8 @@ 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 ${{ env.RESULTS_FILENAME }} "${{ secrets.JOB_SECRET }}" \
+          --slab-url "${{ secrets.SLAB_URL }}"
 
       - name: Slack Notification
         if: ${{ failure() }}
@@ -161,7 +168,7 @@ jobs:
         uses: rtCamp/action-slack-notify@4e5fb42d249be6a45a298f3c9543b111b02f7907
         env:
           SLACK_COLOR: ${{ job.status }}
-          SLACK_MESSAGE: "WASM benchmarks finished with status: ${{ job.status }}. (${{ env.ACTION_RUN_URL }})"
+          SLACK_MESSAGE: "WASM benchmarks (${{ matrix.browser }}) finished with status: ${{ job.status }}. (${{ env.ACTION_RUN_URL }})"
 
   teardown-instance:
     name: Teardown instance (wasm-client-benchmarks)
@@ -171,7 +178,7 @@ jobs:
     steps:
       - name: Stop instance
         id: stop-instance
-        uses: zama-ai/slab-github-runner@58f2cae4bf2c0b6728083f5f009b6dc0eb6dc3ac
+        uses: zama-ai/slab-github-runner@c0e7168795bd78f61f61146951ed9d0c73c9b701
         with:
           mode: stop
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}

.github/workflows/benchmark_zk_pke.yml

@@ -24,19 +24,19 @@ env:
 jobs:
   should-run:
     runs-on: ubuntu-latest
-    if: github.event_name != 'push' ||
-      (github.event_name == 'push' && github.repository == 'zama-ai/tfhe-rs')
+    if: github.event_name == 'workflow_dispatch' ||
+      ((github.event_name == 'push' || github.event_name == 'schedule') && github.repository == 'zama-ai/tfhe-rs')
     outputs:
       zk_pok_changed: ${{ steps.changed-files.outputs.zk_pok_any_changed }}
     steps:
       - name: Checkout tfhe-rs
-        uses: actions/checkout@692973e3d937129bcbf40652eb9f2f61becf3332
+        uses: actions/checkout@d632683dd7b4114ad314bca15554477dd762a938
         with:
           fetch-depth: 0
 
       - name: Check for file changes
         id: changed-files
-        uses: tj-actions/changed-files@eaf854ef0c266753e1abec356dcf17d92695b251
+        uses: tj-actions/changed-files@c3a1bb2c992d77180ae65be6ae6c166cf40f857c
         with:
           since_last_remote_commit: true
           files_yaml: |
@@ -55,7 +55,7 @@ jobs:
     name: Setup instance (pke-zk-benchmarks)
     runs-on: ubuntu-latest
     needs: should-run
-    if: github.event_name != 'push' ||
+    if: github.event_name == 'workflow_dispatch' ||
       (github.event_name == 'schedule' && github.repository == 'zama-ai/tfhe-rs') ||
       (github.event_name == 'push' &&
       github.repository == 'zama-ai/tfhe-rs' &&
@@ -65,7 +65,7 @@ jobs:
     steps:
       - name: Start instance
         id: start-instance
-        uses: zama-ai/slab-github-runner@58f2cae4bf2c0b6728083f5f009b6dc0eb6dc3ac
+        uses: zama-ai/slab-github-runner@c0e7168795bd78f61f61146951ed9d0c73c9b701
         with:
           mode: start
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}
@@ -76,19 +76,18 @@ jobs:
 
   pke-zk-benchmarks:
     name: Execute PKE ZK benchmarks
-    if: github.event_name != 'push' ||
-      ((github.event_name == 'push' || github.event_name == 'schedule') &&
-      needs.setup-instance.result != 'skipped')
-    needs: [ should-run, setup-instance ]
+    if: needs.setup-instance.result != 'skipped'
+    needs: setup-instance
     concurrency:
-      group: ${{ github.workflow }}_${{github.event_name}}_${{ github.ref }}
-      cancel-in-progress: true
+      group: ${{ github.workflow }}_${{github.event_name}}_${{ github.ref }}${{ github.ref == 'refs/heads/main' && github.sha || '' }}
+      cancel-in-progress: ${{ github.ref != 'refs/heads/main' }}
     runs-on: ${{ needs.setup-instance.outputs.runner-name }}
     steps:
       - name: Checkout tfhe-rs repo with tags
-        uses: actions/checkout@692973e3d937129bcbf40652eb9f2f61becf3332
+        uses: actions/checkout@d632683dd7b4114ad314bca15554477dd762a938
         with:
           fetch-depth: 0
+          token: ${{ secrets.FHE_ACTIONS_TOKEN }}
 
       - name: Get benchmark details
         run: |
@@ -104,12 +103,12 @@ jobs:
           echo "HOME=/home/ubuntu" >> "${GITHUB_ENV}"
 
       - name: Install rust
-        uses: dtolnay/rust-toolchain@21dc36fb71dd22e3317045c0c31a3f4249868b17
+        uses: dtolnay/rust-toolchain@7b1c307e0dcbda6122208f10795a713336a9b35a
         with:
           toolchain: nightly
 
       - name: Checkout Slab repo
-        uses: actions/checkout@692973e3d937129bcbf40652eb9f2f61becf3332
+        uses: actions/checkout@d632683dd7b4114ad314bca15554477dd762a938
         with:
           repository: zama-ai/slab
           path: slab
@@ -140,13 +139,13 @@ jobs:
           --append-results
 
       - name: Upload parsed results artifact
-        uses: actions/upload-artifact@65462800fd760344b1a7b4382951275a0abb4808
+        uses: actions/upload-artifact@50769540e7f4bd5e21e526ee35c689e35e0d6874
         with:
           name: ${{ github.sha }}_integer_zk
           path: ${{ env.RESULTS_FILENAME }}
 
       - name: Checkout Slab repo
-        uses: actions/checkout@692973e3d937129bcbf40652eb9f2f61becf3332
+        uses: actions/checkout@d632683dd7b4114ad314bca15554477dd762a938
         with:
           repository: zama-ai/slab
           path: slab
@@ -155,19 +154,11 @@ 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 ${{ env.RESULTS_FILENAME }} "${{ secrets.JOB_SECRET }}" \
+          --slab-url "${{ secrets.SLAB_URL }}"
 
       - name: Slack Notification
-        if: ${{ !success() && !cancelled() }}
+        if: ${{ failure() }}
         continue-on-error: true
         uses: rtCamp/action-slack-notify@4e5fb42d249be6a45a298f3c9543b111b02f7907
         env:
@@ -182,7 +173,7 @@ jobs:
     steps:
       - name: Stop instance
         id: stop-instance
-        uses: zama-ai/slab-github-runner@58f2cae4bf2c0b6728083f5f009b6dc0eb6dc3ac
+        uses: zama-ai/slab-github-runner@c0e7168795bd78f61f61146951ed9d0c73c9b701
         with:
           mode: stop
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}

.github/workflows/boolean_benchmark.yml

@@ -1,136 +0,0 @@
-# Run boolean benchmarks on an AWS instance and return parsed results to Slab CI bot.
-name: Boolean benchmarks
-
-on:
-  workflow_dispatch:
-    inputs:
-      instance_id:
-        description: "Instance ID"
-        type: string
-      instance_image_id:
-        description: "Instance AMI ID"
-        type: string
-      instance_type:
-        description: "Instance product type"
-        type: string
-      runner_name:
-        description: "Action runner name"
-        type: string
-      request_id:
-        description: "Slab request ID"
-        type: string
-      # This input is not used in this workflow but still mandatory since a calling workflow could
-      # use it. If a triggering command include a user_inputs field, then the triggered workflow
-      # must include this very input, otherwise the workflow won't be called.
-      # See start_full_benchmarks.yml as example.
-      user_inputs:
-        description: "Type of benchmarks to run"
-        type: string
-        default: "weekly_benchmarks"
-
-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"
-
-jobs:
-  run-boolean-benchmarks:
-    name: Execute boolean benchmarks in EC2
-    runs-on: ${{ github.event.inputs.runner_name }}
-    if: ${{ !cancelled() }}
-    steps:
-      - name: Instance configuration used
-        run: |
-          echo "IDs: ${{ inputs.instance_id }}"
-          echo "AMI: ${{ inputs.instance_image_id }}"
-          echo "Type: ${{ inputs.instance_type }}"
-          echo "Request ID: ${{ inputs.request_id }}"
-
-      - name: Get benchmark date
-        run: |
-          echo "BENCH_DATE=$(date --iso-8601=seconds)" >> "${GITHUB_ENV}"
-
-      - name: Checkout tfhe-rs repo with tags
-        uses: actions/checkout@692973e3d937129bcbf40652eb9f2f61becf3332
-        with:
-          fetch-depth: 0
-
-      - name: Set up home
-        # "Install rust" step require root user to have a HOME directory which is not set.
-        run: |
-          echo "HOME=/home/ubuntu" >> "${GITHUB_ENV}"
-
-      - name: Install rust
-        uses: dtolnay/rust-toolchain@21dc36fb71dd22e3317045c0c31a3f4249868b17
-        with:
-          toolchain: nightly
-
-      - name: Run benchmarks with AVX512
-        run: |
-          make bench_boolean
-
-      - 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 }} \
-          --database tfhe_rs \
-          --hardware ${{ inputs.instance_type }} \
-          --project-version "${COMMIT_HASH}" \
-          --branch ${{ github.ref_name }} \
-          --commit-date "${COMMIT_DATE}" \
-          --bench-date "${{ env.BENCH_DATE }}" \
-          --walk-subdirs \
-          --name-suffix avx512 \
-          --throughput
-
-      - name: Measure key sizes
-        run: |
-          make measure_boolean_key_sizes
-
-      - name: Parse key sizes results
-        run: |
-          python3 ./ci/benchmark_parser.py tfhe/boolean_key_sizes.csv ${{ env.RESULTS_FILENAME }} \
-          --key-sizes \
-          --append-results
-
-      - name: Upload parsed results artifact
-        uses: actions/upload-artifact@65462800fd760344b1a7b4382951275a0abb4808
-        with:
-          name: ${{ github.sha }}_boolean
-          path: ${{ env.RESULTS_FILENAME }}
-
-      - name: Checkout Slab repo
-        uses: actions/checkout@692973e3d937129bcbf40652eb9f2f61becf3332
-        with:
-          repository: zama-ai/slab
-          path: slab
-          token: ${{ secrets.FHE_ACTIONS_TOKEN }}
-
-      - name: Send data to Slab
-        shell: bash
-        run: |
-          echo "Computing HMac on results file"
-          SIGNATURE="$(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 }}
-
-      - name: Slack Notification
-        if: ${{ failure() }}
-        continue-on-error: true
-        uses: rtCamp/action-slack-notify@4e5fb42d249be6a45a298f3c9543b111b02f7907
-        env:
-          SLACK_COLOR: ${{ job.status }}
-          SLACK_CHANNEL: ${{ secrets.SLACK_CHANNEL }}
-          SLACK_ICON: https://pbs.twimg.com/profile_images/1274014582265298945/OjBKP9kn_400x400.png
-          SLACK_MESSAGE: "Boolean benchmarks finished with status: ${{ job.status }}. (${{ env.ACTION_RUN_URL }})"
-          SLACK_USERNAME: ${{ secrets.BOT_USERNAME }}
-          SLACK_WEBHOOK: ${{ secrets.SLACK_WEBHOOK }}

.github/workflows/cargo_build.yml

@@ -19,14 +19,21 @@ jobs:
 
     strategy:
       matrix:
-        os: [ubuntu-latest, macos-latest-large, windows-latest]
+        # GitHub macos-latest are now M1 macs, so use ours, we limit what runs so it will be fast
+        # even with a few PRs
+        os: [large_ubuntu_16, macos-latest, windows-latest]
       fail-fast: false
 
     steps:
-      - uses: actions/checkout@692973e3d937129bcbf40652eb9f2f61becf3332
+      - uses: actions/checkout@d632683dd7b4114ad314bca15554477dd762a938
+
+      - name: Install latest stable
+        uses: dtolnay/rust-toolchain@7b1c307e0dcbda6122208f10795a713336a9b35a
+        with:
+          toolchain: stable
 
       - name: Install and run newline linter checks
-        if: matrix.os == 'ubuntu-latest'
+        if: ${{ contains(matrix.os, 'ubuntu') }}
         run: |
           wget https://github.com/fernandrone/linelint/releases/download/0.0.6/linelint-linux-amd64
           echo "16b70fb7b471d6f95cbdc0b4e5dc2b0ac9e84ba9ecdc488f7bdf13df823aca4b linelint-linux-amd64" > checksum
@@ -36,27 +43,33 @@ jobs:
           make check_newline
 
       - name: Run pcc checks
+        if: ${{ contains(matrix.os, 'ubuntu') }}
         run: |
           make pcc
 
       - name: Build concrete-csprng
+        if: ${{ contains(matrix.os, 'ubuntu') }}
         run: |
           make build_concrete_csprng
 
       - name: Build Release core
+        if: ${{ contains(matrix.os, 'ubuntu') }}
         run: |
           make build_core AVX512_SUPPORT=ON
           make build_core_experimental AVX512_SUPPORT=ON
 
       - name: Build Release boolean
+        if: ${{ contains(matrix.os, 'ubuntu') }}
         run: |
           make build_boolean
 
       - name: Build Release shortint
+        if: ${{ contains(matrix.os, 'ubuntu') }}
         run: |
           make build_shortint
 
       - name: Build Release integer
+        if: ${{ contains(matrix.os, 'ubuntu') }}
         run: |
           make build_integer
 
@@ -65,10 +78,12 @@ jobs:
           make build_tfhe_full
 
       - name: Build Release c_api
+        if: ${{ contains(matrix.os, 'ubuntu') }}
         run: |
           make build_c_api
 
       - name: Build coverage tests
+        if: ${{ contains(matrix.os, 'ubuntu') }}
         run: |
           make build_tfhe_coverage
 

.github/workflows/check_commit.yml

@@ -10,7 +10,7 @@ jobs:
       - name: Check first line
         uses: gsactions/commit-message-checker@16fa2d5de096ae0d35626443bcd24f1e756cafee
         with:
-          pattern: '^((feat|fix|chore|refactor|style|test|docs|doc)(\([\w\-_]+\))?\:) .+$'
+          pattern: '^((feat|fix|chore|refactor|style|test|docs|doc)(\([\w\-_]+\))?\!?\:) .+$'
           flags: "gs"
           error: 'Your first line has to contain a commit type and scope like "feat(my_feature): msg".'
           excludeDescription: "true" # optional: this excludes the description body of a pull request

.github/workflows/ci_lint.yml

@@ -13,7 +13,7 @@ jobs:
     runs-on: ubuntu-latest
     steps:
       - name: Checkout tfhe-rs
-        uses: actions/checkout@692973e3d937129bcbf40652eb9f2f61becf3332
+        uses: actions/checkout@d632683dd7b4114ad314bca15554477dd762a938
 
       - name: Get actionlint
         run: |
@@ -25,3 +25,9 @@ jobs:
       - name: Lint workflows
         run: |
           make lint_workflow
+
+      - name: Ensure SHA pinned actions
+        uses: zgosalvez/github-actions-ensure-sha-pinned-actions@b8f9a25a51fe633d9215ac7734854dc11cd299cb # v3.0.13
+        with:
+          allowlist: |
+            slsa-framework/slsa-github-generator

.github/workflows/code_coverage.yml

@@ -25,7 +25,7 @@ jobs:
     steps:
       - name: Start instance
         id: start-instance
-        uses: zama-ai/slab-github-runner@58f2cae4bf2c0b6728083f5f009b6dc0eb6dc3ac
+        uses: zama-ai/slab-github-runner@c0e7168795bd78f61f61146951ed9d0c73c9b701
         with:
           mode: start
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}
@@ -44,20 +44,20 @@ jobs:
     timeout-minutes: 5760 # 4 days
     steps:
       - name: Checkout tfhe-rs
-        uses: actions/checkout@692973e3d937129bcbf40652eb9f2f61becf3332
+        uses: actions/checkout@d632683dd7b4114ad314bca15554477dd762a938
 
       - name: Set up home
         run: |
           echo "HOME=/home/ubuntu" >> "${GITHUB_ENV}"
 
       - name: Install latest stable
-        uses: dtolnay/rust-toolchain@21dc36fb71dd22e3317045c0c31a3f4249868b17
+        uses: dtolnay/rust-toolchain@7b1c307e0dcbda6122208f10795a713336a9b35a
         with:
           toolchain: stable
 
       - name: Check for file changes
         id: changed-files
-        uses: tj-actions/changed-files@eaf854ef0c266753e1abec356dcf17d92695b251
+        uses: tj-actions/changed-files@c3a1bb2c992d77180ae65be6ae6c166cf40f857c
         with:
           files_yaml: |
             tfhe:
@@ -87,7 +87,7 @@ jobs:
           make test_shortint_cov
 
       - name: Upload tfhe coverage to Codecov
-        uses: codecov/codecov-action@e28ff129e5465c2c0dcc6f003fc735cb6ae0c673
+        uses: codecov/codecov-action@b9fd7d16f6d7d1b5d2bec1a2887e65ceed900238
         if: steps.changed-files.outputs.tfhe_any_changed == 'true'
         with:
           token: ${{ secrets.CODECOV_TOKEN }}
@@ -101,7 +101,7 @@ jobs:
           make test_integer_cov
 
       - name: Upload tfhe coverage to Codecov
-        uses: codecov/codecov-action@e28ff129e5465c2c0dcc6f003fc735cb6ae0c673
+        uses: codecov/codecov-action@b9fd7d16f6d7d1b5d2bec1a2887e65ceed900238
         if: steps.changed-files.outputs.tfhe_any_changed == 'true'
         with:
           token: ${{ secrets.CODECOV_TOKEN }}
@@ -125,7 +125,7 @@ jobs:
     steps:
       - name: Stop instance
         id: stop-instance
-        uses: zama-ai/slab-github-runner@58f2cae4bf2c0b6728083f5f009b6dc0eb6dc3ac
+        uses: zama-ai/slab-github-runner@c0e7168795bd78f61f61146951ed9d0c73c9b701
         with:
           mode: stop
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}

.github/workflows/core_crypto_benchmark.yml

@@ -1,128 +0,0 @@
-# Run core crypto benchmarks on an AWS instance and return parsed results to Slab CI bot.
-name: Core crypto benchmarks
-
-on:
-  workflow_dispatch:
-    inputs:
-      instance_id:
-        description: "Instance ID"
-        type: string
-      instance_image_id:
-        description: "Instance AMI ID"
-        type: string
-      instance_type:
-        description: "Instance product type"
-        type: string
-      runner_name:
-        description: "Action runner name"
-        type: string
-      request_id:
-        description: "Slab request ID"
-        type: string
-      # This input is not used in this workflow but still mandatory since a calling workflow could
-      # use it. If a triggering command include a user_inputs field, then the triggered workflow
-      # must include this very input, otherwise the workflow won't be called.
-      # See start_full_benchmarks.yml as example.
-      user_inputs:
-        description: "Type of benchmarks to run"
-        type: string
-        default: "weekly_benchmarks"
-
-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"
-
-jobs:
-  run-core-crypto-benchmarks:
-    name: Execute core crypto benchmarks in EC2
-    runs-on: ${{ github.event.inputs.runner_name }}
-    if: ${{ !cancelled() }}
-    steps:
-      - name: Instance configuration used
-        run: |
-          echo "IDs: ${{ inputs.instance_id }}"
-          echo "AMI: ${{ inputs.instance_image_id }}"
-          echo "Type: ${{ inputs.instance_type }}"
-          echo "Request ID: ${{ inputs.request_id }}"
-
-      - name: Get benchmark date
-        run: |
-          echo "BENCH_DATE=$(date --iso-8601=seconds)" >> "${GITHUB_ENV}"
-
-      - name: Checkout tfhe-rs repo with tags
-        uses: actions/checkout@692973e3d937129bcbf40652eb9f2f61becf3332
-        with:
-          fetch-depth: 0
-
-      - name: Set up home
-        # "Install rust" step require root user to have a HOME directory which is not set.
-        run: |
-          echo "HOME=/home/ubuntu" >> "${GITHUB_ENV}"
-
-      - name: Install rust
-        uses: dtolnay/rust-toolchain@21dc36fb71dd22e3317045c0c31a3f4249868b17
-        with:
-          toolchain: nightly
-
-      - name: Run benchmarks with AVX512
-        run: |
-          make bench_pbs
-          make bench_pbs128
-          make bench_ks
-
-      - 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 }} \
-          --database tfhe_rs \
-          --hardware ${{ inputs.instance_type }} \
-          --project-version "${COMMIT_HASH}" \
-          --branch ${{ github.ref_name }} \
-          --commit-date "${COMMIT_DATE}" \
-          --bench-date "${{ env.BENCH_DATE }}" \
-          --name-suffix avx512 \
-          --walk-subdirs \
-          --throughput
-
-      - name: Upload parsed results artifact
-        uses: actions/upload-artifact@65462800fd760344b1a7b4382951275a0abb4808
-        with:
-          name: ${{ github.sha }}_core_crypto
-          path: ${{ env.RESULTS_FILENAME }}
-
-      - name: Checkout Slab repo
-        uses: actions/checkout@692973e3d937129bcbf40652eb9f2f61becf3332
-        with:
-          repository: zama-ai/slab
-          path: slab
-          token: ${{ secrets.FHE_ACTIONS_TOKEN }}
-
-      - name: Send data to Slab
-        shell: bash
-        run: |
-          echo "Computing HMac on downloaded artifact"
-          SIGNATURE="$(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 }}
-
-      - name: Slack Notification
-        if: ${{ failure() }}
-        continue-on-error: true
-        uses: rtCamp/action-slack-notify@4e5fb42d249be6a45a298f3c9543b111b02f7907
-        env:
-          SLACK_COLOR: ${{ job.status }}
-          SLACK_CHANNEL: ${{ secrets.SLACK_CHANNEL }}
-          SLACK_ICON: https://pbs.twimg.com/profile_images/1274014582265298945/OjBKP9kn_400x400.png
-          SLACK_MESSAGE: "PBS benchmarks finished with status: ${{ job.status }}. (${{ env.ACTION_RUN_URL }})"
-          SLACK_USERNAME: ${{ secrets.BOT_USERNAME }}
-          SLACK_WEBHOOK: ${{ secrets.SLACK_WEBHOOK }}

.github/workflows/csprng_randomness_tests.yml

@@ -27,7 +27,7 @@ jobs:
     steps:
       - name: Start instance
         id: start-instance
-        uses: zama-ai/slab-github-runner@58f2cae4bf2c0b6728083f5f009b6dc0eb6dc3ac
+        uses: zama-ai/slab-github-runner@c0e7168795bd78f61f61146951ed9d0c73c9b701
         with:
           mode: start
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}
@@ -45,16 +45,17 @@ jobs:
     runs-on: ${{ needs.setup-instance.outputs.runner-name }}
     steps:
       - name: Checkout tfhe-rs
-        uses: actions/checkout@692973e3d937129bcbf40652eb9f2f61becf3332
+        uses: actions/checkout@d632683dd7b4114ad314bca15554477dd762a938
         with:
           persist-credentials: 'false'
+          token: ${{ secrets.FHE_ACTIONS_TOKEN }}
 
       - name: Set up home
         run: |
           echo "HOME=/home/ubuntu" >> "${GITHUB_ENV}"
 
       - name: Install latest stable
-        uses: dtolnay/rust-toolchain@21dc36fb71dd22e3317045c0c31a3f4249868b17
+        uses: dtolnay/rust-toolchain@7b1c307e0dcbda6122208f10795a713336a9b35a
         with:
           toolchain: stable
 
@@ -78,7 +79,7 @@ jobs:
     steps:
       - name: Stop instance
         id: stop-instance
-        uses: zama-ai/slab-github-runner@58f2cae4bf2c0b6728083f5f009b6dc0eb6dc3ac
+        uses: zama-ai/slab-github-runner@c0e7168795bd78f61f61146951ed9d0c73c9b701
         with:
           mode: stop
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}

.github/workflows/data_pr_close.yml

@@ -0,0 +1,123 @@
+name: Close or Merge corresponding PR on the data repo
+
+# When a PR with the data_PR tag is closed or merged, this will close the corresponding PR in the data repo.
+
+env:
+  TARGET_REPO_API_URL: ${{ github.api_url }}/repos/zama-ai/tfhe-backward-compat-data
+  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 }}
+  PR_BRANCH: ${{ github.head_ref || github.ref_name }}
+  CLOSE_TYPE: ${{ github.event.pull_request.merged && 'merge' || 'close' }}
+
+# only trigger on pull request closed events
+on:
+  pull_request:
+    types: [ closed ]
+
+# The same pattern is used for jobs that use the github api:
+# - save the result of the API call in the env var "GH_API_RES". Since the var is multiline
+# we use this trick: https://docs.github.com/en/actions/using-workflows/workflow-commands-for-github-actions#example-of-a-multiline-string
+# - "set +e" will make sure we reach the last "echo EOF" even in case of error
+# - "set -o" pipefail makes one line piped command return the error of the first failure
+# - 'RES="$?"' and 'exit $RES' are used to return the error code if a command failed. Without it, with "set +e"
+# the script will always return 0 because of the "echo EOF".
+
+
+jobs:
+  auto_close_job:
+    if: ${{ contains(github.event.pull_request.labels.*.name, 'data_PR') }}
+    runs-on: ubuntu-latest
+    steps:
+    - name: Find corresponding Pull Request in the data repo
+      run: |
+        {
+          set +e
+          set -o pipefail
+          echo 'TARGET_REPO_PR<<EOF'
+          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 }}:${{ env.PR_BRANCH }} | jq -e '.[0]' | sed 's/null/{ "message": "corresponding PR not found" }/'
+          RES="$?"
+          echo EOF
+        } >> "${GITHUB_ENV}"
+        exit $RES
+
+    - name: Comment on the PR to indicate the reason of the close
+      run: |
+        {
+          set +e
+          set -o pipefail
+          echo 'GH_API_RES<<EOF'
+          curl --fail-with-body --no-progress-meter -L -X POST \
+          -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  }}" }'
+          RES="$?"
+          echo EOF
+        } >> "${GITHUB_ENV}"
+        exit $RES
+
+    - name: Merge the Pull Request in the data repo
+      if: ${{ github.event.pull_request.merged }}
+      run: |
+        {
+          set +e
+          set -o pipefail
+          echo 'GH_API_RES<<EOF'
+          curl --fail-with-body --no-progress-meter -L -X PUT \
+          -H "Accept: application/vnd.github+json" \
+          -H "Authorization: Bearer ${{ secrets.FHE_ACTIONS_TOKEN }}" \
+          -H "X-GitHub-Api-Version: 2022-11-28" \
+          ${{ fromJson(env.TARGET_REPO_PR).url }}/merge \
+          -d '{ "merge_method": "rebase" }'
+          RES="$?"
+          echo EOF
+        } >> "${GITHUB_ENV}"
+        exit $RES
+
+    - name: Close the Pull Request in the data repo
+      if: ${{ !github.event.pull_request.merged }}
+      run: |
+        {
+          set +e
+          set -o pipefail
+          echo 'GH_API_RES<<EOF'
+          curl --fail-with-body --no-progress-meter -L -X PATCH \
+          -H "Accept: application/vnd.github+json" \
+          -H "Authorization: Bearer ${{ secrets.FHE_ACTIONS_TOKEN }}" \
+          -H "X-GitHub-Api-Version: 2022-11-28" \
+          ${{ fromJson(env.TARGET_REPO_PR).url }} \
+          -d '{ "state": "closed" }'
+          RES="$?"
+          echo EOF
+        } >> "${GITHUB_ENV}"
+        exit $RES
+
+    - name: Delete the associated branch in the data repo
+      run: |
+        {
+          set +e
+          set -o pipefail
+          echo 'GH_API_RES<<EOF'
+          curl --fail-with-body --no-progress-meter -L -X DELETE \
+          -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 }}
+          RES="$?"
+          echo EOF
+        } >> "${GITHUB_ENV}"
+        exit $RES
+
+    - name: Slack Notification
+      if: ${{ always() && job.status == 'failure' }}
+      continue-on-error: true
+      uses: rtCamp/action-slack-notify@4e5fb42d249be6a45a298f3c9543b111b02f7907
+      env:
+        SLACK_COLOR: ${{ job.status }}
+        SLACK_MESSAGE: "Failed to auto-${{ env.CLOSE_TYPE }} PR on data repo: ${{ fromJson(env.GH_API_RES || env.TARGET_REPO_PR).message }}"

.github/workflows/gpu_4090_tests.yml

@@ -17,11 +17,16 @@ on:
   workflow_dispatch:
   pull_request:
     types: [ labeled ]
+  schedule:
+    # Nightly tests @ 1AM after each work day
+    - cron: "0 1 * * MON-FRI"
 
 jobs:
   cuda-tests-linux:
     name: CUDA tests (RTX 4090)
-    if: ${{ github.event_name == 'workflow_dispatch' || contains(github.event.label.name, '4090_test') }}
+    if: github.event_name == 'workflow_dispatch' ||
+      contains(github.event.label.name, '4090_test') ||
+      (github.event_name == 'schedule' &&  github.repository == 'zama-ai/tfhe-rs')
     concurrency:
       group: ${{ github.workflow }}_${{ github.ref }}
       cancel-in-progress: true
@@ -29,12 +34,13 @@ jobs:
 
     steps:
       - name: Checkout tfhe-rs
-        uses: actions/checkout@692973e3d937129bcbf40652eb9f2f61becf3332
+        uses: actions/checkout@d632683dd7b4114ad314bca15554477dd762a938
         with:
           persist-credentials: 'false'
+          token: ${{ secrets.FHE_ACTIONS_TOKEN }}
 
       - name: Install latest stable
-        uses: dtolnay/rust-toolchain@21dc36fb71dd22e3317045c0c31a3f4249868b17
+        uses: dtolnay/rust-toolchain@7b1c307e0dcbda6122208f10795a713336a9b35a
         with:
           toolchain: stable
 
@@ -69,7 +75,7 @@ jobs:
           github_token: ${{ secrets.GITHUB_TOKEN }}
 
       - name: Slack Notification
-        if: ${{ always() }}
+        if: ${{ failure() }}
         continue-on-error: true
         uses: rtCamp/action-slack-notify@4e5fb42d249be6a45a298f3c9543b111b02f7907
         env:

.github/workflows/gpu_fast_h100_tests.yml

@@ -0,0 +1,201 @@
+# Compile and test tfhe-cuda-backend on an H100 VM on hyperstack
+name: TFHE Cuda Backend - Fast tests on H100
+
+env:
+  CARGO_TERM_COLOR: always
+  ACTION_RUN_URL: ${{ github.server_url }}/${{ github.repository }}/actions/runs/${{ github.run_id }}
+  RUSTFLAGS: "-C target-cpu=native"
+  RUST_BACKTRACE: "full"
+  RUST_MIN_STACK: "8388608"
+  SLACK_CHANNEL: ${{ secrets.SLACK_CHANNEL }}
+  SLACK_ICON: https://pbs.twimg.com/profile_images/1274014582265298945/OjBKP9kn_400x400.png
+  SLACK_USERNAME: ${{ secrets.BOT_USERNAME }}
+  SLACK_WEBHOOK: ${{ secrets.SLACK_WEBHOOK }}
+  IS_PULL_REQUEST: ${{ github.event_name == 'pull_request' }}
+
+on:
+  # Allows you to run this workflow manually from the Actions tab as an alternative.
+  workflow_dispatch:
+  pull_request:
+      types: [ labeled ]
+
+jobs:
+  should-run:
+    runs-on: ubuntu-latest
+    permissions:
+      pull-requests: write
+    outputs:
+      gpu_test: ${{ env.IS_PULL_REQUEST == 'false' || steps.changed-files.outputs.gpu_any_changed }}
+    steps:
+      - name: Checkout tfhe-rs
+        uses: actions/checkout@d632683dd7b4114ad314bca15554477dd762a938
+        with:
+          fetch-depth: 0
+
+      - name: Check for file changes
+        id: changed-files
+        uses: tj-actions/changed-files@c3a1bb2c992d77180ae65be6ae6c166cf40f857c
+        with:
+          since_last_remote_commit: true
+          files_yaml: |
+            gpu:
+              - tfhe/Cargo.toml
+              - tfhe/build.rs
+              - backends/tfhe-cuda-backend/**
+              - tfhe/src/core_crypto/gpu/**
+              - tfhe/src/integer/gpu/**
+              - tfhe/src/shortint/parameters/**
+              - tfhe/src/high_level_api/**
+              - tfhe/src/c_api/**
+              - 'tfhe/docs/**.md'
+              - Makefile
+              - '.github/workflows/gpu_fast_h100_tests.yml'
+              - scripts/**
+              - ci/**
+
+  setup-instance:
+    name: Setup instance (cuda-h100-tests)
+    needs: should-run
+    if: github.event_name != 'pull_request' ||
+      (github.event.action != 'labeled' && needs.should-run.outputs.gpu_test == 'true') ||
+      (github.event.action == 'labeled' && github.event.label.name == 'approved' && needs.should-run.outputs.gpu_test == 'true')
+    runs-on: ubuntu-latest
+    outputs:
+      runner-name: ${{ steps.start-instance.outputs.label }}
+    steps:
+      - name: Start instance
+        id: start-instance
+        uses: zama-ai/slab-github-runner@c0e7168795bd78f61f61146951ed9d0c73c9b701
+        with:
+          mode: start
+          github-token: ${{ secrets.SLAB_ACTION_TOKEN }}
+          slab-url: ${{ secrets.SLAB_BASE_URL }}
+          job-secret: ${{ secrets.JOB_SECRET }}
+          backend: hyperstack
+          profile: single-h100
+
+  cuda-tests-linux:
+    name: CUDA H100 tests
+    needs: [ should-run, setup-instance ]
+    if: github.event_name != 'pull_request' ||
+      (github.event_name == 'pull_request' && needs.setup-instance.result != 'skipped')
+    concurrency:
+      group: ${{ github.workflow }}_${{ github.ref }}
+      cancel-in-progress: ${{ github.ref != 'refs/heads/main' }}
+    runs-on: ${{ needs.setup-instance.outputs.runner-name }}
+    strategy:
+      fail-fast: false
+      # explicit include-based build matrix, of known valid options
+      matrix:
+        include:
+          - os: ubuntu-22.04
+            cuda: "12.2"
+            gcc: 11 
+    env:
+      CUDA_PATH: /usr/local/cuda-${{ matrix.cuda }}
+      CMAKE_VERSION: 3.29.6
+    steps:
+      # Mandatory on hyperstack since a bootable volume is not re-usable yet.
+      - name: Install dependencies
+        run: |
+          sudo apt update
+          sudo apt install -y checkinstall zlib1g-dev libssl-dev 
+          wget https://github.com/Kitware/CMake/releases/download/v${{ env.CMAKE_VERSION }}/cmake-${{ env.CMAKE_VERSION }}.tar.gz
+          tar -zxvf cmake-${{ env.CMAKE_VERSION }}.tar.gz
+          cd cmake-${{ env.CMAKE_VERSION }}
+          ./bootstrap
+          make -j"$(nproc)"
+          sudo make install
+
+      - name: Checkout tfhe-rs
+        uses: actions/checkout@d632683dd7b4114ad314bca15554477dd762a938
+        with:
+          persist-credentials: 'false'
+          token: ${{ secrets.FHE_ACTIONS_TOKEN }}
+
+      - name: Set up home
+        run: |
+          echo "HOME=/home/ubuntu" >> "${GITHUB_ENV}"
+
+      - name: Install latest stable
+        uses: dtolnay/rust-toolchain@7b1c307e0dcbda6122208f10795a713336a9b35a
+        with:
+          toolchain: stable
+
+      - name: Export CUDA variables
+        if: ${{ !cancelled() }}
+        run: |
+          echo "CUDA_PATH=$CUDA_PATH" >> "${GITHUB_ENV}"
+          echo "$CUDA_PATH/bin" >> "${GITHUB_PATH}"
+          echo "LD_LIBRARY_PATH=$CUDA_PATH/lib:$LD_LIBRARY_PATH" >> "${GITHUB_ENV}"
+          echo "CUDACXX=/usr/local/cuda-${{ matrix.cuda }}/bin/nvcc" >> "${GITHUB_ENV}"
+
+      # Specify the correct host compilers
+      - name: Export gcc and g++ variables
+        if: ${{ !cancelled() }}
+        run: |
+          {
+            echo "CC=/usr/bin/gcc-${{ matrix.gcc }}";
+            echo "CXX=/usr/bin/g++-${{ matrix.gcc }}";
+            echo "CUDAHOSTCXX=/usr/bin/g++-${{ matrix.gcc }}";
+            echo "HOME=/home/ubuntu";
+          } >> "${GITHUB_ENV}"
+
+      - name: Check device is detected
+        if: ${{ !cancelled() }}
+        run: nvidia-smi
+
+      - name: Run core crypto and internal CUDA backend tests
+        run: |
+          BIG_TESTS_INSTANCE=TRUE make test_core_crypto_gpu
+          BIG_TESTS_INSTANCE=TRUE make test_integer_compression_gpu
+          BIG_TESTS_INSTANCE=TRUE make test_cuda_backend
+
+      - name: Run user docs tests
+        run: |
+          BIG_TESTS_INSTANCE=TRUE make test_user_doc_gpu
+
+      - name: Test C API
+        run: |
+          BIG_TESTS_INSTANCE=TRUE make test_c_api_gpu
+
+      - name: Run High Level API Tests
+        run: |
+          BIG_TESTS_INSTANCE=TRUE make test_high_level_api_gpu
+
+  slack-notify:
+    name: Slack Notification
+    needs: [ setup-instance, cuda-tests-linux ]
+    runs-on: ubuntu-latest
+    if: ${{ always() && needs.cuda-tests-linux.result != 'skipped' && failure() }}
+    continue-on-error: true
+    steps:
+      - name: Send message
+        uses: rtCamp/action-slack-notify@4e5fb42d249be6a45a298f3c9543b111b02f7907
+        env:
+          SLACK_COLOR: ${{ needs.cuda-tests-linux.result }}
+          SLACK_MESSAGE: "Fast H100 tests finished with status: ${{ needs.cuda-tests-linux.result }}. (${{ env.ACTION_RUN_URL }})"
+
+  teardown-instance:
+    name: Teardown instance (cuda-h100-tests)
+    if: ${{ always() && needs.setup-instance.result != 'skipped' }}
+    needs: [ setup-instance, cuda-tests-linux ]
+    runs-on: ubuntu-latest
+    steps:
+      - name: Stop instance
+        id: stop-instance
+        uses: zama-ai/slab-github-runner@c0e7168795bd78f61f61146951ed9d0c73c9b701
+        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@4e5fb42d249be6a45a298f3c9543b111b02f7907
+        env:
+          SLACK_COLOR: ${{ job.status }}
+          SLACK_MESSAGE: "Instance teardown (cuda-h100-tests) finished with status: ${{ job.status }}. (${{ env.ACTION_RUN_URL }})"

.github/workflows/gpu_fast_tests.yml

@@ -1,5 +1,5 @@
 # Compile and test tfhe-cuda-backend on an AWS instance
-name: TFHE Cuda Backend - Full tests
+name: TFHE Cuda Backend - Fast tests
 
 env:
   CARGO_TERM_COLOR: always
@@ -11,6 +11,7 @@ env:
   SLACK_ICON: https://pbs.twimg.com/profile_images/1274014582265298945/OjBKP9kn_400x400.png
   SLACK_USERNAME: ${{ secrets.BOT_USERNAME }}
   SLACK_WEBHOOK: ${{ secrets.SLACK_WEBHOOK }}
+  IS_PULL_REQUEST: ${{ github.event_name == 'pull_request' }}
 
 on:
   # Allows you to run this workflow manually from the Actions tab as an alternative.
@@ -18,94 +19,64 @@ on:
   pull_request:
 
 jobs:
+  should-run:
+    runs-on: ubuntu-latest
+    permissions:
+      pull-requests: write
+    outputs:
+      gpu_test: ${{ env.IS_PULL_REQUEST == 'false' || steps.changed-files.outputs.gpu_any_changed }}
+    steps:
+      - name: Checkout tfhe-rs
+        uses: actions/checkout@d632683dd7b4114ad314bca15554477dd762a938
+        with:
+          fetch-depth: 0
+
+      - name: Check for file changes
+        id: changed-files
+        uses: tj-actions/changed-files@c3a1bb2c992d77180ae65be6ae6c166cf40f857c
+        with:
+          since_last_remote_commit: true
+          files_yaml: |
+            gpu:
+              - tfhe/Cargo.toml
+              - tfhe/build.rs
+              - backends/tfhe-cuda-backend/**
+              - tfhe/src/core_crypto/gpu/**
+              - tfhe/src/integer/gpu/**
+              - tfhe/src/shortint/parameters/**
+              - tfhe/src/high_level_api/**
+              - tfhe/src/c_api/**
+              - 'tfhe/docs/**.md'
+              - '.github/workflows/gpu_fast_tests.yml'
+              - Makefile
+              - scripts/**
+              - ci/**
+
   setup-instance:
     name: Setup instance (cuda-tests)
+    needs: should-run
+    if: github.event_name != 'pull_request' ||
+      needs.should-run.outputs.gpu_test == 'true'
     runs-on: ubuntu-latest
     outputs:
       runner-name: ${{ steps.start-instance.outputs.label }}
     steps:
       - name: Start instance
         id: start-instance
-        uses: zama-ai/slab-github-runner@58f2cae4bf2c0b6728083f5f009b6dc0eb6dc3ac
+        uses: zama-ai/slab-github-runner@c0e7168795bd78f61f61146951ed9d0c73c9b701
         with:
           mode: start
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}
           slab-url: ${{ secrets.SLAB_BASE_URL }}
           job-secret: ${{ secrets.JOB_SECRET }}
-          backend: aws
+          backend: hyperstack
           profile: gpu-test
 
-  cuda-pcc:
-    name: CUDA post-commit checks
-    needs: setup-instance
-    concurrency:
-      group: ${{ github.workflow }}_${{ github.ref }}
-      cancel-in-progress: ${{ github.ref != 'refs/heads/main' }}
-    runs-on: ${{ needs.setup-instance.outputs.runner-name }}
-    strategy:
-      fail-fast: false
-      # explicit include-based build matrix, of known valid options
-      matrix:
-        include:
-          - os: ubuntu-22.04
-            cuda: "12.2"
-            gcc: 9
-    env:
-      CUDA_PATH: /usr/local/cuda-${{ matrix.cuda }}
-
-    steps:
-      - name: Checkout tfhe-rs
-        uses: actions/checkout@692973e3d937129bcbf40652eb9f2f61becf3332
-        with:
-          persist-credentials: 'false'
-
-      - name: Set up home
-        run: |
-          echo "HOME=/home/ubuntu" >> "${GITHUB_ENV}"
-
-      - name: Install latest stable
-        uses: dtolnay/rust-toolchain@21dc36fb71dd22e3317045c0c31a3f4249868b17
-        with:
-          toolchain: stable
-
-      - name: Export CUDA variables
-        if: ${{ !cancelled() }}
-        run: |
-          echo "CUDA_PATH=$CUDA_PATH" >> "${GITHUB_ENV}"
-          echo "$CUDA_PATH/bin" >> "${GITHUB_PATH}"
-          echo "LD_LIBRARY_PATH=$CUDA_PATH/lib:$LD_LIBRARY_PATH" >> "${GITHUB_ENV}"
-          echo "CUDACXX=/usr/local/cuda-${{ matrix.cuda }}/bin/nvcc" >> "${GITHUB_ENV}"
-
-      # Specify the correct host compilers
-      - name: Export gcc and g++ variables
-        if: ${{ !cancelled() }}
-        run: |
-          {
-            echo "CC=/usr/bin/gcc-${{ matrix.gcc }}";
-            echo "CXX=/usr/bin/g++-${{ matrix.gcc }}";
-            echo "CUDAHOSTCXX=/usr/bin/g++-${{ matrix.gcc }}";
-            echo "HOME=/home/ubuntu";
-          } >> "${GITHUB_ENV}"
-
-      - name: Run fmt checks
-        run: |
-          make check_fmt_gpu
-
-      - name: Run clippy checks
-        run: |
-          make pcc_gpu
-
-      - name: Slack Notification
-        if: ${{ always() }}
-        continue-on-error: true
-        uses: rtCamp/action-slack-notify@4e5fb42d249be6a45a298f3c9543b111b02f7907
-        env:
-          SLACK_COLOR: ${{ job.status }}
-          SLACK_MESSAGE: "CUDA AWS post-commit checks finished with status: ${{ job.status }}. (${{ env.ACTION_RUN_URL }})"
-
   cuda-tests-linux:
     name: CUDA tests
-    needs: [ setup-instance, cuda-pcc ]
+    needs: [ should-run, setup-instance ]
+    if: github.event_name != 'pull_request' ||
+      (github.event_name == 'pull_request' && needs.setup-instance.result != 'skipped')
     concurrency:
       group: ${{ github.workflow }}_${{ github.ref }}
       cancel-in-progress: ${{ github.ref != 'refs/heads/main' }}
@@ -117,22 +88,35 @@ jobs:
         include:
           - os: ubuntu-22.04
             cuda: "12.2"
-            gcc: 9
+            gcc: 11 
     env:
       CUDA_PATH: /usr/local/cuda-${{ matrix.cuda }}
-
+      CMAKE_VERSION: 3.29.6
     steps:
+      # Mandatory on hyperstack since a bootable volume is not re-usable yet.
+      - name: Install dependencies
+        run: |
+          sudo apt update
+          sudo apt install -y checkinstall zlib1g-dev libssl-dev 
+          wget https://github.com/Kitware/CMake/releases/download/v${{ env.CMAKE_VERSION }}/cmake-${{ env.CMAKE_VERSION }}.tar.gz
+          tar -zxvf cmake-${{ env.CMAKE_VERSION }}.tar.gz
+          cd cmake-${{ env.CMAKE_VERSION }}
+          ./bootstrap
+          make -j"$(nproc)"
+          sudo make install
+
       - name: Checkout tfhe-rs
-        uses: actions/checkout@692973e3d937129bcbf40652eb9f2f61becf3332
+        uses: actions/checkout@d632683dd7b4114ad314bca15554477dd762a938
         with:
           persist-credentials: 'false'
+          token: ${{ secrets.FHE_ACTIONS_TOKEN }}
 
       - name: Set up home
         run: |
           echo "HOME=/home/ubuntu" >> "${GITHUB_ENV}"
 
       - name: Install latest stable
-        uses: dtolnay/rust-toolchain@21dc36fb71dd22e3317045c0c31a3f4249868b17
+        uses: dtolnay/rust-toolchain@7b1c307e0dcbda6122208f10795a713336a9b35a
         with:
           toolchain: stable
 
@@ -155,9 +139,15 @@ jobs:
             echo "HOME=/home/ubuntu";
           } >> "${GITHUB_ENV}"
 
-      - name: Run core crypto, integer and internal CUDA backend tests
+      - name: Check device is detected
+        if: ${{ !cancelled() }}
+        run: nvidia-smi
+
+      - name: Run core crypto and internal CUDA backend tests
         run: |
-          make test_gpu
+          make test_core_crypto_gpu
+          make test_integer_compression_gpu
+          make test_cuda_backend
 
       - name: Run user docs tests
         run: |
@@ -171,23 +161,28 @@ jobs:
         run: |
           make test_high_level_api_gpu
 
-      - name: Slack Notification
-        if: ${{ always() }}
-        continue-on-error: true
+  slack-notify:
+    name: Slack Notification
+    needs: [ setup-instance, cuda-tests-linux ]
+    runs-on: ubuntu-latest
+    if: ${{ always() && needs.cuda-tests-linux.result != 'skipped' && failure() }}
+    continue-on-error: true
+    steps:
+      - name: Send message
         uses: rtCamp/action-slack-notify@4e5fb42d249be6a45a298f3c9543b111b02f7907
         env:
-          SLACK_COLOR: ${{ job.status }}
-          SLACK_MESSAGE: "CUDA AWS tests finished with status: ${{ job.status }}. (${{ env.ACTION_RUN_URL }})"
+          SLACK_COLOR: ${{ needs.cuda-tests-linux.result }}
+          SLACK_MESSAGE: "Base GPU tests finished with status: ${{ needs.cuda-tests-linux.result }}. (${{ env.ACTION_RUN_URL }})"
 
   teardown-instance:
     name: Teardown instance (cuda-tests)
     if: ${{ always() && needs.setup-instance.result != 'skipped' }}
-    needs: [ setup-instance, cuda-pcc, cuda-tests-linux ]
+    needs: [ setup-instance, cuda-tests-linux ]
     runs-on: ubuntu-latest
     steps:
       - name: Stop instance
         id: stop-instance
-        uses: zama-ai/slab-github-runner@58f2cae4bf2c0b6728083f5f009b6dc0eb6dc3ac
+        uses: zama-ai/slab-github-runner@c0e7168795bd78f61f61146951ed9d0c73c9b701
         with:
           mode: stop
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}

.github/workflows/gpu_full_h100_tests.yml

@@ -11,11 +11,10 @@ env:
   SLACK_ICON: https://pbs.twimg.com/profile_images/1274014582265298945/OjBKP9kn_400x400.png
   SLACK_USERNAME: ${{ secrets.BOT_USERNAME }}
   SLACK_WEBHOOK: ${{ secrets.SLACK_WEBHOOK }}
+  IS_PULL_REQUEST: ${{ github.event_name == 'pull_request' }}
 
 on:
-  # Allows you to run this workflow manually from the Actions tab as an alternative.
   workflow_dispatch:
-  pull_request:
 
 jobs:
   setup-instance:
@@ -26,7 +25,7 @@ jobs:
     steps:
       - name: Start instance
         id: start-instance
-        uses: zama-ai/slab-github-runner@58f2cae4bf2c0b6728083f5f009b6dc0eb6dc3ac
+        uses: zama-ai/slab-github-runner@447a2d0fd2d1a9d647aa0d0723a6e9255372f261
         with:
           mode: start
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}
@@ -52,22 +51,13 @@ jobs:
             gcc: 11 
     env:
       CUDA_PATH: /usr/local/cuda-${{ matrix.cuda }}
-      CMAKE_VERSION: 3.29.1
+      CMAKE_VERSION: 3.29.6
     steps:
       # Mandatory on hyperstack since a bootable volume is not re-usable yet.
       - name: Install dependencies
         run: |
           sudo apt update
-          sudo apt install ca-certificates curl
-          sudo install -m 0755 -d /etc/apt/keyrings
-          sudo curl -fsSL https://download.docker.com/linux/ubuntu/gpg -o /etc/apt/keyrings/docker.asc
-          sudo chmod a+r /etc/apt/keyrings/docker.asc
-          echo \
-          "deb [arch=$(dpkg --print-architecture) signed-by=/etc/apt/keyrings/docker.asc] https://download.docker.com/linux/ubuntu \
-           $(. /etc/os-release && echo "$VERSION_CODENAME") stable" | \
-          sudo tee /etc/apt/sources.list.d/docker.list > /dev/null
-          sudo apt update
-          sudo apt install -y checkinstall zlib1g-dev libssl-dev docker-ce docker-ce-cli containerd.io docker-buildx-plugin docker-compose-plugin
+          sudo apt install -y checkinstall zlib1g-dev libssl-dev 
           wget https://github.com/Kitware/CMake/releases/download/v${{ env.CMAKE_VERSION }}/cmake-${{ env.CMAKE_VERSION }}.tar.gz
           tar -zxvf cmake-${{ env.CMAKE_VERSION }}.tar.gz
           cd cmake-${{ env.CMAKE_VERSION }}
@@ -76,14 +66,17 @@ jobs:
           sudo make install
 
       - name: Checkout tfhe-rs
-        uses: actions/checkout@692973e3d937129bcbf40652eb9f2f61becf3332
+        uses: actions/checkout@d632683dd7b4114ad314bca15554477dd762a938
+        with:
+          persist-credentials: 'false'
+          token: ${{ secrets.FHE_ACTIONS_TOKEN }}
 
       - name: Set up home
         run: |
           echo "HOME=/home/ubuntu" >> "${GITHUB_ENV}"
 
       - name: Install latest stable
-        uses: dtolnay/rust-toolchain@21dc36fb71dd22e3317045c0c31a3f4249868b17
+        uses: dtolnay/rust-toolchain@7b1c307e0dcbda6122208f10795a713336a9b35a
         with:
           toolchain: stable
 
@@ -106,6 +99,10 @@ jobs:
             echo "HOME=/home/ubuntu";
           } >> "${GITHUB_ENV}"
 
+      - name: Check device is detected
+        if: ${{ !cancelled() }}
+        run: nvidia-smi
+
       - name: Run core crypto, integer and internal CUDA backend tests
         run: |
           make test_gpu
@@ -125,25 +122,24 @@ jobs:
   slack-notify:
     name: Slack Notification
     needs: [ setup-instance, cuda-tests-linux ]
-    runs-on: ${{ needs.setup-instance.outputs.runner-name }}
-    if: ${{ !success() && !cancelled() }}
+    runs-on: ubuntu-latest
+    if: ${{ failure() }}
     continue-on-error: true
     steps:
       - name: Send message
         uses: rtCamp/action-slack-notify@4e5fb42d249be6a45a298f3c9543b111b02f7907
         env:
           SLACK_COLOR: ${{ needs.cuda-tests-linux.result }}
-          SLACK_MESSAGE: "Integer GPU H100 tests finished with status: ${{ needs.cuda-tests-linux.result }}. (${{ env.ACTION_RUN_URL }})"
+          SLACK_MESSAGE: "Full H100 tests finished with status: ${{ needs.cuda-tests-linux.result }}. (${{ env.ACTION_RUN_URL }})"
 
   teardown-instance:
     name: Teardown instance (cuda-h100-tests)
-    if: ${{ always() && needs.setup-instance.result != 'skipped' }}
     needs: [ setup-instance, cuda-tests-linux ]
     runs-on: ubuntu-latest
     steps:
       - name: Stop instance
         id: stop-instance
-        uses: zama-ai/slab-github-runner@58f2cae4bf2c0b6728083f5f009b6dc0eb6dc3ac
+        uses: zama-ai/slab-github-runner@447a2d0fd2d1a9d647aa0d0723a6e9255372f261
         with:
           mode: stop
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}

.github/workflows/gpu_full_multi_gpu_tests.yml

@@ -0,0 +1,204 @@
+# Compile and test tfhe-cuda-backend on an AWS instance
+name: TFHE Cuda Backend - Full tests multi-GPU
+
+env:
+  CARGO_TERM_COLOR: always
+  ACTION_RUN_URL: ${{ github.server_url }}/${{ github.repository }}/actions/runs/${{ github.run_id }}
+  RUSTFLAGS: "-C target-cpu=native"
+  RUST_BACKTRACE: "full"
+  RUST_MIN_STACK: "8388608"
+  SLACK_CHANNEL: ${{ secrets.SLACK_CHANNEL }}
+  SLACK_ICON: https://pbs.twimg.com/profile_images/1274014582265298945/OjBKP9kn_400x400.png
+  SLACK_USERNAME: ${{ secrets.BOT_USERNAME }}
+  SLACK_WEBHOOK: ${{ secrets.SLACK_WEBHOOK }}
+  IS_PULL_REQUEST: ${{ github.event_name == 'pull_request' }}
+
+on:
+  # Allows you to run this workflow manually from the Actions tab as an alternative.
+  workflow_dispatch:
+  pull_request:
+    types: [ labeled ]
+
+jobs:
+  should-run:
+    runs-on: ubuntu-latest
+    permissions:
+      pull-requests: write
+    outputs:
+      gpu_test: ${{ env.IS_PULL_REQUEST == 'false' || steps.changed-files.outputs.gpu_any_changed }}
+    steps:
+      - name: Checkout tfhe-rs
+        uses: actions/checkout@d632683dd7b4114ad314bca15554477dd762a938
+        with:
+          fetch-depth: 0
+
+      - name: Check for file changes
+        id: changed-files
+        uses: tj-actions/changed-files@c3a1bb2c992d77180ae65be6ae6c166cf40f857c
+        with:
+          since_last_remote_commit: true
+          files_yaml: |
+            gpu:
+              - tfhe/Cargo.toml
+              - tfhe/build.rs
+              - backends/tfhe-cuda-backend/**
+              - tfhe/src/core_crypto/gpu/**
+              - tfhe/src/integer/gpu/**
+              - tfhe/src/shortint/parameters/**
+              - tfhe/src/high_level_api/**
+              - tfhe/src/c_api/**
+              - 'tfhe/docs/**.md'
+              - Makefile
+              - '.github/workflows/**_multi_gpu_tests.yml'
+              - scripts/**
+              - ci/**
+
+  setup-instance:
+    name: Setup instance (cuda-tests-multi-gpu)
+    needs: should-run
+    if: github.event_name != 'pull_request' ||
+      (github.event.action != 'labeled' && needs.should-run.outputs.gpu_test == 'true') ||
+      (github.event.action == 'labeled' && github.event.label.name == 'approved' && needs.should-run.outputs.gpu_test == 'true')
+    runs-on: ubuntu-latest
+    outputs:
+      runner-name: ${{ steps.start-instance.outputs.label }}
+    steps:
+      - name: Start instance
+        id: start-instance
+        uses: zama-ai/slab-github-runner@c0e7168795bd78f61f61146951ed9d0c73c9b701
+        with:
+          mode: start
+          github-token: ${{ secrets.SLAB_ACTION_TOKEN }}
+          slab-url: ${{ secrets.SLAB_BASE_URL }}
+          job-secret: ${{ secrets.JOB_SECRET }}
+          backend: hyperstack
+          profile: multi-gpu-test
+
+  cuda-tests-linux:
+    name: CUDA multi-GPU tests
+    needs: [ should-run, setup-instance ]
+    if: github.event_name != 'pull_request' ||
+      (github.event_name == 'pull_request' && needs.setup-instance.result != 'skipped')
+    concurrency:
+      group: ${{ github.workflow }}_${{ github.ref }}
+      cancel-in-progress: ${{ github.ref != 'refs/heads/main' }}
+    runs-on: ${{ needs.setup-instance.outputs.runner-name }}
+    strategy:
+      fail-fast: false
+      # explicit include-based build matrix, of known valid options
+      matrix:
+        include:
+          - os: ubuntu-22.04
+            cuda: "12.2"
+            gcc: 11 
+    env:
+      CUDA_PATH: /usr/local/cuda-${{ matrix.cuda }}
+      CMAKE_VERSION: 3.29.6
+    steps:
+      # Mandatory on hyperstack since a bootable volume is not re-usable yet.
+      - name: Install dependencies
+        run: |
+          sudo apt update
+          sudo apt install -y checkinstall zlib1g-dev libssl-dev 
+          wget https://github.com/Kitware/CMake/releases/download/v${{ env.CMAKE_VERSION }}/cmake-${{ env.CMAKE_VERSION }}.tar.gz
+          tar -zxvf cmake-${{ env.CMAKE_VERSION }}.tar.gz
+          cd cmake-${{ env.CMAKE_VERSION }}
+          ./bootstrap
+          make -j"$(nproc)"
+          sudo make install
+
+      - name: Checkout tfhe-rs
+        uses: actions/checkout@d632683dd7b4114ad314bca15554477dd762a938
+        with:
+          persist-credentials: 'false'
+          token: ${{ secrets.FHE_ACTIONS_TOKEN }}
+
+      - name: Set up home
+        run: |
+          echo "HOME=/home/ubuntu" >> "${GITHUB_ENV}"
+
+      - name: Install latest stable
+        uses: dtolnay/rust-toolchain@7b1c307e0dcbda6122208f10795a713336a9b35a
+        with:
+          toolchain: stable
+
+      - name: Export CUDA variables
+        if: ${{ !cancelled() }}
+        run: |
+          echo "CUDA_PATH=$CUDA_PATH" >> "${GITHUB_ENV}"
+          echo "$CUDA_PATH/bin" >> "${GITHUB_PATH}"
+          echo "LD_LIBRARY_PATH=$CUDA_PATH/lib:$LD_LIBRARY_PATH" >> "${GITHUB_ENV}"
+          echo "CUDACXX=/usr/local/cuda-${{ matrix.cuda }}/bin/nvcc" >> "${GITHUB_ENV}"
+
+      # Specify the correct host compilers
+      - name: Export gcc and g++ variables
+        if: ${{ !cancelled() }}
+        run: |
+          {
+            echo "CC=/usr/bin/gcc-${{ matrix.gcc }}";
+            echo "CXX=/usr/bin/g++-${{ matrix.gcc }}";
+            echo "CUDAHOSTCXX=/usr/bin/g++-${{ matrix.gcc }}";
+            echo "HOME=/home/ubuntu";
+          } >> "${GITHUB_ENV}"
+
+      - name: Check device is detected
+        if: ${{ !cancelled() }}
+        run: nvidia-smi
+
+      - name: Run multi-bit CUDA integer compression tests
+        run: |
+          BIG_TESTS_INSTANCE=TRUE make test_integer_compression_gpu
+
+      # No need to test core_crypto and classic PBS in integer since it's already tested on single GPU.
+      - name: Run multi-bit CUDA integer tests
+        run: |
+          BIG_TESTS_INSTANCE=TRUE make test_integer_multi_bit_gpu_ci
+
+      - name: Run user docs tests
+        run: |
+          BIG_TESTS_INSTANCE=TRUE make test_user_doc_gpu
+
+      - name: Test C API
+        run: |
+          BIG_TESTS_INSTANCE=TRUE make test_c_api_gpu
+
+      - name: Run High Level API Tests
+        run: |
+          BIG_TESTS_INSTANCE=TRUE make test_high_level_api_gpu
+
+  slack-notify:
+    name: Slack Notification
+    needs: [ setup-instance, cuda-tests-linux ]
+    runs-on: ubuntu-latest
+    if: ${{ always() && needs.cuda-tests-linux.result != 'skipped' && failure() }}
+    continue-on-error: true
+    steps:
+      - name: Send message
+        uses: rtCamp/action-slack-notify@4e5fb42d249be6a45a298f3c9543b111b02f7907
+        env:
+          SLACK_COLOR: ${{ needs.cuda-tests-linux.result }}
+          SLACK_MESSAGE: "Multi-GPU tests finished with status: ${{ needs.cuda-tests-linux.result }}. (${{ env.ACTION_RUN_URL }})"
+
+  teardown-instance:
+    name: Teardown instance (cuda-tests-multi-gpu)
+    if: ${{ always() && needs.setup-instance.result != 'skipped' }}
+    needs: [ setup-instance, cuda-tests-linux ]
+    runs-on: ubuntu-latest
+    steps:
+      - name: Stop instance
+        id: stop-instance
+        uses: zama-ai/slab-github-runner@c0e7168795bd78f61f61146951ed9d0c73c9b701
+        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@4e5fb42d249be6a45a298f3c9543b111b02f7907
+        env:
+          SLACK_COLOR: ${{ job.status }}
+          SLACK_MESSAGE: "Instance teardown (cuda-tests-multi-gpu) finished with status: ${{ job.status }}. (${{ env.ACTION_RUN_URL }})"

.github/workflows/gpu_pcc.yml

@@ -1,5 +1,5 @@
-# Compile and test tfhe-cuda-backend on an AWS instance
-name: TFHE Cuda Backend - Full tests multi-GPU
+# Perfom tfhe-cuda-backend post-commit checks on an AWS instance
+name: TFHE Cuda Backend - Post-commit Checks
 
 env:
   CARGO_TERM_COLOR: always
@@ -13,34 +13,32 @@ env:
   SLACK_WEBHOOK: ${{ secrets.SLACK_WEBHOOK }}
 
 on:
-  # Allows you to run this workflow manually from the Actions tab as an alternative.
-  workflow_dispatch:
   pull_request:
 
 jobs:
   setup-instance:
-    name: Setup instance (cuda-tests-multi-gpu)
+    name: Setup instance (cuda-pcc)
     runs-on: ubuntu-latest
     outputs:
       runner-name: ${{ steps.start-instance.outputs.label }}
     steps:
       - name: Start instance
         id: start-instance
-        uses: zama-ai/slab-github-runner@58f2cae4bf2c0b6728083f5f009b6dc0eb6dc3ac
+        uses: zama-ai/slab-github-runner@c0e7168795bd78f61f61146951ed9d0c73c9b701
         with:
           mode: start
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}
           slab-url: ${{ secrets.SLAB_BASE_URL }}
           job-secret: ${{ secrets.JOB_SECRET }}
           backend: aws
-          profile: multi-gpu-test
+          profile: gpu-build
 
-  cuda-tests-linux:
-    name: CUDA multi-GPU tests
-    needs: [ setup-instance ]
+  cuda-pcc:
+    name: CUDA post-commit checks
+    needs: setup-instance
     concurrency:
       group: ${{ github.workflow }}_${{ github.ref }}
-      cancel-in-progress: ${{ github.ref != 'refs/heads/main' }}
+      cancel-in-progress: true
     runs-on: ${{ needs.setup-instance.outputs.runner-name }}
     strategy:
       fail-fast: false
@@ -55,14 +53,17 @@ jobs:
 
     steps:
       - name: Checkout tfhe-rs
-        uses: actions/checkout@692973e3d937129bcbf40652eb9f2f61becf3332
+        uses: actions/checkout@d632683dd7b4114ad314bca15554477dd762a938
+        with:
+          persist-credentials: 'false'
+          token: ${{ secrets.FHE_ACTIONS_TOKEN }}
 
       - name: Set up home
         run: |
           echo "HOME=/home/ubuntu" >> "${GITHUB_ENV}"
 
       - name: Install latest stable
-        uses: dtolnay/rust-toolchain@21dc36fb71dd22e3317045c0c31a3f4249868b17
+        uses: dtolnay/rust-toolchain@7b1c307e0dcbda6122208f10795a713336a9b35a
         with:
           toolchain: stable
 
@@ -85,39 +86,31 @@ jobs:
             echo "HOME=/home/ubuntu";
           } >> "${GITHUB_ENV}"
 
-      - name: Run core crypto, integer and internal CUDA backend tests
+      - name: Run fmt checks
         run: |
-          make test_gpu
+          make check_fmt_gpu
 
-      - name: Run user docs tests
+      - name: Run clippy checks
         run: |
-          make test_user_doc_gpu
-
-      - name: Test C API
-        run: |
-          make test_c_api_gpu
-
-      - name: Run High Level API Tests
-        run: |
-          make test_high_level_api_gpu
+          make pcc_gpu
 
       - name: Slack Notification
-        if: ${{ always() }}
+        if: ${{ failure() }}
         continue-on-error: true
         uses: rtCamp/action-slack-notify@4e5fb42d249be6a45a298f3c9543b111b02f7907
         env:
           SLACK_COLOR: ${{ job.status }}
-          SLACK_MESSAGE: "CUDA AWS multi-GPU tests finished with status: ${{ job.status }}. (${{ env.ACTION_RUN_URL }})"
+          SLACK_MESSAGE: "CUDA AWS post-commit checks finished with status: ${{ job.status }}. (${{ env.ACTION_RUN_URL }})"
 
   teardown-instance:
-    name: Teardown instance (cuda-tests-multi-gpu)
+    name: Teardown instance (cuda-pcc)
     if: ${{ always() && needs.setup-instance.result != 'skipped' }}
-    needs: [ setup-instance, cuda-tests-linux ]
+    needs: [ setup-instance, cuda-pcc ]
     runs-on: ubuntu-latest
     steps:
       - name: Stop instance
         id: stop-instance
-        uses: zama-ai/slab-github-runner@58f2cae4bf2c0b6728083f5f009b6dc0eb6dc3ac
+        uses: zama-ai/slab-github-runner@c0e7168795bd78f61f61146951ed9d0c73c9b701
         with:
           mode: stop
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}
@@ -131,4 +124,4 @@ jobs:
         uses: rtCamp/action-slack-notify@4e5fb42d249be6a45a298f3c9543b111b02f7907
         env:
           SLACK_COLOR: ${{ job.status }}
-          SLACK_MESSAGE: "Instance teardown (cuda-tests-multi-gpu) finished with status: ${{ job.status }}. (${{ env.ACTION_RUN_URL }})"
+          SLACK_MESSAGE: "Instance teardown (cuda-pcc) finished with status: ${{ job.status }}. (${{ env.ACTION_RUN_URL }})"

.github/workflows/gpu_signed_integer_h100_tests.yml

@@ -0,0 +1,189 @@
+# Signed integer GPU tests on an H100 VM on hyperstack
+name: TFHE Cuda Backend - Signed integer tests on H100
+
+env:
+  CARGO_TERM_COLOR: always
+  ACTION_RUN_URL: ${{ github.server_url }}/${{ github.repository }}/actions/runs/${{ github.run_id }}
+  RUSTFLAGS: "-C target-cpu=native"
+  RUST_BACKTRACE: "full"
+  RUST_MIN_STACK: "8388608"
+  SLACK_CHANNEL: ${{ secrets.SLACK_CHANNEL }}
+  SLACK_ICON: https://pbs.twimg.com/profile_images/1274014582265298945/OjBKP9kn_400x400.png
+  SLACK_USERNAME: ${{ secrets.BOT_USERNAME }}
+  SLACK_WEBHOOK: ${{ secrets.SLACK_WEBHOOK }}
+  IS_PULL_REQUEST: ${{ github.event_name == 'pull_request' }}
+
+on:
+  # Allows you to run this workflow manually from the Actions tab as an alternative.
+  workflow_dispatch:
+  pull_request:
+      types: [ labeled ]
+
+jobs:
+  should-run:
+    runs-on: ubuntu-latest
+    permissions:
+      pull-requests: write
+    outputs:
+      gpu_test: ${{ env.IS_PULL_REQUEST == 'false' || steps.changed-files.outputs.gpu_any_changed }}
+    steps:
+      - name: Checkout tfhe-rs
+        uses: actions/checkout@d632683dd7b4114ad314bca15554477dd762a938
+        with:
+          fetch-depth: 0
+          token: ${{ secrets.FHE_ACTIONS_TOKEN }}
+
+      - name: Check for file changes
+        id: changed-files
+        uses: tj-actions/changed-files@c3a1bb2c992d77180ae65be6ae6c166cf40f857c
+        with:
+          since_last_remote_commit: true
+          files_yaml: |
+            gpu:
+              - tfhe/Cargo.toml
+              - tfhe/build.rs
+              - backends/tfhe-cuda-backend/**
+              - tfhe/src/core_crypto/gpu/**
+              - tfhe/src/integer/gpu/**
+              - tfhe/src/shortint/parameters/**
+              - tfhe/src/high_level_api/**
+              - tfhe/src/c_api/**
+              - 'tfhe/docs/**.md'
+              - Makefile
+              - '.github/workflows/gpu_signed_integer_h100_tests.yml'
+              - scripts/**
+              - ci/**
+
+  setup-instance:
+    name: Setup instance (cuda-h100-tests)
+    needs: should-run
+    if: github.event_name != 'pull_request' ||
+      (github.event.action != 'labeled' && needs.should-run.outputs.gpu_test == 'true') ||
+      (github.event.action == 'labeled' && github.event.label.name == 'approved' && needs.should-run.outputs.gpu_test == 'true')
+    runs-on: ubuntu-latest
+    outputs:
+      runner-name: ${{ steps.start-instance.outputs.label }}
+    steps:
+      - name: Start instance
+        id: start-instance
+        uses: zama-ai/slab-github-runner@c0e7168795bd78f61f61146951ed9d0c73c9b701
+        with:
+          mode: start
+          github-token: ${{ secrets.SLAB_ACTION_TOKEN }}
+          slab-url: ${{ secrets.SLAB_BASE_URL }}
+          job-secret: ${{ secrets.JOB_SECRET }}
+          backend: hyperstack
+          profile: single-h100
+
+  cuda-tests-linux:
+    name: CUDA H100 signed integer tests
+    needs: [ should-run, setup-instance ]
+    if: github.event_name != 'pull_request' ||
+      (github.event_name == 'pull_request' && needs.setup-instance.result != 'skipped')
+    concurrency:
+      group: ${{ github.workflow }}_${{ github.ref }}
+      cancel-in-progress: ${{ github.ref != 'refs/heads/main' }}
+    runs-on: ${{ needs.setup-instance.outputs.runner-name }}
+    strategy:
+      fail-fast: false
+      # explicit include-based build matrix, of known valid options
+      matrix:
+        include:
+          - os: ubuntu-22.04
+            cuda: "12.2"
+            gcc: 11 
+    env:
+      CUDA_PATH: /usr/local/cuda-${{ matrix.cuda }}
+      CMAKE_VERSION: 3.29.6
+    steps:
+      # Mandatory on hyperstack since a bootable volume is not re-usable yet.
+      - name: Install dependencies
+        run: |
+          sudo apt update
+          sudo apt install -y checkinstall zlib1g-dev libssl-dev 
+          wget https://github.com/Kitware/CMake/releases/download/v${{ env.CMAKE_VERSION }}/cmake-${{ env.CMAKE_VERSION }}.tar.gz
+          tar -zxvf cmake-${{ env.CMAKE_VERSION }}.tar.gz
+          cd cmake-${{ env.CMAKE_VERSION }}
+          ./bootstrap
+          make -j"$(nproc)"
+          sudo make install
+
+      - name: Checkout tfhe-rs
+        uses: actions/checkout@d632683dd7b4114ad314bca15554477dd762a938
+
+      - name: Set up home
+        run: |
+          echo "HOME=/home/ubuntu" >> "${GITHUB_ENV}"
+
+      - name: Install latest stable
+        uses: dtolnay/rust-toolchain@7b1c307e0dcbda6122208f10795a713336a9b35a
+        with:
+          toolchain: stable
+
+      - name: Export CUDA variables
+        if: ${{ !cancelled() }}
+        run: |
+          echo "CUDA_PATH=$CUDA_PATH" >> "${GITHUB_ENV}"
+          echo "$CUDA_PATH/bin" >> "${GITHUB_PATH}"
+          echo "LD_LIBRARY_PATH=$CUDA_PATH/lib:$LD_LIBRARY_PATH" >> "${GITHUB_ENV}"
+          echo "CUDACXX=/usr/local/cuda-${{ matrix.cuda }}/bin/nvcc" >> "${GITHUB_ENV}"
+
+      # Specify the correct host compilers
+      - name: Export gcc and g++ variables
+        if: ${{ !cancelled() }}
+        run: |
+          {
+            echo "CC=/usr/bin/gcc-${{ matrix.gcc }}";
+            echo "CXX=/usr/bin/g++-${{ matrix.gcc }}";
+            echo "CUDAHOSTCXX=/usr/bin/g++-${{ matrix.gcc }}";
+            echo "HOME=/home/ubuntu";
+          } >> "${GITHUB_ENV}"
+
+      - name: Check device is detected
+        if: ${{ !cancelled() }}
+        run: nvidia-smi
+
+      - name: Run signed integer tests
+        run: |
+          BIG_TESTS_INSTANCE=TRUE make test_signed_integer_gpu_ci
+
+      - name: Run signed integer multi-bit tests
+        run: |
+          BIG_TESTS_INSTANCE=TRUE make test_signed_integer_multi_bit_gpu_ci
+
+  slack-notify:
+    name: Slack Notification
+    needs: [ setup-instance, cuda-tests-linux ]
+    runs-on: ubuntu-latest
+    if: ${{ always() && needs.cuda-tests-linux.result != 'skipped' && failure() }}
+    continue-on-error: true
+    steps:
+      - name: Send message
+        uses: rtCamp/action-slack-notify@4e5fb42d249be6a45a298f3c9543b111b02f7907
+        env:
+          SLACK_COLOR: ${{ needs.cuda-tests-linux.result }}
+          SLACK_MESSAGE: "Integer GPU H100 tests finished with status: ${{ needs.cuda-tests-linux.result }}. (${{ env.ACTION_RUN_URL }})"
+
+  teardown-instance:
+    name: Teardown instance (cuda-h100-tests)
+    if: ${{ always() && needs.setup-instance.result != 'skipped' }}
+    needs: [ setup-instance, cuda-tests-linux ]
+    runs-on: ubuntu-latest
+    steps:
+      - name: Stop instance
+        id: stop-instance
+        uses: zama-ai/slab-github-runner@c0e7168795bd78f61f61146951ed9d0c73c9b701
+        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@4e5fb42d249be6a45a298f3c9543b111b02f7907
+        env:
+          SLACK_COLOR: ${{ job.status }}
+          SLACK_MESSAGE: "Instance teardown (cuda-h100-tests) finished with status: ${{ job.status }}. (${{ env.ACTION_RUN_URL }})"

.github/workflows/gpu_signed_integer_tests.yml

@@ -0,0 +1,203 @@
+# Compile and test tfhe-cuda-backend signed integer on an AWS instance
+name: TFHE Cuda Backend - Signed integer tests
+
+env:
+  CARGO_TERM_COLOR: always
+  ACTION_RUN_URL: ${{ github.server_url }}/${{ github.repository }}/actions/runs/${{ github.run_id }}
+  RUSTFLAGS: "-C target-cpu=native"
+  RUST_BACKTRACE: "full"
+  RUST_MIN_STACK: "8388608"
+  SLACK_CHANNEL: ${{ secrets.SLACK_CHANNEL }}
+  SLACK_ICON: https://pbs.twimg.com/profile_images/1274014582265298945/OjBKP9kn_400x400.png
+  SLACK_USERNAME: ${{ secrets.BOT_USERNAME }}
+  SLACK_WEBHOOK: ${{ secrets.SLACK_WEBHOOK }}
+  FAST_TESTS: TRUE
+  NIGHTLY_TESTS: FALSE
+  IS_PULL_REQUEST: ${{ github.event_name == 'pull_request' }}
+
+on:
+  # Allows you to run this workflow manually from the Actions tab as an alternative.
+  workflow_dispatch:
+  pull_request:
+    types:
+      - opened
+      - synchronize
+      - labeled
+  schedule:
+    # Nightly tests @ 1AM after each work day
+    - cron: "0 1 * * MON-FRI"
+
+jobs:
+  should-run:
+    runs-on: ubuntu-latest
+    permissions:
+      pull-requests: write
+    outputs:
+      gpu_test: ${{ env.IS_PULL_REQUEST == 'false' || steps.changed-files.outputs.gpu_any_changed }}
+    steps:
+      - name: Checkout tfhe-rs
+        uses: actions/checkout@d632683dd7b4114ad314bca15554477dd762a938
+        with:
+          fetch-depth: 0
+
+      - name: Check for file changes
+        id: changed-files
+        uses: tj-actions/changed-files@c3a1bb2c992d77180ae65be6ae6c166cf40f857c
+        with:
+          since_last_remote_commit: true
+          files_yaml: |
+            gpu:
+              - tfhe/Cargo.toml
+              - tfhe/build.rs
+              - backends/tfhe-cuda-backend/**
+              - tfhe/src/core_crypto/gpu/**
+              - tfhe/src/integer/gpu/**
+              - tfhe/src/shortint/parameters/**
+              - tfhe/src/high_level_api/**
+              - tfhe/src/c_api/**
+              - 'tfhe/docs/**.md'
+              - '.github/workflows/gpu_signed_integer_tests.yml'
+              - Makefile
+              - scripts/**
+              - ci/**
+
+  setup-instance:
+    name: Setup instance (cuda-signed-integer-tests)
+    runs-on: ubuntu-latest
+    needs: should-run
+    if: (github.event_name == 'schedule' && github.repository == 'zama-ai/tfhe-rs') ||
+      github.event_name == 'workflow_dispatch' ||
+      (github.event.action != 'labeled' && needs.should-run.outputs.gpu_test == 'true')
+    outputs:
+      runner-name: ${{ steps.start-instance.outputs.label }}
+    steps:
+      - name: Start instance
+        id: start-instance
+        uses: zama-ai/slab-github-runner@c0e7168795bd78f61f61146951ed9d0c73c9b701
+        with:
+          mode: start
+          github-token: ${{ secrets.SLAB_ACTION_TOKEN }}
+          slab-url: ${{ secrets.SLAB_BASE_URL }}
+          job-secret: ${{ secrets.JOB_SECRET }}
+          backend: hyperstack
+          profile: gpu-test
+
+  cuda-signed-integer-tests:
+    name: CUDA signed integer tests
+    needs: [ should-run, setup-instance ]
+    if: github.event_name != 'pull_request' ||
+      (github.event_name == 'pull_request' && needs.setup-instance.result != 'skipped')
+    concurrency:
+      group: ${{ github.workflow }}_${{ github.ref }}
+      cancel-in-progress: ${{ github.ref != 'refs/heads/main' }}
+    runs-on: ${{ needs.setup-instance.outputs.runner-name }}
+    strategy:
+      fail-fast: false
+      # explicit include-based build matrix, of known valid options
+      matrix:
+        include:
+          - os: ubuntu-22.04
+            cuda: "12.2"
+            gcc: 11 
+    env:
+      CUDA_PATH: /usr/local/cuda-${{ matrix.cuda }}
+      CMAKE_VERSION: 3.29.6
+    steps:
+      # Mandatory on hyperstack since a bootable volume is not re-usable yet.
+      - name: Install dependencies
+        run: |
+          sudo apt update
+          sudo apt install -y checkinstall zlib1g-dev libssl-dev 
+          wget https://github.com/Kitware/CMake/releases/download/v${{ env.CMAKE_VERSION }}/cmake-${{ env.CMAKE_VERSION }}.tar.gz
+          tar -zxvf cmake-${{ env.CMAKE_VERSION }}.tar.gz
+          cd cmake-${{ env.CMAKE_VERSION }}
+          ./bootstrap
+          make -j"$(nproc)"
+          sudo make install
+
+      - name: Checkout tfhe-rs
+        uses: actions/checkout@d632683dd7b4114ad314bca15554477dd762a938
+        with:
+          persist-credentials: 'false'
+          token: ${{ secrets.FHE_ACTIONS_TOKEN }}
+
+      - name: Set up home
+        run: |
+          echo "HOME=/home/ubuntu" >> "${GITHUB_ENV}"
+
+      - name: Install latest stable
+        uses: dtolnay/rust-toolchain@7b1c307e0dcbda6122208f10795a713336a9b35a
+        with:
+          toolchain: stable
+
+      - name: Export CUDA variables
+        if: ${{ !cancelled() }}
+        run: |
+          echo "CUDA_PATH=$CUDA_PATH" >> "${GITHUB_ENV}"
+          echo "$CUDA_PATH/bin" >> "${GITHUB_PATH}"
+          echo "LD_LIBRARY_PATH=$CUDA_PATH/lib:$LD_LIBRARY_PATH" >> "${GITHUB_ENV}"
+          echo "CUDACXX=/usr/local/cuda-${{ matrix.cuda }}/bin/nvcc" >> "${GITHUB_ENV}"
+
+      # Specify the correct host compilers
+      - name: Export gcc and g++ variables
+        if: ${{ !cancelled() }}
+        run: |
+          {
+            echo "CC=/usr/bin/gcc-${{ matrix.gcc }}";
+            echo "CXX=/usr/bin/g++-${{ matrix.gcc }}";
+            echo "CUDAHOSTCXX=/usr/bin/g++-${{ matrix.gcc }}";
+            echo "HOME=/home/ubuntu";
+          } >> "${GITHUB_ENV}"
+
+      - name: Should run nightly tests
+        if: github.event_name == 'schedule'
+        run: |
+          {
+            echo "FAST_TESTS=FALSE";
+            echo "NIGHTLY_TESTS=TRUE";
+          } >> "${GITHUB_ENV}"
+
+      - name: Check device is detected
+        if: ${{ !cancelled() }}
+        run: nvidia-smi
+
+      - name: Run signed integer multi-bit tests
+        run: |
+          make test_signed_integer_multi_bit_gpu_ci
+
+  slack-notify:
+    name: Slack Notification
+    needs: [ setup-instance, cuda-signed-integer-tests ]
+    runs-on: ubuntu-latest
+    if: ${{ always() && needs.cuda-signed-integer-tests.result != 'skipped' && failure() }}
+    continue-on-error: true
+    steps:
+      - name: Send message
+        uses: rtCamp/action-slack-notify@4e5fb42d249be6a45a298f3c9543b111b02f7907
+        env:
+          SLACK_COLOR: ${{ needs.cuda-signed-integer-tests.result }}
+          SLACK_MESSAGE: "Base GPU tests finished with status: ${{ needs.cuda-signed-integer-tests.result }}. (${{ env.ACTION_RUN_URL }})"
+
+  teardown-instance:
+    name: Teardown instance (cuda-tests)
+    if: ${{ always() && needs.setup-instance.result != 'skipped' }}
+    needs: [ setup-instance, cuda-signed-integer-tests ]
+    runs-on: ubuntu-latest
+    steps:
+      - name: Stop instance
+        id: stop-instance
+        uses: zama-ai/slab-github-runner@c0e7168795bd78f61f61146951ed9d0c73c9b701
+        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@4e5fb42d249be6a45a298f3c9543b111b02f7907
+        env:
+          SLACK_COLOR: ${{ job.status }}
+          SLACK_MESSAGE: "Instance teardown (cuda-signed-integer-tests) finished with status: ${{ job.status }}. (${{ env.ACTION_RUN_URL }})"

.github/workflows/gpu_unsigned_integer_h100_tests.yml

@@ -0,0 +1,189 @@
+# Test unsigned integers on an H100 VM on hyperstack
+name: TFHE Cuda Backend - Unsigned integer tests on H100
+
+env:
+  CARGO_TERM_COLOR: always
+  ACTION_RUN_URL: ${{ github.server_url }}/${{ github.repository }}/actions/runs/${{ github.run_id }}
+  RUSTFLAGS: "-C target-cpu=native"
+  RUST_BACKTRACE: "full"
+  RUST_MIN_STACK: "8388608"
+  SLACK_CHANNEL: ${{ secrets.SLACK_CHANNEL }}
+  SLACK_ICON: https://pbs.twimg.com/profile_images/1274014582265298945/OjBKP9kn_400x400.png
+  SLACK_USERNAME: ${{ secrets.BOT_USERNAME }}
+  SLACK_WEBHOOK: ${{ secrets.SLACK_WEBHOOK }}
+  IS_PULL_REQUEST: ${{ github.event_name == 'pull_request' }}
+
+on:
+  # Allows you to run this workflow manually from the Actions tab as an alternative.
+  workflow_dispatch:
+  pull_request:
+      types: [ labeled ]
+
+jobs:
+  should-run:
+    runs-on: ubuntu-latest
+    permissions:
+      pull-requests: write
+    outputs:
+      gpu_test: ${{ env.IS_PULL_REQUEST == 'false' || steps.changed-files.outputs.gpu_any_changed }}
+    steps:
+      - name: Checkout tfhe-rs
+        uses: actions/checkout@d632683dd7b4114ad314bca15554477dd762a938
+        with:
+          fetch-depth: 0
+          token: ${{ secrets.FHE_ACTIONS_TOKEN }}
+
+      - name: Check for file changes
+        id: changed-files
+        uses: tj-actions/changed-files@c3a1bb2c992d77180ae65be6ae6c166cf40f857c
+        with:
+          since_last_remote_commit: true
+          files_yaml: |
+            gpu:
+              - tfhe/Cargo.toml
+              - tfhe/build.rs
+              - backends/tfhe-cuda-backend/**
+              - tfhe/src/core_crypto/gpu/**
+              - tfhe/src/integer/gpu/**
+              - tfhe/src/shortint/parameters/**
+              - tfhe/src/high_level_api/**
+              - tfhe/src/c_api/**
+              - 'tfhe/docs/**.md'
+              - Makefile
+              - '.github/workflows/gpu_unsigned_integer_tests.yml'
+              - scripts/**
+              - ci/**
+
+  setup-instance:
+    name: Setup instance (cuda-h100-tests)
+    needs: should-run
+    if: github.event_name != 'pull_request' ||
+      (github.event.action != 'labeled' && needs.should-run.outputs.gpu_test == 'true') ||
+      (github.event.action == 'labeled' && github.event.label.name == 'approved' && needs.should-run.outputs.gpu_test == 'true')
+    runs-on: ubuntu-latest
+    outputs:
+      runner-name: ${{ steps.start-instance.outputs.label }}
+    steps:
+      - name: Start instance
+        id: start-instance
+        uses: zama-ai/slab-github-runner@c0e7168795bd78f61f61146951ed9d0c73c9b701
+        with:
+          mode: start
+          github-token: ${{ secrets.SLAB_ACTION_TOKEN }}
+          slab-url: ${{ secrets.SLAB_BASE_URL }}
+          job-secret: ${{ secrets.JOB_SECRET }}
+          backend: hyperstack
+          profile: single-h100
+
+  cuda-tests-linux:
+    name: CUDA H100 unsigned integer tests
+    needs: [ should-run, setup-instance ]
+    if: github.event_name != 'pull_request' ||
+      (github.event_name == 'pull_request' && needs.setup-instance.result != 'skipped')
+    concurrency:
+      group: ${{ github.workflow }}_${{ github.ref }}
+      cancel-in-progress: ${{ github.ref != 'refs/heads/main' }}
+    runs-on: ${{ needs.setup-instance.outputs.runner-name }}
+    strategy:
+      fail-fast: false
+      # explicit include-based build matrix, of known valid options
+      matrix:
+        include:
+          - os: ubuntu-22.04
+            cuda: "12.2"
+            gcc: 11 
+    env:
+      CUDA_PATH: /usr/local/cuda-${{ matrix.cuda }}
+      CMAKE_VERSION: 3.29.6
+    steps:
+      # Mandatory on hyperstack since a bootable volume is not re-usable yet.
+      - name: Install dependencies
+        run: |
+          sudo apt update
+          sudo apt install -y checkinstall zlib1g-dev libssl-dev 
+          wget https://github.com/Kitware/CMake/releases/download/v${{ env.CMAKE_VERSION }}/cmake-${{ env.CMAKE_VERSION }}.tar.gz
+          tar -zxvf cmake-${{ env.CMAKE_VERSION }}.tar.gz
+          cd cmake-${{ env.CMAKE_VERSION }}
+          ./bootstrap
+          make -j"$(nproc)"
+          sudo make install
+
+      - name: Checkout tfhe-rs
+        uses: actions/checkout@d632683dd7b4114ad314bca15554477dd762a938
+
+      - name: Set up home
+        run: |
+          echo "HOME=/home/ubuntu" >> "${GITHUB_ENV}"
+
+      - name: Install latest stable
+        uses: dtolnay/rust-toolchain@7b1c307e0dcbda6122208f10795a713336a9b35a
+        with:
+          toolchain: stable
+
+      - name: Export CUDA variables
+        if: ${{ !cancelled() }}
+        run: |
+          echo "CUDA_PATH=$CUDA_PATH" >> "${GITHUB_ENV}"
+          echo "$CUDA_PATH/bin" >> "${GITHUB_PATH}"
+          echo "LD_LIBRARY_PATH=$CUDA_PATH/lib:$LD_LIBRARY_PATH" >> "${GITHUB_ENV}"
+          echo "CUDACXX=/usr/local/cuda-${{ matrix.cuda }}/bin/nvcc" >> "${GITHUB_ENV}"
+
+      # Specify the correct host compilers
+      - name: Export gcc and g++ variables
+        if: ${{ !cancelled() }}
+        run: |
+          {
+            echo "CC=/usr/bin/gcc-${{ matrix.gcc }}";
+            echo "CXX=/usr/bin/g++-${{ matrix.gcc }}";
+            echo "CUDAHOSTCXX=/usr/bin/g++-${{ matrix.gcc }}";
+            echo "HOME=/home/ubuntu";
+          } >> "${GITHUB_ENV}"
+
+      - name: Check device is detected
+        if: ${{ !cancelled() }}
+        run: nvidia-smi
+
+      - name: Run unsigned integer tests
+        run: |
+          BIG_TESTS_INSTANCE=TRUE make test_unsigned_integer_gpu_ci
+
+      - name: Run unsigned integer multi-bit tests
+        run: |
+          BIG_TESTS_INSTANCE=TRUE make test_unsigned_integer_multi_bit_gpu_ci
+
+  slack-notify:
+    name: Slack Notification
+    needs: [ setup-instance, cuda-tests-linux ]
+    runs-on: ubuntu-latest
+    if: ${{ always() && needs.cuda-tests-linux.result != 'skipped' && failure() }}
+    continue-on-error: true
+    steps:
+      - name: Send message
+        uses: rtCamp/action-slack-notify@4e5fb42d249be6a45a298f3c9543b111b02f7907
+        env:
+          SLACK_COLOR: ${{ needs.cuda-tests-linux.result }}
+          SLACK_MESSAGE: "Unsigned integer GPU H100 tests finished with status: ${{ needs.cuda-tests-linux.result }}. (${{ env.ACTION_RUN_URL }})"
+
+  teardown-instance:
+    name: Teardown instance (cuda-h100-tests)
+    if: ${{ always() && needs.setup-instance.result != 'skipped' }}
+    needs: [ setup-instance, cuda-tests-linux ]
+    runs-on: ubuntu-latest
+    steps:
+      - name: Stop instance
+        id: stop-instance
+        uses: zama-ai/slab-github-runner@c0e7168795bd78f61f61146951ed9d0c73c9b701
+        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@4e5fb42d249be6a45a298f3c9543b111b02f7907
+        env:
+          SLACK_COLOR: ${{ job.status }}
+          SLACK_MESSAGE: "Instance teardown (cuda-h100-tests) finished with status: ${{ job.status }}. (${{ env.ACTION_RUN_URL }})"

.github/workflows/gpu_unsigned_integer_tests.yml

@@ -0,0 +1,200 @@
+# Compile and test tfhe-cuda-backend unsigned integer on an AWS instance
+name: TFHE Cuda Backend - Unsigned integer tests
+
+env:
+  CARGO_TERM_COLOR: always
+  ACTION_RUN_URL: ${{ github.server_url }}/${{ github.repository }}/actions/runs/${{ github.run_id }}
+  RUSTFLAGS: "-C target-cpu=native"
+  RUST_BACKTRACE: "full"
+  RUST_MIN_STACK: "8388608"
+  SLACK_CHANNEL: ${{ secrets.SLACK_CHANNEL }}
+  SLACK_ICON: https://pbs.twimg.com/profile_images/1274014582265298945/OjBKP9kn_400x400.png
+  SLACK_USERNAME: ${{ secrets.BOT_USERNAME }}
+  SLACK_WEBHOOK: ${{ secrets.SLACK_WEBHOOK }}
+  FAST_TESTS: TRUE
+  NIGHTLY_TESTS: FALSE
+
+on:
+  # Allows you to run this workflow manually from the Actions tab as an alternative.
+  workflow_dispatch:
+  pull_request:
+    types:
+      - opened
+      - synchronize
+      - labeled
+  schedule:
+    # Nightly tests @ 1AM after each work day
+    - cron: "0 1 * * MON-FRI"
+
+jobs:
+  should-run:
+    runs-on: ubuntu-latest
+    permissions:
+      pull-requests: write
+    outputs:
+      gpu_test: ${{ env.IS_PULL_REQUEST == 'false' || steps.changed-files.outputs.gpu_any_changed }}
+    steps:
+      - name: Checkout tfhe-rs
+        uses: actions/checkout@d632683dd7b4114ad314bca15554477dd762a938
+        with:
+          fetch-depth: 0
+          token: ${{ secrets.FHE_ACTIONS_TOKEN }}
+
+      - name: Check for file changes
+        id: changed-files
+        uses: tj-actions/changed-files@c3a1bb2c992d77180ae65be6ae6c166cf40f857c
+        with:
+          since_last_remote_commit: true
+          files_yaml: |
+            gpu:
+              - tfhe/Cargo.toml
+              - tfhe/build.rs
+              - backends/tfhe-cuda-backend/**
+              - tfhe/src/core_crypto/gpu/**
+              - tfhe/src/integer/gpu/**
+              - tfhe/src/shortint/parameters/**
+              - tfhe/src/high_level_api/**
+              - tfhe/src/c_api/**
+              - 'tfhe/docs/**.md'
+              - '.github/workflows/gpu_unsigned_integer_tests.yml'
+              - Makefile
+              - scripts/**
+              - ci/**
+
+  setup-instance:
+    name: Setup instance (cuda-unsigned-integer-tests)
+    needs: should-run
+    if: (github.event_name == 'schedule' && github.repository == 'zama-ai/tfhe-rs') ||
+      github.event_name == 'workflow_dispatch' ||
+      (github.event.action != 'labeled' && needs.should-run.outputs.gpu_test == 'true')
+    runs-on: ubuntu-latest
+    outputs:
+      runner-name: ${{ steps.start-instance.outputs.label }}
+    steps:
+      - name: Start instance
+        id: start-instance
+        uses: zama-ai/slab-github-runner@c0e7168795bd78f61f61146951ed9d0c73c9b701
+        with:
+          mode: start
+          github-token: ${{ secrets.SLAB_ACTION_TOKEN }}
+          slab-url: ${{ secrets.SLAB_BASE_URL }}
+          job-secret: ${{ secrets.JOB_SECRET }}
+          backend: hyperstack
+          profile: gpu-test
+
+  cuda-unsigned-integer-tests:
+    name: CUDA unsigned integer tests
+    needs: [ should-run, setup-instance ]
+    if: github.event_name != 'pull_request' ||
+      (github.event_name == 'pull_request' && needs.setup-instance.result != 'skipped')
+    concurrency:
+      group: ${{ github.workflow }}_${{ github.ref }}
+      cancel-in-progress: ${{ github.ref != 'refs/heads/main' }}
+    runs-on: ${{ needs.setup-instance.outputs.runner-name }}
+    strategy:
+      fail-fast: false
+      # explicit include-based build matrix, of known valid options
+      matrix:
+        include:
+          - os: ubuntu-22.04
+            cuda: "12.2"
+            gcc: 11
+    env:
+      CUDA_PATH: /usr/local/cuda-${{ matrix.cuda }}
+      CMAKE_VERSION: 3.29.6
+    steps:
+      # Mandatory on hyperstack since a bootable volume is not re-usable yet.
+      - name: Install dependencies
+        run: |
+          sudo apt update
+          sudo apt install -y checkinstall zlib1g-dev libssl-dev 
+          wget https://github.com/Kitware/CMake/releases/download/v${{ env.CMAKE_VERSION }}/cmake-${{ env.CMAKE_VERSION }}.tar.gz
+          tar -zxvf cmake-${{ env.CMAKE_VERSION }}.tar.gz
+          cd cmake-${{ env.CMAKE_VERSION }}
+          ./bootstrap
+          make -j"$(nproc)"
+          sudo make install
+
+      - name: Checkout tfhe-rs
+        uses: actions/checkout@d632683dd7b4114ad314bca15554477dd762a938
+
+      - name: Set up home
+        run: |
+          echo "HOME=/home/ubuntu" >> "${GITHUB_ENV}"
+
+      - name: Install latest stable
+        uses: dtolnay/rust-toolchain@7b1c307e0dcbda6122208f10795a713336a9b35a
+        with:
+          toolchain: stable
+
+      - name: Export CUDA variables
+        if: ${{ !cancelled() }}
+        run: |
+          echo "CUDA_PATH=$CUDA_PATH" >> "${GITHUB_ENV}"
+          echo "$CUDA_PATH/bin" >> "${GITHUB_PATH}"
+          echo "LD_LIBRARY_PATH=$CUDA_PATH/lib:$LD_LIBRARY_PATH" >> "${GITHUB_ENV}"
+          echo "CUDACXX=/usr/local/cuda-${{ matrix.cuda }}/bin/nvcc" >> "${GITHUB_ENV}"
+
+      # Specify the correct host compilers
+      - name: Export gcc and g++ variables
+        if: ${{ !cancelled() }}
+        run: |
+          {
+            echo "CC=/usr/bin/gcc-${{ matrix.gcc }}";
+            echo "CXX=/usr/bin/g++-${{ matrix.gcc }}";
+            echo "CUDAHOSTCXX=/usr/bin/g++-${{ matrix.gcc }}";
+            echo "HOME=/home/ubuntu";
+          } >> "${GITHUB_ENV}"
+
+      - name: Should run nightly tests
+        if: github.event_name == 'schedule'
+        run: |
+          {
+            echo "FAST_TESTS=FALSE";
+            echo "NIGHTLY_TESTS=TRUE";
+          } >> "${GITHUB_ENV}"
+
+      - name: Check device is detected
+        if: ${{ !cancelled() }}
+        run: nvidia-smi
+
+      - name: Run unsigned integer multi-bit tests
+        run: |
+          make test_unsigned_integer_multi_bit_gpu_ci
+
+  slack-notify:
+    name: Slack Notification
+    needs: [ setup-instance, cuda-unsigned-integer-tests ]
+    runs-on: ubuntu-latest
+    if: ${{ always() && needs.cuda-unsigned-integer-tests.result != 'skipped' && failure() }}
+    continue-on-error: true
+    steps:
+      - name: Send message
+        uses: rtCamp/action-slack-notify@4e5fb42d249be6a45a298f3c9543b111b02f7907
+        env:
+          SLACK_COLOR: ${{ needs.cuda-unsigned-integer-tests.result }}
+          SLACK_MESSAGE: "Unsigned integer GPU tests finished with status: ${{ needs.cuda-unsigned-integer-tests.result }}. (${{ env.ACTION_RUN_URL }})"
+
+  teardown-instance:
+    name: Teardown instance (cuda-tests)
+    if: ${{ always() && needs.setup-instance.result != 'skipped' }}
+    needs: [ setup-instance, cuda-unsigned-integer-tests ]
+    runs-on: ubuntu-latest
+    steps:
+      - name: Stop instance
+        id: stop-instance
+        uses: zama-ai/slab-github-runner@c0e7168795bd78f61f61146951ed9d0c73c9b701
+        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@4e5fb42d249be6a45a298f3c9543b111b02f7907
+        env:
+          SLACK_COLOR: ${{ job.status }}
+          SLACK_MESSAGE: "Instance teardown (cuda-unsigned-integer-tests) finished with status: ${{ job.status }}. (${{ env.ACTION_RUN_URL }})"

.github/workflows/integer_benchmark.yml

@@ -1,130 +0,0 @@
-# Run integer benchmarks on an AWS instance and return parsed results to Slab CI bot.
-name: Integer benchmarks
-
-on:
-  workflow_dispatch:
-    inputs:
-      instance_id:
-        description: "Instance ID"
-        type: string
-      instance_image_id:
-        description: "Instance AMI ID"
-        type: string
-      instance_type:
-        description: "Instance product type"
-        type: string
-      runner_name:
-        description: "Action runner name"
-        type: string
-      request_id:
-        description: "Slab request ID"
-        type: string
-
-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"
-
-jobs:
-  run-integer-benchmarks:
-    name: Execute integer benchmarks in EC2
-    runs-on: ${{ github.event.inputs.runner_name }}
-    if: ${{ !cancelled() }}
-    steps:
-      - name: Instance configuration used
-        run: |
-          echo "IDs: ${{ inputs.instance_id }}"
-          echo "AMI: ${{ inputs.instance_image_id }}"
-          echo "Type: ${{ inputs.instance_type }}"
-          echo "Request ID: ${{ inputs.request_id }}"
-
-      - name: Get benchmark date
-        run: |
-          echo "BENCH_DATE=$(date --iso-8601=seconds)" >> "${GITHUB_ENV}"
-
-      - name: Checkout tfhe-rs repo with tags
-        uses: actions/checkout@692973e3d937129bcbf40652eb9f2f61becf3332
-        with:
-          fetch-depth: 0
-
-      - name: Set up home
-        # "Install rust" step require root user to have a HOME directory which is not set.
-        run: |
-          echo "HOME=/home/ubuntu" >> "${GITHUB_ENV}"
-
-      - name: Install rust
-        uses: dtolnay/rust-toolchain@21dc36fb71dd22e3317045c0c31a3f4249868b17
-        with:
-          toolchain: nightly
-
-      - name: Run benchmarks with AVX512
-        run: |
-          make FAST_BENCH=TRUE bench_integer
-
-      - name: Parse benchmarks to csv
-        run: |
-          make PARSE_INTEGER_BENCH_CSV_FILE=${{ env.PARSE_INTEGER_BENCH_CSV_FILE }} \
-            parse_integer_benches
-
-      - name: Upload csv results artifact
-        uses: actions/upload-artifact@65462800fd760344b1a7b4382951275a0abb4808
-        with:
-          name: ${{ github.sha }}_csv_integer
-          path: ${{ env.PARSE_INTEGER_BENCH_CSV_FILE }}
-
-      - name: Parse results
-        run: |
-          COMMIT_DATE="$(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 }} \
-          --database tfhe_rs \
-          --hardware ${{ inputs.instance_type }} \
-          --project-version "${COMMIT_HASH}" \
-          --branch ${{ github.ref_name }} \
-          --commit-date "${COMMIT_DATE}" \
-          --bench-date "${{ env.BENCH_DATE }}" \
-          --walk-subdirs \
-          --name-suffix avx512 \
-          --throughput
-
-      - name: Upload parsed results artifact
-        uses: actions/upload-artifact@65462800fd760344b1a7b4382951275a0abb4808
-        with:
-          name: ${{ github.sha }}_integer
-          path: ${{ env.RESULTS_FILENAME }}
-
-      - name: Checkout Slab repo
-        uses: actions/checkout@692973e3d937129bcbf40652eb9f2f61becf3332
-        with:
-          repository: zama-ai/slab
-          path: slab
-          token: ${{ secrets.FHE_ACTIONS_TOKEN }}
-
-      - name: Send data to Slab
-        shell: bash
-        run: |
-          echo "Computing HMac on results file"
-          SIGNATURE="$(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 }}
-
-      - name: Slack Notification
-        if: ${{ failure() }}
-        continue-on-error: true
-        uses: rtCamp/action-slack-notify@4e5fb42d249be6a45a298f3c9543b111b02f7907
-        env:
-          SLACK_COLOR: ${{ job.status }}
-          SLACK_CHANNEL: ${{ secrets.SLACK_CHANNEL }}
-          SLACK_ICON: https://pbs.twimg.com/profile_images/1274014582265298945/OjBKP9kn_400x400.png
-          SLACK_MESSAGE: "Integer benchmarks finished with status: ${{ job.status }}. (${{ env.ACTION_RUN_URL }})"
-          SLACK_USERNAME: ${{ secrets.BOT_USERNAME }}
-          SLACK_WEBHOOK: ${{ secrets.SLACK_WEBHOOK }}

.github/workflows/integer_full_benchmark.yml

@@ -1,158 +0,0 @@
-# Run all integer benchmarks on an AWS instance and return parsed results to Slab CI bot.
-name: Integer full benchmarks
-
-on:
-  workflow_dispatch:
-    inputs:
-      instance_id:
-        description: "Instance ID"
-        type: string
-      instance_image_id:
-        description: "Instance AMI ID"
-        type: string
-      instance_type:
-        description: "Instance product type"
-        type: string
-      runner_name:
-        description: "Action runner name"
-        type: string
-      request_id:
-        description: "Slab request ID"
-        type: string
-      user_inputs:
-        description: "Type of benchmarks to run"
-        type: string
-        default: "weekly_benchmarks"
-
-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"
-
-jobs:
-  prepare-matrix:
-    name: Prepare operations matrix
-    runs-on: ubuntu-latest
-    outputs:
-      op_flavor: ${{ steps.set_op_flavor.outputs.op_flavor }}
-    steps:
-      - name: Weekly benchmarks
-        if: ${{ github.event.inputs.user_inputs == 'weekly_benchmarks' }}
-        run: |
-          echo "OP_FLAVOR=[\"default\"]" >> "${GITHUB_ENV}"
-
-      - name: Quarterly benchmarks
-        if: ${{ github.event.inputs.user_inputs == 'quarterly_benchmarks' }}
-        run: |
-          echo "OP_FLAVOR=[\"default\", \"smart\", \"unchecked\", \"misc\"]" >> "${GITHUB_ENV}"
-
-      -  name: Set operation flavor output
-         id: set_op_flavor
-         run: |
-          echo "op_flavor=${{ toJSON(env.OP_FLAVOR) }}" >> "${GITHUB_OUTPUT}"
-
-  integer-benchmarks:
-    name: Execute integer benchmarks for all operations flavor
-    needs: prepare-matrix
-    runs-on: ${{ github.event.inputs.runner_name }}
-    if: ${{ !cancelled() }}
-    continue-on-error: true
-    timeout-minutes: 1440  # 24 hours
-    strategy:
-      max-parallel: 1
-      matrix:
-        command: [ integer, integer_multi_bit]
-        op_flavor: ${{ fromJson(needs.prepare-matrix.outputs.op_flavor) }}
-    steps:
-      - name: Instance configuration used
-        run: |
-          echo "IDs: ${{ inputs.instance_id }}"
-          echo "AMI: ${{ inputs.instance_image_id }}"
-          echo "Type: ${{ inputs.instance_type }}"
-          echo "Request ID: ${{ inputs.request_id }}"
-
-      - name: Checkout tfhe-rs repo with tags
-        uses: actions/checkout@692973e3d937129bcbf40652eb9f2f61becf3332
-        with:
-          fetch-depth: 0
-
-      - name: Get benchmark details
-        run: |
-          {
-            echo "BENCH_DATE=$(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)";
-          } >> "${GITHUB_ENV}"
-
-      - name: Set up home
-        # "Install rust" step require root user to have a HOME directory which is not set.
-        run: |
-          echo "HOME=/home/ubuntu" >> "${GITHUB_ENV}"
-
-      - name: Install rust
-        uses: dtolnay/rust-toolchain@21dc36fb71dd22e3317045c0c31a3f4249868b17
-        with:
-          toolchain: nightly
-
-      - name: Checkout Slab repo
-        uses: actions/checkout@692973e3d937129bcbf40652eb9f2f61becf3332
-        with:
-          repository: zama-ai/slab
-          path: slab
-          token: ${{ secrets.FHE_ACTIONS_TOKEN }}
-
-      - name: Run benchmarks with AVX512
-        run: |
-          make BENCH_OP_FLAVOR=${{ matrix.op_flavor }} bench_${{ matrix.command }}
-
-      - name: Parse results
-        run: |
-          python3 ./ci/benchmark_parser.py target/criterion ${{ env.RESULTS_FILENAME }} \
-          --database tfhe_rs \
-          --hardware ${{ inputs.instance_type }} \
-          --project-version "${{ env.COMMIT_HASH }}" \
-          --branch ${{ github.ref_name }} \
-          --commit-date "${{ env.COMMIT_DATE }}" \
-          --bench-date "${{ env.BENCH_DATE }}" \
-          --walk-subdirs \
-          --name-suffix avx512 \
-          --throughput
-
-      - name: Upload parsed results artifact
-        uses: actions/upload-artifact@65462800fd760344b1a7b4382951275a0abb4808
-        with:
-          name: ${{ github.sha }}_${{ matrix.command }}_${{ matrix.op_flavor }}
-          path: ${{ env.RESULTS_FILENAME }}
-
-      - name: Send data to Slab
-        shell: bash
-        run: |
-          echo "Computing HMac on results file"
-          SIGNATURE="$(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 }}
-
-  slack-notification:
-    name: Slack Notification
-    runs-on: ${{ github.event.inputs.runner_name }}
-    if: ${{ failure() }}
-    needs: integer-benchmarks
-    steps:
-      - name: Notify
-        continue-on-error: true
-        uses: rtCamp/action-slack-notify@4e5fb42d249be6a45a298f3c9543b111b02f7907
-        env:
-          SLACK_COLOR: ${{ job.status }}
-          SLACK_CHANNEL: ${{ secrets.SLACK_CHANNEL }}
-          SLACK_ICON: https://pbs.twimg.com/profile_images/1274014582265298945/OjBKP9kn_400x400.png
-          SLACK_MESSAGE: "Integer full benchmarks finished with status: ${{ job.status }}. (${{ env.ACTION_RUN_URL }})"
-          SLACK_USERNAME: ${{ secrets.BOT_USERNAME }}
-          SLACK_WEBHOOK: ${{ secrets.SLACK_WEBHOOK }}

.github/workflows/integer_multi_bit_benchmark.yml

@@ -1,130 +0,0 @@
-# Run integer benchmarks with multi-bit cryptographic parameters on an AWS instance and return parsed results to Slab CI bot.
-name: Integer Multi-bit benchmarks
-
-on:
-  workflow_dispatch:
-    inputs:
-      instance_id:
-        description: "Instance ID"
-        type: string
-      instance_image_id:
-        description: "Instance AMI ID"
-        type: string
-      instance_type:
-        description: "Instance product type"
-        type: string
-      runner_name:
-        description: "Action runner name"
-        type: string
-      request_id:
-        description: "Slab request ID"
-        type: string
-
-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"
-
-jobs:
-  run-integer-benchmarks:
-    name: Execute integer multi-bit benchmarks in EC2
-    runs-on: ${{ github.event.inputs.runner_name }}
-    if: ${{ !cancelled() }}
-    steps:
-      - name: Instance configuration used
-        run: |
-          echo "IDs: ${{ inputs.instance_id }}"
-          echo "AMI: ${{ inputs.instance_image_id }}"
-          echo "Type: ${{ inputs.instance_type }}"
-          echo "Request ID: ${{ inputs.request_id }}"
-
-      - name: Get benchmark date
-        run: |
-          echo "BENCH_DATE=$(date --iso-8601=seconds)" >> "${GITHUB_ENV}"
-
-      - name: Checkout tfhe-rs repo with tags
-        uses: actions/checkout@692973e3d937129bcbf40652eb9f2f61becf3332
-        with:
-          fetch-depth: 0
-
-      - name: Set up home
-        # "Install rust" step require root user to have a HOME directory which is not set.
-        run: |
-          echo "HOME=/home/ubuntu" >> "${GITHUB_ENV}"
-
-      - name: Install rust
-        uses: dtolnay/rust-toolchain@21dc36fb71dd22e3317045c0c31a3f4249868b17
-        with:
-          toolchain: nightly
-
-      - name: Run multi-bit benchmarks with AVX512
-        run: |
-          make FAST_BENCH=TRUE bench_integer_multi_bit
-
-      - name: Parse benchmarks to csv
-        run: |
-          make PARSE_INTEGER_BENCH_CSV_FILE=${{ env.PARSE_INTEGER_BENCH_CSV_FILE }} \
-            parse_integer_benches
-
-      - name: Upload csv results artifact
-        uses: actions/upload-artifact@65462800fd760344b1a7b4382951275a0abb4808
-        with:
-          name: ${{ github.sha }}_csv_integer
-          path: ${{ env.PARSE_INTEGER_BENCH_CSV_FILE }}
-
-      - name: Parse results
-        run: |
-          COMMIT_DATE="$(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 }} \
-          --database tfhe_rs \
-          --hardware ${{ inputs.instance_type }} \
-          --project-version "${COMMIT_HASH}" \
-          --branch ${{ github.ref_name }} \
-          --commit-date "${COMMIT_DATE}" \
-          --bench-date "${{ env.BENCH_DATE }}" \
-          --walk-subdirs \
-          --name-suffix avx512 \
-          --throughput
-
-      - name: Upload parsed results artifact
-        uses: actions/upload-artifact@65462800fd760344b1a7b4382951275a0abb4808
-        with:
-          name: ${{ github.sha }}_integer
-          path: ${{ env.RESULTS_FILENAME }}
-
-      - name: Checkout Slab repo
-        uses: actions/checkout@692973e3d937129bcbf40652eb9f2f61becf3332
-        with:
-          repository: zama-ai/slab
-          path: slab
-          token: ${{ secrets.FHE_ACTIONS_TOKEN }}
-
-      - name: Send data to Slab
-        shell: bash
-        run: |
-          echo "Computing HMac on results file"
-          SIGNATURE="$(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 }}
-
-      - name: Slack Notification
-        if: ${{ failure() }}
-        continue-on-error: true
-        uses: rtCamp/action-slack-notify@4e5fb42d249be6a45a298f3c9543b111b02f7907
-        env:
-          SLACK_COLOR: ${{ job.status }}
-          SLACK_CHANNEL: ${{ secrets.SLACK_CHANNEL }}
-          SLACK_ICON: https://pbs.twimg.com/profile_images/1274014582265298945/OjBKP9kn_400x400.png
-          SLACK_MESSAGE: "Integer benchmarks finished with status: ${{ job.status }}. (${{ env.ACTION_RUN_URL }})"
-          SLACK_USERNAME: ${{ secrets.BOT_USERNAME }}
-          SLACK_WEBHOOK: ${{ secrets.SLACK_WEBHOOK }}

.github/workflows/m1_tests.yml

@@ -34,12 +34,12 @@ jobs:
     timeout-minutes: 720
 
     steps:
-      - uses: actions/checkout@692973e3d937129bcbf40652eb9f2f61becf3332
+      - uses: actions/checkout@d632683dd7b4114ad314bca15554477dd762a938
         with:
           persist-credentials: 'false'
 
       - name: Install latest stable
-        uses: dtolnay/rust-toolchain@21dc36fb71dd22e3317045c0c31a3f4249868b17
+        uses: dtolnay/rust-toolchain@7b1c307e0dcbda6122208f10795a713336a9b35a
         with:
           toolchain: stable
 

.github/workflows/make_release.yml

@@ -30,19 +30,62 @@ env:
   NPM_TAG: ""
 
 jobs:
-  publish_release:
-    name: Publish Release
+  package:
     runs-on: ubuntu-latest
+    outputs:
+      hash: ${{ steps.hash.outputs.hash }}
     steps:
       - name: Checkout
-        uses: actions/checkout@692973e3d937129bcbf40652eb9f2f61becf3332
+        uses: actions/checkout@d632683dd7b4114ad314bca15554477dd762a938
         with:
           fetch-depth: 0
-
-      - name: Create NPM version tag
+      - name: Prepare package
         run: |
-          echo "NPM_TAG=$(sed -n -e '1,/^version/p' tfhe/Cargo.toml | grep '^version[[:space:]]*=' | cut -d '=' -f 2 | tr -d ' ')" >> "${GITHUB_ENV}"
+          cargo package -p tfhe
+      - uses: actions/upload-artifact@50769540e7f4bd5e21e526ee35c689e35e0d6874 # v4.4.0
+        with:
+          name: crate
+          path: target/package/*.crate
+      - name: generate hash
+        id: hash
+        run: cd target/package && echo "hash=$(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@v2.0.0
+    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 Release
+    needs: [package] # for comparing hashes
+    runs-on: ubuntu-latest
+    permissions:
+      contents: read
+      id-token: write
+    steps:
+      - name: Checkout
+        uses: actions/checkout@d632683dd7b4114ad314bca15554477dd762a938
+        with:
+          fetch-depth: 0
+      - name: Create NPM version tag
+        if: ${{ inputs.npm_latest_tag }}
+        run: |
+          echo "NPM_TAG=latest" >> "${GITHUB_ENV}"
+      - name: Download artifact
+        uses: actions/download-artifact@fa0a91b85d4f404e444e00e005971372dc801d16 # v4.1.8
+        with:
+          name: crate
+          path: target/package
       - name: Publish crate.io package
         if: ${{ inputs.push_to_crates }}
         env:
@@ -51,6 +94,22 @@ jobs:
         run: |
           cargo publish -p tfhe --token ${{ env.CRATES_TOKEN }} ${{ env.DRY_RUN }}
 
+      - name: Generate hash
+        id: published_hash
+        run: cd target/package && echo "pub_hash=$(sha256sum ./*.crate | base64 -w0)" >> "${GITHUB_OUTPUT}"
+
+      - name: Slack notification (hashes comparison)
+        if: ${{ needs.package.outputs.hash != steps.published_hash.outputs.pub_hash }}
+        continue-on-error: true
+        uses: rtCamp/action-slack-notify@4e5fb42d249be6a45a298f3c9543b111b02f7907
+        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 }}
         run: |
@@ -64,14 +123,7 @@ jobs:
           package: tfhe/pkg/package.json
           dry-run: ${{ inputs.dry_run }}
           tag: ${{ env.NPM_TAG }}
-
-      - name: Publish web package as latest
-        if: ${{ inputs.push_web_package && inputs.npm_latest_tag }}
-        uses: JS-DevTools/npm-publish@19c28f1ef146469e409470805ea4279d47c3d35c
-        with:
-          token: ${{ secrets.NPM_TOKEN }}
-          package: tfhe/pkg/package.json
-          dry-run: ${{ inputs.dry_run }}
+          provenance: true
 
       - name: Build Node package
         if: ${{ inputs.push_node_package }}
@@ -89,14 +141,7 @@ jobs:
           package: tfhe/pkg/package.json
           dry-run: ${{ inputs.dry_run }}
           tag: ${{ env.NPM_TAG }}
-
-      - name: Publish Node package as latest
-        if: ${{ inputs.push_node_package && inputs.npm_latest_tag }}
-        uses: JS-DevTools/npm-publish@19c28f1ef146469e409470805ea4279d47c3d35c
-        with:
-          token: ${{ secrets.NPM_TOKEN }}
-          package: tfhe/pkg/package.json
-          dry-run: ${{ inputs.dry_run }}
+          provenance: true
 
       - name: Slack Notification
         if: ${{ failure() }}

.github/workflows/make_release_concrete_csprng.yml

@@ -1,4 +1,3 @@
-# Publish new release of tfhe-rs on various platform.
 name: Publish concrete-csprng release
 
 on:
@@ -18,7 +17,7 @@ jobs:
     runs-on: ubuntu-latest
     steps:
       - name: Checkout
-        uses: actions/checkout@692973e3d937129bcbf40652eb9f2f61becf3332
+        uses: actions/checkout@d632683dd7b4114ad314bca15554477dd762a938
         with:
           fetch-depth: 0
 
@@ -37,6 +36,6 @@ jobs:
           SLACK_COLOR: ${{ job.status }}
           SLACK_CHANNEL: ${{ secrets.SLACK_CHANNEL }}
           SLACK_ICON: https://pbs.twimg.com/profile_images/1274014582265298945/OjBKP9kn_400x400.png
-          SLACK_MESSAGE: "concrete-csprng release failed: (${{ env.ACTION_RUN_URL }})"
+          SLACK_MESSAGE: "concrete-csprng release finished with status: ${{ job.status }}. (${{ env.ACTION_RUN_URL }})"
           SLACK_USERNAME: ${{ secrets.BOT_USERNAME }}
           SLACK_WEBHOOK: ${{ secrets.SLACK_WEBHOOK }}

.github/workflows/make_release_cuda.yml

@@ -29,14 +29,14 @@ jobs:
     steps:
       - name: Start instance
         id: start-instance
-        uses: zama-ai/slab-github-runner@58f2cae4bf2c0b6728083f5f009b6dc0eb6dc3ac
+        uses: zama-ai/slab-github-runner@c0e7168795bd78f61f61146951ed9d0c73c9b701
         with:
           mode: start
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}
           slab-url: ${{ secrets.SLAB_BASE_URL }}
           job-secret: ${{ secrets.JOB_SECRET }}
           backend: aws
-          profile: gpu-test
+          profile: gpu-build
 
   publish-cuda-release:
     name: Publish CUDA Release
@@ -54,7 +54,7 @@ jobs:
       CUDA_PATH: /usr/local/cuda-${{ matrix.cuda }}
     steps:
       - name: Checkout
-        uses: actions/checkout@692973e3d937129bcbf40652eb9f2f61becf3332
+        uses: actions/checkout@d632683dd7b4114ad314bca15554477dd762a938
         with:
           fetch-depth: 0
 
@@ -63,7 +63,7 @@ jobs:
           echo "HOME=/home/ubuntu" >> "${GITHUB_ENV}"
 
       - name: Install latest stable
-        uses: dtolnay/rust-toolchain@21dc36fb71dd22e3317045c0c31a3f4249868b17
+        uses: dtolnay/rust-toolchain@7b1c307e0dcbda6122208f10795a713336a9b35a
         with:
           toolchain: stable
 
@@ -112,7 +112,7 @@ jobs:
     steps:
       - name: Stop instance
         id: stop-instance
-        uses: zama-ai/slab-github-runner@58f2cae4bf2c0b6728083f5f009b6dc0eb6dc3ac
+        uses: zama-ai/slab-github-runner@c0e7168795bd78f61f61146951ed9d0c73c9b701
         with:
           mode: stop
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}

.github/workflows/make_release_tfhe_versionable.yml

@@ -0,0 +1,48 @@
+name: Publish tfhe-versionable release
+
+on:
+  workflow_dispatch:
+    inputs:
+      dry_run:
+        description: "Dry-run"
+        type: boolean
+        default: true
+
+env:
+  ACTION_RUN_URL: ${{ github.server_url }}/${{ github.repository }}/actions/runs/${{ github.run_id }}
+
+jobs:
+  publish_release:
+    name: Publish tfhe-versionable Release
+    runs-on: ubuntu-latest
+    steps:
+      - name: Checkout
+        uses: actions/checkout@d632683dd7b4114ad314bca15554477dd762a938
+        with:
+          fetch-depth: 0
+
+      - name: Publish proc-macro crate
+        env:
+          CRATES_TOKEN: ${{ secrets.CARGO_REGISTRY_TOKEN }}
+          DRY_RUN: ${{ inputs.dry_run && '--dry-run' || '' }}
+        run: |
+          cargo publish -p tfhe-versionable-derive --token ${{ env.CRATES_TOKEN }} ${{ env.DRY_RUN }}
+
+      - name: Publish main crate
+        if: ${{ ! inputs.dry_run }}
+        env:
+          CRATES_TOKEN: ${{ secrets.CARGO_REGISTRY_TOKEN }}
+        run: |
+          cargo publish -p tfhe-versionable --token ${{ env.CRATES_TOKEN }}
+
+      - name: Slack Notification
+        if: ${{ failure() }}
+        continue-on-error: true
+        uses: rtCamp/action-slack-notify@4e5fb42d249be6a45a298f3c9543b111b02f7907
+        env:
+          SLACK_COLOR: ${{ job.status }}
+          SLACK_CHANNEL: ${{ secrets.SLACK_CHANNEL }}
+          SLACK_ICON: https://pbs.twimg.com/profile_images/1274014582265298945/OjBKP9kn_400x400.png
+          SLACK_MESSAGE: "tfhe-versionable release finished with status: ${{ job.status }}. (${{ env.ACTION_RUN_URL }})"
+          SLACK_USERNAME: ${{ secrets.BOT_USERNAME }}
+          SLACK_WEBHOOK: ${{ secrets.SLACK_WEBHOOK }}

.github/workflows/make_release_zk_pok.yml

@@ -18,7 +18,7 @@ jobs:
     runs-on: ubuntu-latest
     steps:
       - name: Checkout
-        uses: actions/checkout@692973e3d937129bcbf40652eb9f2f61becf3332
+        uses: actions/checkout@d632683dd7b4114ad314bca15554477dd762a938
         with:
           fetch-depth: 0
 

.github/workflows/parameters_check.yml

@@ -14,17 +14,17 @@ on:
 
 jobs:
   params-curves-security-check:
-    runs-on: ubuntu-latest
+    runs-on: large_ubuntu_16
     steps:
       - name: Checkout tfhe-rs
-        uses: actions/checkout@692973e3d937129bcbf40652eb9f2f61becf3332
+        uses: actions/checkout@d632683dd7b4114ad314bca15554477dd762a938
 
       - name: Checkout lattice-estimator
-        uses: actions/checkout@692973e3d937129bcbf40652eb9f2f61becf3332
+        uses: actions/checkout@d632683dd7b4114ad314bca15554477dd762a938
         with:
           repository: malb/lattice-estimator
           path: lattice_estimator
-          ref: '53508253629d3b5d31a2ad110e85dc69391ccb95'
+          ref: 'e80ec6bbbba212428b0e92d0467c18629cf9ed67'
 
       - name: Install Sage
         run: |

.github/workflows/shortint_benchmark.yml

@@ -1,128 +0,0 @@
-# Run shortint benchmarks on an AWS instance and return parsed results to Slab CI bot.
-name: Shortint benchmarks
-
-on:
-  workflow_dispatch:
-    inputs:
-      instance_id:
-        description: "Instance ID"
-        type: string
-      instance_image_id:
-        description: "Instance AMI ID"
-        type: string
-      instance_type:
-        description: "Instance product type"
-        type: string
-      runner_name:
-        description: "Action runner name"
-        type: string
-      request_id:
-        description: "Slab request ID"
-        type: string
-
-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"
-
-jobs:
-  run-shortint-benchmarks:
-    name: Execute shortint benchmarks in EC2
-    runs-on: ${{ github.event.inputs.runner_name }}
-    if: ${{ !cancelled() }}
-    steps:
-      - name: Instance configuration used
-        run: |
-          echo "IDs: ${{ inputs.instance_id }}"
-          echo "AMI: ${{ inputs.instance_image_id }}"
-          echo "Type: ${{ inputs.instance_type }}"
-          echo "Request ID: ${{ inputs.request_id }}"
-
-      - name: Get benchmark date
-        run: |
-          echo "BENCH_DATE=$(date --iso-8601=seconds)" >> "${GITHUB_ENV}"
-
-      - name: Checkout tfhe-rs repo with tags
-        uses: actions/checkout@692973e3d937129bcbf40652eb9f2f61becf3332
-        with:
-          fetch-depth: 0
-
-      - name: Set up home
-        # "Install rust" step require root user to have a HOME directory which is not set.
-        run: |
-          echo "HOME=/home/ubuntu" >> "${GITHUB_ENV}"
-
-      - name: Install rust
-        uses: dtolnay/rust-toolchain@21dc36fb71dd22e3317045c0c31a3f4249868b17
-        with:
-          toolchain: nightly
-
-      - name: Run benchmarks with AVX512
-        run: |
-          make bench_shortint
-
-      - 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 }} \
-          --database tfhe_rs \
-          --hardware ${{ inputs.instance_type }} \
-          --project-version "${COMMIT_HASH}" \
-          --branch ${{ github.ref_name }} \
-          --commit-date "${COMMIT_DATE}" \
-          --bench-date "${{ env.BENCH_DATE }}" \
-          --walk-subdirs \
-          --name-suffix avx512 \
-          --throughput
-
-      - name: Measure key sizes
-        run: |
-          make measure_shortint_key_sizes
-
-      - name: Parse key sizes results
-        run: |
-          python3 ./ci/benchmark_parser.py tfhe/shortint_key_sizes.csv ${{ env.RESULTS_FILENAME }} \
-          --key-sizes \
-          --append-results
-
-      - name: Upload parsed results artifact
-        uses: actions/upload-artifact@65462800fd760344b1a7b4382951275a0abb4808
-        with:
-          name: ${{ github.sha }}_shortint
-          path: ${{ env.RESULTS_FILENAME }}
-
-      - name: Checkout Slab repo
-        uses: actions/checkout@692973e3d937129bcbf40652eb9f2f61becf3332
-        with:
-          repository: zama-ai/slab
-          path: slab
-          token: ${{ secrets.FHE_ACTIONS_TOKEN }}
-
-      - name: Send data to Slab
-        shell: bash
-        run: |
-          echo "Computing HMac on results file"
-          SIGNATURE="$(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 }}
-
-      - name: Slack Notification
-        if: ${{ failure() }}
-        continue-on-error: true
-        uses: rtCamp/action-slack-notify@4e5fb42d249be6a45a298f3c9543b111b02f7907
-        env:
-          SLACK_COLOR: ${{ job.status }}
-          SLACK_CHANNEL: ${{ secrets.SLACK_CHANNEL }}
-          SLACK_ICON: https://pbs.twimg.com/profile_images/1274014582265298945/OjBKP9kn_400x400.png
-          SLACK_MESSAGE: "Shortint benchmarks finished with status: ${{ job.status }}. (${{ env.ACTION_RUN_URL }})"
-          SLACK_USERNAME: ${{ secrets.BOT_USERNAME }}
-          SLACK_WEBHOOK: ${{ secrets.SLACK_WEBHOOK }}

.github/workflows/shortint_full_benchmark.yml

@@ -1,152 +0,0 @@
-# Run all shortint benchmarks on an AWS instance and return parsed results to Slab CI bot.
-name: Shortint full benchmarks
-
-on:
-  workflow_dispatch:
-    inputs:
-      instance_id:
-        description: "Instance ID"
-        type: string
-      instance_image_id:
-        description: "Instance AMI ID"
-        type: string
-      instance_type:
-        description: "Instance product type"
-        type: string
-      runner_name:
-        description: "Action runner name"
-        type: string
-      request_id:
-        description: "Slab request ID"
-        type: string
-      # This input is not used in this workflow but still mandatory since a calling workflow could
-      # use it. If a triggering command include a user_inputs field, then the triggered workflow
-      # must include this very input, otherwise the workflow won't be called.
-      # See start_full_benchmarks.yml as example.
-      user_inputs:
-        description: "Type of benchmarks to run"
-        type: string
-        default: "weekly_benchmarks"
-
-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"
-
-jobs:
-  shortint-benchmarks:
-    name: Execute shortint benchmarks for all operations flavor
-    runs-on: ${{ github.event.inputs.runner_name }}
-    if: ${{ !cancelled() }}
-    strategy:
-      max-parallel: 1
-      matrix:
-        op_flavor: [ default, smart, unchecked ]
-    steps:
-      - name: Instance configuration used
-        run: |
-          echo "IDs: ${{ inputs.instance_id }}"
-          echo "AMI: ${{ inputs.instance_image_id }}"
-          echo "Type: ${{ inputs.instance_type }}"
-          echo "Request ID: ${{ inputs.request_id }}"
-
-      - name: Checkout tfhe-rs repo with tags
-        uses: actions/checkout@692973e3d937129bcbf40652eb9f2f61becf3332
-        with:
-          fetch-depth: 0
-
-      - name: Get benchmark details
-        run: |
-          {
-            echo "BENCH_DATE=$(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)";
-          } >> "${GITHUB_ENV}"
-
-      - name: Set up home
-        # "Install rust" step require root user to have a HOME directory which is not set.
-        run: |
-          echo "HOME=/home/ubuntu" >> "${GITHUB_ENV}"
-
-      - name: Install rust
-        uses: dtolnay/rust-toolchain@21dc36fb71dd22e3317045c0c31a3f4249868b17
-        with:
-          toolchain: nightly
-
-      - name: Checkout Slab repo
-        uses: actions/checkout@692973e3d937129bcbf40652eb9f2f61becf3332
-        with:
-          repository: zama-ai/slab
-          path: slab
-          token: ${{ secrets.FHE_ACTIONS_TOKEN }}
-
-      - name: Run benchmarks with AVX512
-        run: |
-          make BENCH_OP_FLAVOR=${{ matrix.op_flavor }} bench_shortint
-
-      - 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 }} \
-          --database tfhe_rs \
-          --hardware ${{ inputs.instance_type }} \
-          --project-version "${COMMIT_HASH}" \
-          --branch ${{ github.ref_name }} \
-          --commit-date "${COMMIT_DATE}" \
-          --bench-date "${{ env.BENCH_DATE }}" \
-          --walk-subdirs \
-          --name-suffix avx512 \
-          --throughput
-
-      # This small benchmark needs to be executed only once.
-      - name: Measure key sizes
-        if: matrix.op_flavor == 'default'
-        run: |
-          make measure_shortint_key_sizes
-
-      - name: Parse key sizes results
-        if: matrix.op_flavor == 'default'
-        run: |
-          python3 ./ci/benchmark_parser.py tfhe/shortint_key_sizes.csv ${{ env.RESULTS_FILENAME }} \
-          --key-sizes \
-          --append-results
-
-      - name: Upload parsed results artifact
-        uses: actions/upload-artifact@65462800fd760344b1a7b4382951275a0abb4808
-        with:
-          name: ${{ github.sha }}_shortint_${{ matrix.op_flavor }}
-          path: ${{ env.RESULTS_FILENAME }}
-
-      - name: Send data to Slab
-        shell: bash
-        run: |
-          echo "Computing HMac on results file"
-          SIGNATURE="$(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 }}
-
-  slack-notification:
-    name: Slack Notification
-    runs-on: ${{ github.event.inputs.runner_name }}
-    if: ${{ failure() }}
-    needs: shortint-benchmarks
-    steps:
-      - name: Notify
-        continue-on-error: true
-        uses: rtCamp/action-slack-notify@4e5fb42d249be6a45a298f3c9543b111b02f7907
-        env:
-          SLACK_COLOR: ${{ job.status }}
-          SLACK_CHANNEL: ${{ secrets.SLACK_CHANNEL }}
-          SLACK_ICON: https://pbs.twimg.com/profile_images/1274014582265298945/OjBKP9kn_400x400.png
-          SLACK_MESSAGE: "Shortint full benchmarks finished with status: ${{ job.status }}. (${{ env.ACTION_RUN_URL }})"
-          SLACK_USERNAME: ${{ secrets.BOT_USERNAME }}
-          SLACK_WEBHOOK: ${{ secrets.SLACK_WEBHOOK }}

.github/workflows/signed_integer_benchmark.yml

@@ -1,130 +0,0 @@
-# Run signed integer benchmarks on an AWS instance and return parsed results to Slab CI bot.
-name: Signed Integer benchmarks
-
-on:
-  workflow_dispatch:
-    inputs:
-      instance_id:
-        description: "Instance ID"
-        type: string
-      instance_image_id:
-        description: "Instance AMI ID"
-        type: string
-      instance_type:
-        description: "Instance product type"
-        type: string
-      runner_name:
-        description: "Action runner name"
-        type: string
-      request_id:
-        description: "Slab request ID"
-        type: string
-
-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"
-
-jobs:
-  run-integer-benchmarks:
-    name: Execute signed integer benchmarks in EC2
-    runs-on: ${{ github.event.inputs.runner_name }}
-    if: ${{ !cancelled() }}
-    steps:
-      - name: Instance configuration used
-        run: |
-          echo "IDs: ${{ inputs.instance_id }}"
-          echo "AMI: ${{ inputs.instance_image_id }}"
-          echo "Type: ${{ inputs.instance_type }}"
-          echo "Request ID: ${{ inputs.request_id }}"
-
-      - name: Get benchmark date
-        run: |
-          echo "BENCH_DATE=$(date --iso-8601=seconds)" >> "${GITHUB_ENV}"
-
-      - name: Checkout tfhe-rs repo with tags
-        uses: actions/checkout@692973e3d937129bcbf40652eb9f2f61becf3332
-        with:
-          fetch-depth: 0
-
-      - name: Set up home
-        # "Install rust" step require root user to have a HOME directory which is not set.
-        run: |
-          echo "HOME=/home/ubuntu" >> "${GITHUB_ENV}"
-
-      - name: Install rust
-        uses: dtolnay/rust-toolchain@21dc36fb71dd22e3317045c0c31a3f4249868b17
-        with:
-          toolchain: nightly
-
-      - name: Run benchmarks with AVX512
-        run: |
-          make FAST_BENCH=TRUE bench_signed_integer
-
-      - name: Parse benchmarks to csv
-        run: |
-          make PARSE_INTEGER_BENCH_CSV_FILE=${{ env.PARSE_INTEGER_BENCH_CSV_FILE }} \
-            parse_integer_benches
-
-      - name: Upload csv results artifact
-        uses: actions/upload-artifact@65462800fd760344b1a7b4382951275a0abb4808
-        with:
-          name: ${{ github.sha }}_csv_integer
-          path: ${{ env.PARSE_INTEGER_BENCH_CSV_FILE }}
-
-      - name: Parse results
-        run: |
-          COMMIT_DATE="$(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 }} \
-          --database tfhe_rs \
-          --hardware ${{ inputs.instance_type }} \
-          --project-version "${COMMIT_HASH}" \
-          --branch ${{ github.ref_name }} \
-          --commit-date "${COMMIT_DATE}" \
-          --bench-date "${{ env.BENCH_DATE }}" \
-          --walk-subdirs \
-          --name-suffix avx512 \
-          --throughput
-
-      - name: Upload parsed results artifact
-        uses: actions/upload-artifact@65462800fd760344b1a7b4382951275a0abb4808
-        with:
-          name: ${{ github.sha }}_integer
-          path: ${{ env.RESULTS_FILENAME }}
-
-      - name: Checkout Slab repo
-        uses: actions/checkout@692973e3d937129bcbf40652eb9f2f61becf3332
-        with:
-          repository: zama-ai/slab
-          path: slab
-          token: ${{ secrets.FHE_ACTIONS_TOKEN }}
-
-      - name: Send data to Slab
-        shell: bash
-        run: |
-          echo "Computing HMac on results file"
-          SIGNATURE="$(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 }}
-
-      - name: Slack Notification
-        if: ${{ failure() }}
-        continue-on-error: true
-        uses: rtCamp/action-slack-notify@4e5fb42d249be6a45a298f3c9543b111b02f7907
-        env:
-          SLACK_COLOR: ${{ job.status }}
-          SLACK_CHANNEL: ${{ secrets.SLACK_CHANNEL }}
-          SLACK_ICON: https://pbs.twimg.com/profile_images/1274014582265298945/OjBKP9kn_400x400.png
-          SLACK_MESSAGE: "Signed integer benchmarks finished with status: ${{ job.status }}. (${{ env.ACTION_RUN_URL }})"
-          SLACK_USERNAME: ${{ secrets.BOT_USERNAME }}
-          SLACK_WEBHOOK: ${{ secrets.SLACK_WEBHOOK }}

.github/workflows/signed_integer_full_benchmark.yml

@@ -1,136 +0,0 @@
-# Run all signed integer benchmarks on an AWS instance and return parsed results to Slab CI bot.
-name: Signed Integer full benchmarks
-
-on:
-  workflow_dispatch:
-    inputs:
-      instance_id:
-        description: "Instance ID"
-        type: string
-      instance_image_id:
-        description: "Instance AMI ID"
-        type: string
-      instance_type:
-        description: "Instance product type"
-        type: string
-      runner_name:
-        description: "Action runner name"
-        type: string
-      request_id:
-        description: "Slab request ID"
-        type: string
-      user_inputs:
-        description: "Type of benchmarks to run"
-        type: string
-        default: "weekly_benchmarks"
-
-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"
-
-jobs:
-  integer-benchmarks:
-    name: Execute signed integer benchmarks for all operations flavor
-    runs-on: ${{ github.event.inputs.runner_name }}
-    if: ${{ !cancelled() }}
-    continue-on-error: true
-    timeout-minutes: 1440  # 24 hours
-    strategy:
-      max-parallel: 1
-      matrix:
-        command: [ integer, integer_multi_bit ]
-        op_flavor: [ default, unchecked ]
-    steps:
-      - name: Instance configuration used
-        run: |
-          echo "IDs: ${{ inputs.instance_id }}"
-          echo "AMI: ${{ inputs.instance_image_id }}"
-          echo "Type: ${{ inputs.instance_type }}"
-          echo "Request ID: ${{ inputs.request_id }}"
-
-      - name: Checkout tfhe-rs repo with tags
-        uses: actions/checkout@692973e3d937129bcbf40652eb9f2f61becf3332
-        with:
-          fetch-depth: 0
-
-      - name: Get benchmark details
-        run: |
-          {
-            echo "BENCH_DATE=$(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)";
-          } >> "${GITHUB_ENV}"
-
-      - name: Set up home
-        # "Install rust" step require root user to have a HOME directory which is not set.
-        run: |
-          echo "HOME=/home/ubuntu" >> "${GITHUB_ENV}"
-
-      - name: Install rust
-        uses: dtolnay/rust-toolchain@21dc36fb71dd22e3317045c0c31a3f4249868b17
-        with:
-          toolchain: nightly
-
-      - name: Checkout Slab repo
-        uses: actions/checkout@692973e3d937129bcbf40652eb9f2f61becf3332
-        with:
-          repository: zama-ai/slab
-          path: slab
-          token: ${{ secrets.FHE_ACTIONS_TOKEN }}
-
-      - name: Run benchmarks with AVX512
-        run: |
-          make BENCH_OP_FLAVOR=${{ matrix.op_flavor }} bench_signed_${{ matrix.command }}
-
-      - name: Parse results
-        run: |
-          python3 ./ci/benchmark_parser.py target/criterion ${{ env.RESULTS_FILENAME }} \
-          --database tfhe_rs \
-          --hardware ${{ inputs.instance_type }} \
-          --project-version "${{ env.COMMIT_HASH }}" \
-          --branch ${{ github.ref_name }} \
-          --commit-date "${{ env.COMMIT_DATE }}" \
-          --bench-date "${{ env.BENCH_DATE }}" \
-          --walk-subdirs \
-          --name-suffix avx512 \
-          --throughput
-
-      - name: Upload parsed results artifact
-        uses: actions/upload-artifact@65462800fd760344b1a7b4382951275a0abb4808
-        with:
-          name: ${{ github.sha }}_${{ matrix.command }}_${{ matrix.op_flavor }}
-          path: ${{ env.RESULTS_FILENAME }}
-
-      - name: Send data to Slab
-        shell: bash
-        run: |
-          echo "Computing HMac on results file"
-          SIGNATURE="$(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 }}
-
-  slack-notification:
-    name: Slack Notification
-    runs-on: ${{ github.event.inputs.runner_name }}
-    if: ${{ failure() }}
-    needs: integer-benchmarks
-    steps:
-      - name: Notify
-        continue-on-error: true
-        uses: rtCamp/action-slack-notify@4e5fb42d249be6a45a298f3c9543b111b02f7907
-        env:
-          SLACK_COLOR: ${{ job.status }}
-          SLACK_CHANNEL: ${{ secrets.SLACK_CHANNEL }}
-          SLACK_ICON: https://pbs.twimg.com/profile_images/1274014582265298945/OjBKP9kn_400x400.png
-          SLACK_MESSAGE: "Signed integer full benchmarks finished with status: ${{ job.status }}. (${{ env.ACTION_RUN_URL }})"
-          SLACK_USERNAME: ${{ secrets.BOT_USERNAME }}
-          SLACK_WEBHOOK: ${{ secrets.SLACK_WEBHOOK }}

.github/workflows/signed_integer_multi_bit_benchmark.yml

@@ -1,130 +0,0 @@
-# Run signed integer benchmarks with multi-bit cryptographic parameters on an AWS instance and return parsed results to Slab CI bot.
-name: Signed Integer Multi-bit benchmarks
-
-on:
-  workflow_dispatch:
-    inputs:
-      instance_id:
-        description: "Instance ID"
-        type: string
-      instance_image_id:
-        description: "Instance AMI ID"
-        type: string
-      instance_type:
-        description: "Instance product type"
-        type: string
-      runner_name:
-        description: "Action runner name"
-        type: string
-      request_id:
-        description: "Slab request ID"
-        type: string
-
-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"
-
-jobs:
-  run-integer-benchmarks:
-    name: Execute signed integer multi-bit benchmarks in EC2
-    runs-on: ${{ github.event.inputs.runner_name }}
-    if: ${{ !cancelled() }}
-    steps:
-      - name: Instance configuration used
-        run: |
-          echo "IDs: ${{ inputs.instance_id }}"
-          echo "AMI: ${{ inputs.instance_image_id }}"
-          echo "Type: ${{ inputs.instance_type }}"
-          echo "Request ID: ${{ inputs.request_id }}"
-
-      - name: Get benchmark date
-        run: |
-          echo "BENCH_DATE=$(date --iso-8601=seconds)" >> "${GITHUB_ENV}"
-
-      - name: Checkout tfhe-rs repo with tags
-        uses: actions/checkout@692973e3d937129bcbf40652eb9f2f61becf3332
-        with:
-          fetch-depth: 0
-
-      - name: Set up home
-        # "Install rust" step require root user to have a HOME directory which is not set.
-        run: |
-          echo "HOME=/home/ubuntu" >> "${GITHUB_ENV}"
-
-      - name: Install rust
-        uses: dtolnay/rust-toolchain@21dc36fb71dd22e3317045c0c31a3f4249868b17
-        with:
-          toolchain: nightly
-
-      - name: Run multi-bit benchmarks with AVX512
-        run: |
-          make FAST_BENCH=TRUE bench_signed_integer_multi_bit
-
-      - name: Parse benchmarks to csv
-        run: |
-          make PARSE_INTEGER_BENCH_CSV_FILE=${{ env.PARSE_INTEGER_BENCH_CSV_FILE }} \
-            parse_integer_benches
-
-      - name: Upload csv results artifact
-        uses: actions/upload-artifact@65462800fd760344b1a7b4382951275a0abb4808
-        with:
-          name: ${{ github.sha }}_csv_integer
-          path: ${{ env.PARSE_INTEGER_BENCH_CSV_FILE }}
-
-      - name: Parse results
-        run: |
-          COMMIT_DATE="$(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 }} \
-          --database tfhe_rs \
-          --hardware ${{ inputs.instance_type }} \
-          --project-version "${COMMIT_HASH}" \
-          --branch ${{ github.ref_name }} \
-          --commit-date "${COMMIT_DATE}" \
-          --bench-date "${{ env.BENCH_DATE }}" \
-          --walk-subdirs \
-          --name-suffix avx512 \
-          --throughput
-
-      - name: Upload parsed results artifact
-        uses: actions/upload-artifact@65462800fd760344b1a7b4382951275a0abb4808
-        with:
-          name: ${{ github.sha }}_integer
-          path: ${{ env.RESULTS_FILENAME }}
-
-      - name: Checkout Slab repo
-        uses: actions/checkout@692973e3d937129bcbf40652eb9f2f61becf3332
-        with:
-          repository: zama-ai/slab
-          path: slab
-          token: ${{ secrets.FHE_ACTIONS_TOKEN }}
-
-      - name: Send data to Slab
-        shell: bash
-        run: |
-          echo "Computing HMac on results file"
-          SIGNATURE="$(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 }}
-
-      - name: Slack Notification
-        if: ${{ failure() }}
-        continue-on-error: true
-        uses: rtCamp/action-slack-notify@4e5fb42d249be6a45a298f3c9543b111b02f7907
-        env:
-          SLACK_COLOR: ${{ job.status }}
-          SLACK_CHANNEL: ${{ secrets.SLACK_CHANNEL }}
-          SLACK_ICON: https://pbs.twimg.com/profile_images/1274014582265298945/OjBKP9kn_400x400.png
-          SLACK_MESSAGE: "Signed integer benchmarks finished with status: ${{ job.status }}. (${{ env.ACTION_RUN_URL }})"
-          SLACK_USERNAME: ${{ secrets.BOT_USERNAME }}
-          SLACK_WEBHOOK: ${{ secrets.SLACK_WEBHOOK }}

.github/workflows/start_benchmarks.yml

@@ -1,123 +0,0 @@
-# Start all benchmark jobs on Slab CI bot.
-name: Start all benchmarks
-
-on:
-  push:
-    branches:
-      - "main"
-  workflow_dispatch:
-    inputs:
-      # The input name must be the name of the slab command to launch
-      boolean_bench:
-        description: "Run Boolean benches"
-        type: boolean
-        default: true
-      shortint_bench:
-        description: "Run shortint benches"
-        type: boolean
-        default: true
-      integer_bench:
-        description: "Run integer benches"
-        type: boolean
-        default: true
-      signed_integer_bench:
-        description: "Run signed integer benches"
-        type: boolean
-        default: true
-      integer_multi_bit_bench:
-        description: "Run integer multi bit benches"
-        type: boolean
-        default: true
-      signed_integer_multi_bit_bench:
-        description: "Run signed integer multi bit benches"
-        type: boolean
-        default: true
-      core_crypto_bench:
-        description: "Run core crypto benches"
-        type: boolean
-        default: true
-
-jobs:
-  start-benchmarks:
-    if: ${{ (github.event_name == 'push' && github.repository == 'zama-ai/tfhe-rs') || github.event_name == 'workflow_dispatch' }}
-    strategy:
-      matrix:
-        command: [ boolean_bench, shortint_bench,
-                   integer_bench, integer_multi_bit_bench,
-                   signed_integer_bench, signed_integer_multi_bit_bench,
-                   core_crypto_bench ]
-    runs-on: ubuntu-latest
-    steps:
-      - name: Checkout tfhe-rs
-        uses: actions/checkout@692973e3d937129bcbf40652eb9f2f61becf3332
-        with:
-          fetch-depth: 0
-
-      - name: Check for file changes
-        id: changed-files
-        uses: tj-actions/changed-files@eaf854ef0c266753e1abec356dcf17d92695b251
-        with:
-          files_yaml: |
-            common_benches:
-              - toolchain.txt
-              - Makefile
-              - ci/slab.toml
-              - tfhe/Cargo.toml
-              - tfhe/src/core_crypto/**
-              - .github/workflows/start_benchmarks.yml
-            boolean_bench:
-              - tfhe/src/boolean/**
-              - tfhe/benches/boolean/**
-              - .github/workflows/boolean_benchmark.yml
-            shortint_bench:
-              - tfhe/src/shortint/**
-              - tfhe/benches/shortint/**
-              - .github/workflows/shortint_benchmark.yml
-            integer_bench:
-              - tfhe/src/shortint/**
-              - tfhe/src/integer/**
-              - tfhe/benches/integer/bench.rs
-              - .github/workflows/integer_benchmark.yml
-            integer_multi_bit_bench:
-              - tfhe/src/shortint/**
-              - tfhe/src/integer/**
-              - tfhe/benches/integer/bench.rs
-              - .github/workflows/integer_multi_bit_benchmark.yml
-            signed_integer_bench:
-              - tfhe/src/shortint/**
-              - tfhe/src/integer/**
-              - tfhe/benches/integer/signed_bench.rs
-              - .github/workflows/signed_integer_benchmark.yml
-            signed_integer_multi_bit_bench:
-              - tfhe/src/shortint/**
-              - tfhe/src/integer/**
-              - tfhe/benches/integer/signed_bench.rs
-              - .github/workflows/signed_integer_multi_bit_benchmark.yml
-            core_crypto_bench:
-              - tfhe/src/core_crypto/**
-              - tfhe/benches/core_crypto/**
-              - .github/workflows/core_crypto_benchmark.yml
-
-      - name: Checkout Slab repo
-        uses: actions/checkout@692973e3d937129bcbf40652eb9f2f61becf3332
-        with:
-          repository: zama-ai/slab
-          path: slab
-          token: ${{ secrets.FHE_ACTIONS_TOKEN }}
-
-      - name: Start AWS job in Slab
-        # If manually triggered check that the current bench has been requested
-        # Otherwise if it's on push check that files relevant to benchmarks have changed
-        if: (github.event_name == 'workflow_dispatch' && github.event.inputs[matrix.command] == 'true') || (github.event_name == 'push' && (steps.changed-files.outputs.common_benches_any_changed == 'true' || steps.changed-files.outputs[format('{0}_any_changed', matrix.command)] == 'true'))
-        shell: bash
-        run: |
-          echo -n '{"command": "${{ matrix.command }}", "git_ref": "${{ github.ref }}", "sha": "${{ github.sha }}"}' > command.json
-          SIGNATURE="$(slab/scripts/hmac_calculator.sh command.json '${{ secrets.JOB_SECRET }}')"
-          curl -v -k \
-          --fail-with-body \
-          -H "Content-Type: application/json" \
-          -H "X-Slab-Repository: ${{ github.repository }}" \
-          -H "X-Slab-Command: start_aws" \
-          -H "X-Hub-Signature-256: sha256=${SIGNATURE}" \
-          -d @command.json \
-          ${{ secrets.SLAB_URL }}

.github/workflows/start_full_benchmarks.yml

@@ -1,66 +0,0 @@
-# Start all benchmark jobs, including full shortint and integer, on Slab CI bot.
-name: Start full suite benchmarks
-
-on:
-  schedule:
-    # Weekly benchmarks will be triggered each Saturday at 1a.m.
-    - cron: '0 1 * * 6'
-    # Quarterly benchmarks will be triggered right before end of quarter, the 25th of the current month at 4a.m.
-    # These benchmarks are far longer to execute hence the reason to run them only four time a year.
-    - cron: '0 4 25 MAR,JUN,SEP,DEC *'
-  workflow_dispatch:
-    inputs:
-      benchmark_type:
-        description: 'Benchmark type'
-        required: true
-        default: 'weekly'
-        type: choice
-        options:
-          - weekly
-          - quarterly
-
-jobs:
-  start-benchmarks:
-    if: ${{ (github.event_name == 'schedule' && github.repository == 'zama-ai/tfhe-rs') || github.event_name == 'workflow_dispatch' }}
-    strategy:
-      matrix:
-        command: [ boolean_bench, shortint_full_bench,
-                   integer_full_bench, signed_integer_full_bench,
-                   core_crypto_bench, wasm_client_bench ]
-    runs-on: ubuntu-latest
-    steps:
-      - name: Checkout tfhe-rs
-        uses: actions/checkout@692973e3d937129bcbf40652eb9f2f61becf3332
-        with:
-          fetch-depth: 0
-
-      - name: Checkout Slab repo
-        uses: actions/checkout@692973e3d937129bcbf40652eb9f2f61becf3332
-        with:
-          repository: zama-ai/slab
-          path: slab
-          token: ${{ secrets.FHE_ACTIONS_TOKEN }}
-
-      - name: Set benchmarks type as weekly
-        if: (github.event_name == 'workflow_dispatch' && inputs.benchmark_type == 'weekly') || github.event.schedule == '0 1 * * 6'
-        run: |
-          echo "BENCH_TYPE=weekly_benchmarks" >> "${GITHUB_ENV}"
-
-      - name: Set benchmarks type as quarterly
-        if: (github.event_name == 'workflow_dispatch' && inputs.benchmark_type == 'quarterly') || github.event.schedule == '0 4 25 MAR,JUN,SEP,DEC *'
-        run: |
-          echo "BENCH_TYPE=quarterly_benchmarks" >> "${GITHUB_ENV}"
-
-      - name: Start AWS job in Slab
-        shell: bash
-        run: |
-          echo -n '{"command": "${{ matrix.command }}", "git_ref": "${{ github.ref }}", "sha": "${{ github.sha }}", "user_inputs": "${{ env.BENCH_TYPE }}"}' > command.json
-          SIGNATURE="$(slab/scripts/hmac_calculator.sh command.json '${{ secrets.JOB_SECRET }}')"
-          curl -v -k \
-          --fail-with-body \
-          -H "Content-Type: application/json" \
-          -H "X-Slab-Repository: ${{ github.repository }}" \
-          -H "X-Slab-Command: start_aws" \
-          -H "X-Hub-Signature-256: sha256=${SIGNATURE}" \
-          -d @command.json \
-          ${{ secrets.SLAB_URL }}

.github/workflows/sync_on_push.yml

@@ -13,7 +13,7 @@ jobs:
     runs-on: ubuntu-latest
     steps:
       - name: Checkout repo
-        uses: actions/checkout@692973e3d937129bcbf40652eb9f2f61becf3332
+        uses: actions/checkout@d632683dd7b4114ad314bca15554477dd762a938
         with:
           fetch-depth: 0
       - name: git-sync

.gitignore

@@ -7,6 +7,7 @@ target/
 # In case of symlinked keys
 /keys
 
+**/*.rmeta
 **/Cargo.lock
 **/*.bin
 
@@ -25,6 +26,8 @@ backends/tfhe-cuda-backend/cuda/cmake-build-debug/
 
 # WASM tests
 tfhe/web_wasm_parallel_tests/server.PID
+venv/
+web-test-runner/
 
 # Dir used for backward compatibility test data
 tfhe/tfhe-backward-compat-data/

Cargo.toml

@@ -8,10 +8,13 @@ members = [
     "concrete-csprng",
     "backends/tfhe-cuda-backend",
     "utils/tfhe-versionable",
-    "utils/tfhe-versionable-derive"
+    "utils/tfhe-versionable-derive",
 ]
+
 exclude = [
-    "tfhe/backward_compatibility_tests"
+    "tfhe/backward_compatibility_tests",
+    "utils/cargo-tfhe-lints-inner",
+    "utils/cargo-tfhe-lints"
 ]
 
 [profile.bench]

Makefile

@@ -16,21 +16,17 @@ GEN_KEY_CACHE_COVERAGE_ONLY?=FALSE
 PARSE_INTEGER_BENCH_CSV_FILE?=tfhe_rs_integer_benches.csv
 FAST_TESTS?=FALSE
 FAST_BENCH?=FALSE
+NIGHTLY_TESTS?=FALSE
 BENCH_OP_FLAVOR?=DEFAULT
-NODE_VERSION=20
+NODE_VERSION=22.6
 FORWARD_COMPAT?=OFF
 BACKWARD_COMPAT_DATA_URL=https://github.com/zama-ai/tfhe-backward-compat-data.git
-BACKWARD_COMPAT_DATA_DIR=tfhe-backward-compat-data
-# sed: -n, do not print input stream, -e means a script/expression
-# 1,/version/ indicates from the first line, to the line matching version at the start of the line
-# p indicates to print, so we keep only the start of the Cargo.toml until we hit the first version
-# entry which should be the version of tfhe
-TFHE_CURRENT_VERSION:=\
-$(shell sed -n -e '1,/^version/p' tfhe/Cargo.toml | \
-grep '^version[[:space:]]*=' | cut -d '=' -f 2 | xargs)
-# Cargo has a hard time distinguishing between our package from the workspace and a package that
-# could be a dependency, so we build an unambiguous spec here
-TFHE_SPEC:=tfhe@$(TFHE_CURRENT_VERSION)
+BACKWARD_COMPAT_DATA_BRANCH?=v0.2
+BACKWARD_COMPAT_DATA_PROJECT=tfhe-backward-compat-data
+BACKWARD_COMPAT_DATA_DIR=$(BACKWARD_COMPAT_DATA_PROJECT)
+TFHE_SPEC:=tfhe
+WEB_RUNNER_DIR=web-test-runner
+WEB_SERVER_DIR=tfhe/web_wasm_parallel_tests
 # This is done to avoid forgetting it, we still precise the RUSTFLAGS in the commands to be able to
 # copy paste the command in the terminal and change them if required without forgetting the flags
 export RUSTFLAGS?=-C target-cpu=native
@@ -117,7 +113,7 @@ install_cargo_nextest: install_rs_build_toolchain
 .PHONY: install_wasm_pack # Install wasm-pack to build JS packages
 install_wasm_pack: install_rs_build_toolchain
 	@wasm-pack --version > /dev/null 2>&1 || \
-	cargo $(CARGO_RS_BUILD_TOOLCHAIN) install wasm-pack || \
+	cargo $(CARGO_RS_BUILD_TOOLCHAIN) install --locked wasm-pack@0.13.0 || \
 	( echo "Unable to install cargo wasm-pack, unknown error." && exit 1 )
 
 .PHONY: install_node # Install last version of NodeJS via nvm
@@ -147,6 +143,68 @@ install_tarpaulin: install_rs_build_toolchain
 	cargo $(CARGO_RS_BUILD_TOOLCHAIN) install cargo-tarpaulin --locked || \
 	( echo "Unable to install cargo tarpaulin, unknown error." && exit 1 )
 
+.PHONY: install_tfhe_lints # Install custom tfhe-rs lints
+install_tfhe_lints:
+	(cd utils/cargo-tfhe-lints-inner && cargo install --path .) && \
+	cd utils/cargo-tfhe-lints && cargo install --path .
+
+.PHONY: install_typos_checker # Install typos checker
+install_typos_checker: install_rs_build_toolchain
+	@typos --version > /dev/null 2>&1 || \
+	cargo $(CARGO_RS_BUILD_TOOLCHAIN) install typos-cli || \
+	( echo "Unable to install typos-cli, unknown error." && exit 1 )
+
+.PHONY: setup_venv # Setup Python virtualenv for wasm tests
+setup_venv:
+	python3 -m venv venv
+	@source venv/bin/activate && \
+	pip3 install -r ci/webdriver_requirements.txt
+
+# This is an internal target, not meant to be called on its own.
+install_web_resource:
+	wget -P $(dest) $(url)
+	@cd $(dest) && \
+	echo "$(checksum) $(filename)" > checksum && \
+	sha256sum -c checksum && \
+	rm checksum && \
+	$(decompress_cmd) $(filename)
+
+install_chrome_browser: url = "https://storage.googleapis.com/chrome-for-testing-public/128.0.6613.137/linux64/chrome-linux64.zip"
+install_chrome_browser: checksum = "c5d7da679f3a353ae4e4420ab113de06d4bd459152f5b17558390c02d9520566"
+install_chrome_browser: dest = "$(WEB_RUNNER_DIR)/chrome"
+install_chrome_browser: filename = "chrome-linux64.zip"
+install_chrome_browser: decompress_cmd = unzip
+
+.PHONY: install_chrome_browser # Install Chrome browser for Linux
+install_chrome_browser: install_web_resource
+
+install_chrome_web_driver: url = "https://storage.googleapis.com/chrome-for-testing-public/128.0.6613.137/linux64/chromedriver-linux64.zip"
+install_chrome_web_driver: checksum = "f041092f403fb7455a6da2871070b6587c32814a3e3c2b0a794d3d4aa4739151"
+install_chrome_web_driver: dest = "$(WEB_RUNNER_DIR)/chrome"
+install_chrome_web_driver: filename = "chromedriver-linux64.zip"
+install_chrome_web_driver: decompress_cmd = unzip
+
+.PHONY: install_chrome_web_driver # Install Chrome web driver for Linux
+install_chrome_web_driver: install_web_resource
+
+install_firefox_browser: url = "https://download-installer.cdn.mozilla.net/pub/firefox/releases/131.0/linux-x86_64/en-US/firefox-131.0.tar.bz2"
+install_firefox_browser: checksum = "4ca8504a62a31472ecb8c3a769d4301dd4ac692d4cc5d51b8fe2cf41e7b11106"
+install_firefox_browser: dest = "$(WEB_RUNNER_DIR)/firefox"
+install_firefox_browser: filename = "firefox-131.0.tar.bz2"
+install_firefox_browser: decompress_cmd = tar -xvf
+
+.PHONY: install_firefox_browser # Install firefox browser for Linux
+install_firefox_browser: install_web_resource
+
+install_firefox_web_driver: url = "https://github.com/mozilla/geckodriver/releases/download/v0.35.0/geckodriver-v0.35.0-linux64.tar.gz"
+install_firefox_web_driver: checksum = "ac26e9ba8f3b8ce0fbf7339b9c9020192f6dcfcbf04a2bcd2af80dfe6bb24260"
+install_firefox_web_driver: dest = "$(WEB_RUNNER_DIR)/firefox"
+install_firefox_web_driver: filename = "geckodriver-v0.35.0-linux64.tar.gz"
+install_firefox_web_driver: decompress_cmd = tar -xvf
+
+.PHONY: install_firefox_web_driver # Install firefox web driver for Linux
+install_firefox_web_driver: install_web_resource
+
 .PHONY: check_linelint_installed # Check if linelint newline linter is installed
 check_linelint_installed:
 	@printf "\n" | linelint - > /dev/null 2>&1 || \
@@ -208,6 +266,10 @@ check_fmt_js: check_nvm_installed
 	nvm use $(NODE_VERSION) && \
 	$(MAKE) -C tfhe/web_wasm_parallel_tests check_fmt
 
+.PHONY: check_typos # Check for typos in codebase
+check_typos: install_typos_checker
+	@typos && echo "No typos found"
+
 .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 \
@@ -215,6 +277,13 @@ clippy_gpu: install_rs_check_toolchain
 		--all-targets \
 		-p $(TFHE_SPEC) -- --no-deps -D warnings
 
+.PHONY: check_gpu # Run check on tfhe with "gpu" enabled
+check_gpu: install_rs_check_toolchain
+	RUSTFLAGS="$(RUSTFLAGS)" cargo "$(CARGO_RS_CHECK_TOOLCHAIN)" check \
+		--features=$(TARGET_ARCH_FEATURE),boolean,shortint,integer,internal-keycache,gpu \
+		--all-targets \
+		-p $(TFHE_SPEC)
+
 .PHONY: fix_newline # Fix newline at end of file issues to be UNIX compliant
 fix_newline: check_linelint_installed
 	linelint -a .
@@ -253,12 +322,18 @@ clippy_shortint: install_rs_check_toolchain
 	RUSTFLAGS="$(RUSTFLAGS)" cargo "$(CARGO_RS_CHECK_TOOLCHAIN)" clippy \
 		--features=$(TARGET_ARCH_FEATURE),shortint \
 		-p $(TFHE_SPEC) -- --no-deps -D warnings
+	RUSTFLAGS="$(RUSTFLAGS)" cargo "$(CARGO_RS_CHECK_TOOLCHAIN)" clippy \
+		--features=$(TARGET_ARCH_FEATURE),shortint,experimental \
+		-p $(TFHE_SPEC) -- --no-deps -D warnings
 
 .PHONY: clippy_integer # Run clippy lints enabling the integer features
 clippy_integer: install_rs_check_toolchain
 	RUSTFLAGS="$(RUSTFLAGS)" cargo "$(CARGO_RS_CHECK_TOOLCHAIN)" clippy \
 		--features=$(TARGET_ARCH_FEATURE),integer \
 		-p $(TFHE_SPEC) -- --no-deps -D warnings
+	RUSTFLAGS="$(RUSTFLAGS)" cargo "$(CARGO_RS_CHECK_TOOLCHAIN)" clippy \
+		--features=$(TARGET_ARCH_FEATURE),integer,experimental \
+		-p $(TFHE_SPEC) -- --no-deps -D warnings
 
 .PHONY: clippy # Run clippy lints enabling the boolean, shortint, integer
 clippy: install_rs_check_toolchain
@@ -266,6 +341,17 @@ clippy: install_rs_check_toolchain
 		--features=$(TARGET_ARCH_FEATURE),boolean,shortint,integer \
 		-p $(TFHE_SPEC) -- --no-deps -D warnings
 
+.PHONY: clippy_rustdoc # Run clippy lints on doctests enabling the boolean, shortint, integer and zk-pok
+clippy_rustdoc: install_rs_check_toolchain
+	if [[ "$(OS)" != "Linux" && "$(OS)" != "Darwin" ]]; then \
+		echo "WARNING: skipped clippy_rustdoc, unsupported OS $(OS)"; \
+		exit 0; \
+	fi && \
+	CLIPPYFLAGS="-D warnings" RUSTDOCFLAGS="--no-run --nocapture --test-builder ./scripts/clippy_driver.sh -Z unstable-options" \
+		cargo "$(CARGO_RS_CHECK_TOOLCHAIN)" test --doc \
+		--features=$(TARGET_ARCH_FEATURE),boolean,shortint,integer,zk-pok,pbs-stats \
+		-p $(TFHE_SPEC)
+
 .PHONY: clippy_c_api # Run clippy lints enabling the boolean, shortint and the C API
 clippy_c_api: install_rs_check_toolchain
 	RUSTFLAGS="$(RUSTFLAGS)" cargo "$(CARGO_RS_CHECK_TOOLCHAIN)" clippy \
@@ -274,6 +360,9 @@ clippy_c_api: install_rs_check_toolchain
 
 .PHONY: clippy_js_wasm_api # Run clippy lints enabling the boolean, shortint, integer and the js wasm API
 clippy_js_wasm_api: install_rs_check_toolchain
+	RUSTFLAGS="$(RUSTFLAGS)" cargo "$(CARGO_RS_CHECK_TOOLCHAIN)" clippy \
+		--features=boolean-client-js-wasm-api,shortint-client-js-wasm-api,integer-client-js-wasm-api,high-level-client-js-wasm-api,zk-pok \
+		-p $(TFHE_SPEC) -- --no-deps -D warnings
 	RUSTFLAGS="$(RUSTFLAGS)" cargo "$(CARGO_RS_CHECK_TOOLCHAIN)" clippy \
 		--features=boolean-client-js-wasm-api,shortint-client-js-wasm-api,integer-client-js-wasm-api,high-level-client-js-wasm-api \
 		-p $(TFHE_SPEC) -- --no-deps -D warnings
@@ -293,6 +382,9 @@ clippy_all_targets: install_rs_check_toolchain
 	RUSTFLAGS="$(RUSTFLAGS)" cargo "$(CARGO_RS_CHECK_TOOLCHAIN)" clippy --all-targets \
 		--features=$(TARGET_ARCH_FEATURE),boolean,shortint,integer,internal-keycache,zk-pok \
 		-p $(TFHE_SPEC) -- --no-deps -D warnings
+	RUSTFLAGS="$(RUSTFLAGS)" cargo "$(CARGO_RS_CHECK_TOOLCHAIN)" clippy --all-targets \
+		--features=$(TARGET_ARCH_FEATURE),boolean,shortint,integer,internal-keycache,zk-pok,experimental \
+		-p $(TFHE_SPEC) -- --no-deps -D warnings
 
 .PHONY: clippy_concrete_csprng # Run clippy lints on concrete-csprng
 clippy_concrete_csprng: install_rs_check_toolchain
@@ -305,19 +397,32 @@ clippy_zk_pok: install_rs_check_toolchain
 	RUSTFLAGS="$(RUSTFLAGS)" cargo "$(CARGO_RS_CHECK_TOOLCHAIN)" clippy --all-targets \
 		-p tfhe-zk-pok -- --no-deps -D warnings
 
+.PHONY: clippy_versionable # Run clippy lints on tfhe-versionable
+clippy_versionable: install_rs_check_toolchain
+	RUSTFLAGS="$(RUSTFLAGS)" cargo "$(CARGO_RS_CHECK_TOOLCHAIN)" clippy --all-targets \
+		-p tfhe-versionable-derive -- --no-deps -D warnings
+	RUSTFLAGS="$(RUSTFLAGS)" cargo "$(CARGO_RS_CHECK_TOOLCHAIN)" clippy --all-targets \
+		-p tfhe-versionable -- --no-deps -D warnings
+
 .PHONY: clippy_all # Run all clippy targets
-clippy_all: clippy clippy_boolean clippy_shortint clippy_integer clippy_all_targets clippy_c_api \
-clippy_js_wasm_api clippy_tasks clippy_core clippy_concrete_csprng clippy_zk_pok clippy_trivium
+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_concrete_csprng clippy_zk_pok clippy_trivium \
+clippy_versionable
 
 .PHONY: clippy_fast # Run main clippy targets
-clippy_fast: clippy clippy_all_targets clippy_c_api clippy_js_wasm_api clippy_tasks clippy_core \
-clippy_concrete_csprng
+clippy_fast: clippy_rustdoc clippy clippy_all_targets clippy_c_api clippy_js_wasm_api clippy_tasks \
+clippy_core clippy_concrete_csprng
 
 .PHONY: clippy_cuda_backend # Run clippy lints on the tfhe-cuda-backend
 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: tfhe_lints # Run custom tfhe-rs lints
+tfhe_lints: install_tfhe_lints
+	cd tfhe && RUSTFLAGS="$(RUSTFLAGS)" cargo tfhe-lints \
+		--features=$(TARGET_ARCH_FEATURE),boolean,shortint,integer,zk-pok -- -D warnings
+
 .PHONY: build_core # Build core_crypto without experimental features
 build_core: install_rs_build_toolchain install_rs_check_toolchain
 	RUSTFLAGS="$(RUSTFLAGS)" cargo $(CARGO_RS_BUILD_TOOLCHAIN) build --profile $(CARGO_PROFILE) \
@@ -361,32 +466,23 @@ build_tfhe_coverage: install_rs_build_toolchain
 	RUSTFLAGS="$(RUSTFLAGS) --cfg tarpaulin" cargo $(CARGO_RS_BUILD_TOOLCHAIN) build --profile $(CARGO_PROFILE) \
 		--features=$(TARGET_ARCH_FEATURE),boolean,shortint,integer,internal-keycache -p $(TFHE_SPEC) --tests
 
-.PHONY: symlink_c_libs_without_fingerprint # Link the .a and .so files without the changing hash part in target
-symlink_c_libs_without_fingerprint:
-	@./scripts/symlink_c_libs_without_fingerprint.sh \
-		--cargo-profile "$(CARGO_PROFILE)" \
-		--lib-name tfhe-c-api-dynamic-buffer
-
 .PHONY: build_c_api # Build the C API for boolean, shortint and integer
 build_c_api: install_rs_check_toolchain
 	RUSTFLAGS="$(RUSTFLAGS)" cargo $(CARGO_RS_CHECK_TOOLCHAIN) build --profile $(CARGO_PROFILE) \
 		--features=$(TARGET_ARCH_FEATURE),boolean-c-api,shortint-c-api,high-level-c-api,zk-pok,$(FORWARD_COMPAT_FEATURE) \
 		-p $(TFHE_SPEC)
-	@"$(MAKE)" symlink_c_libs_without_fingerprint
 
 .PHONY: build_c_api_gpu # Build the C API for boolean, shortint and integer
 build_c_api_gpu: install_rs_check_toolchain
 	RUSTFLAGS="$(RUSTFLAGS)" cargo $(CARGO_RS_CHECK_TOOLCHAIN) build --profile $(CARGO_PROFILE) \
 		--features=$(TARGET_ARCH_FEATURE),boolean-c-api,shortint-c-api,high-level-c-api,zk-pok,gpu \
 		-p $(TFHE_SPEC)
-	@"$(MAKE)" symlink_c_libs_without_fingerprint
 
 .PHONY: build_c_api_experimental_deterministic_fft # Build the C API for boolean, shortint and integer with experimental deterministic FFT
 build_c_api_experimental_deterministic_fft: install_rs_check_toolchain
 	RUSTFLAGS="$(RUSTFLAGS)" cargo $(CARGO_RS_CHECK_TOOLCHAIN) build --profile $(CARGO_PROFILE) \
 		--features=$(TARGET_ARCH_FEATURE),boolean-c-api,shortint-c-api,high-level-c-api,zk-pok,experimental-force_fft_algo_dif4,$(FORWARD_COMPAT_FEATURE) \
 		-p $(TFHE_SPEC)
-	@"$(MAKE)" symlink_c_libs_without_fingerprint
 
 .PHONY: build_web_js_api # Build the js API targeting the web browser
 build_web_js_api: install_rs_build_toolchain install_wasm_pack
@@ -402,7 +498,9 @@ build_web_js_api_parallel: install_rs_check_toolchain install_wasm_pack
 	RUSTFLAGS="$(WASM_RUSTFLAGS) -C target-feature=+atomics,+bulk-memory,+mutable-globals" rustup run $(RS_CHECK_TOOLCHAIN) \
 		wasm-pack build --release --target=web \
 		-- --features=boolean-client-js-wasm-api,shortint-client-js-wasm-api,integer-client-js-wasm-api,parallel-wasm-api,zk-pok \
-		-Z build-std=panic_abort,std
+		-Z build-std=panic_abort,std && \
+	find pkg/snippets -type f -iname workerHelpers.worker.js -exec sed -i "s|from '..\/..\/..\/';|from '..\/..\/..\/tfhe.js';|" {} \;
+	jq '.files += ["snippets"]' tfhe/pkg/package.json > tmp_pkg.json && mv -f tmp_pkg.json tfhe/pkg/package.json
 
 .PHONY: build_node_js_api # Build the js API targeting nodejs
 build_node_js_api: install_rs_build_toolchain install_wasm_pack
@@ -445,8 +543,8 @@ test_cuda_backend:
 	mkdir -p "$(TFHECUDA_BUILD)" && \
 		cd "$(TFHECUDA_BUILD)" && \
 		cmake .. -DCMAKE_BUILD_TYPE=Release -DTFHE_CUDA_BACKEND_BUILD_TESTS=ON && \
-		make -j "$(CPU_COUNT)" && \
-		make test
+		"$(MAKE)" -j "$(CPU_COUNT)" && \
+		"$(MAKE)" test
 
 .PHONY: test_gpu # Run the tests of the core_crypto module including experimental on the gpu backend
 test_gpu: test_core_crypto_gpu test_integer_gpu test_cuda_backend
@@ -465,6 +563,67 @@ test_integer_gpu: install_rs_build_toolchain
 	RUSTFLAGS="$(RUSTFLAGS)" cargo $(CARGO_RS_BUILD_TOOLCHAIN) test --doc --profile $(CARGO_PROFILE) \
 		--features=$(TARGET_ARCH_FEATURE),integer,gpu -p $(TFHE_SPEC) -- integer::gpu::server_key::
 
+.PHONY: test_integer_compression_gpu
+test_integer_compression_gpu: install_rs_build_toolchain
+	RUSTFLAGS="$(RUSTFLAGS)" cargo $(CARGO_RS_BUILD_TOOLCHAIN) test --profile $(CARGO_PROFILE) \
+		--features=$(TARGET_ARCH_FEATURE),integer,gpu -p $(TFHE_SPEC) -- integer::gpu::ciphertext::compressed_ciphertext_list::tests::
+	RUSTFLAGS="$(RUSTFLAGS)" cargo $(CARGO_RS_BUILD_TOOLCHAIN) test --doc --profile $(CARGO_PROFILE) \
+		--features=$(TARGET_ARCH_FEATURE),integer,gpu -p $(TFHE_SPEC) -- integer::gpu::ciphertext::compress
+
+.PHONY: test_integer_gpu_ci # Run the tests for integer ci on gpu backend
+test_integer_gpu_ci: install_rs_check_toolchain install_cargo_nextest
+	BIG_TESTS_INSTANCE="$(BIG_TESTS_INSTANCE)" \
+	FAST_TESTS="$(FAST_TESTS)" \
+	NIGHTLY_TESTS="$(NIGHTLY_TESTS)" \
+		./scripts/integer-tests.sh --rust-toolchain $(CARGO_RS_CHECK_TOOLCHAIN) \
+		--cargo-profile "$(CARGO_PROFILE)" --backend "gpu" \
+		--tfhe-package "$(TFHE_SPEC)"
+
+.PHONY: test_unsigned_integer_gpu_ci # Run the tests for unsigned integer ci on gpu backend
+test_unsigned_integer_gpu_ci: install_rs_check_toolchain install_cargo_nextest
+	BIG_TESTS_INSTANCE="$(BIG_TESTS_INSTANCE)" \
+	FAST_TESTS="$(FAST_TESTS)" \
+	NIGHTLY_TESTS="$(NIGHTLY_TESTS)" \
+		./scripts/integer-tests.sh --rust-toolchain $(CARGO_RS_CHECK_TOOLCHAIN) \
+		--cargo-profile "$(CARGO_PROFILE)" --backend "gpu" \
+		--unsigned-only --tfhe-package "$(TFHE_SPEC)"
+
+.PHONY: test_signed_integer_gpu_ci # Run the tests for signed integer ci on gpu backend
+test_signed_integer_gpu_ci: install_rs_check_toolchain install_cargo_nextest
+	BIG_TESTS_INSTANCE="$(BIG_TESTS_INSTANCE)" \
+	FAST_TESTS="$(FAST_TESTS)" \
+	NIGHTLY_TESTS="$(NIGHTLY_TESTS)" \
+		./scripts/integer-tests.sh --rust-toolchain $(CARGO_RS_CHECK_TOOLCHAIN) \
+		--cargo-profile "$(CARGO_PROFILE)" --backend "gpu" \
+		--signed-only --tfhe-package "$(TFHE_SPEC)"
+
+.PHONY: test_integer_multi_bit_gpu_ci # Run the tests for integer ci on gpu backend running only multibit tests
+test_integer_multi_bit_gpu_ci: install_rs_check_toolchain install_cargo_nextest
+	BIG_TESTS_INSTANCE="$(BIG_TESTS_INSTANCE)" \
+	FAST_TESTS="$(FAST_TESTS)" \
+	NIGHTLY_TESTS="$(NIGHTLY_TESTS)" \
+		./scripts/integer-tests.sh --rust-toolchain $(CARGO_RS_CHECK_TOOLCHAIN) \
+		--cargo-profile "$(CARGO_PROFILE)" --multi-bit --backend "gpu" \
+		--tfhe-package "$(TFHE_SPEC)"
+
+.PHONY: test_unsigned_integer_multi_bit_gpu_ci # Run the tests for unsigned integer ci on gpu backend running only multibit tests
+test_unsigned_integer_multi_bit_gpu_ci: install_rs_check_toolchain install_cargo_nextest
+	BIG_TESTS_INSTANCE="$(BIG_TESTS_INSTANCE)" \
+	FAST_TESTS="$(FAST_TESTS)" \
+	NIGHTLY_TESTS="$(NIGHTLY_TESTS)" \
+		./scripts/integer-tests.sh --rust-toolchain $(CARGO_RS_CHECK_TOOLCHAIN) \
+		--cargo-profile "$(CARGO_PROFILE)" --multi-bit --backend "gpu" \
+		--unsigned-only --tfhe-package "$(TFHE_SPEC)"
+
+.PHONY: test_signed_integer_multi_bit_gpu_ci # Run the tests for signed integer ci on gpu backend running only multibit tests
+test_signed_integer_multi_bit_gpu_ci: install_rs_check_toolchain install_cargo_nextest
+	BIG_TESTS_INSTANCE="$(BIG_TESTS_INSTANCE)" \
+	FAST_TESTS="$(FAST_TESTS)" \
+	NIGHTLY_TESTS="$(NIGHTLY_TESTS)" \
+		./scripts/integer-tests.sh --rust-toolchain $(CARGO_RS_CHECK_TOOLCHAIN) \
+		--cargo-profile "$(CARGO_PROFILE)" --multi-bit --backend "gpu" \
+		--signed-only --tfhe-package "$(TFHE_SPEC)"
+
 .PHONY: test_boolean # Run the tests of the boolean module
 test_boolean: install_rs_build_toolchain
 	RUSTFLAGS="$(RUSTFLAGS)" cargo $(CARGO_RS_BUILD_TOOLCHAIN) test --profile $(CARGO_PROFILE) \
@@ -527,6 +686,7 @@ test_shortint_cov: install_rs_check_toolchain install_tarpaulin
 test_integer_ci: install_rs_check_toolchain install_cargo_nextest
 	BIG_TESTS_INSTANCE="$(BIG_TESTS_INSTANCE)" \
 	FAST_TESTS="$(FAST_TESTS)" \
+	NIGHTLY_TESTS="$(NIGHTLY_TESTS)" \
 		./scripts/integer-tests.sh --rust-toolchain $(CARGO_RS_CHECK_TOOLCHAIN) \
 		--cargo-profile "$(CARGO_PROFILE)" --avx512-support "$(AVX512_SUPPORT)" \
 		--tfhe-package "$(TFHE_SPEC)"
@@ -535,6 +695,7 @@ test_integer_ci: install_rs_check_toolchain install_cargo_nextest
 test_unsigned_integer_ci: install_rs_check_toolchain install_cargo_nextest
 	BIG_TESTS_INSTANCE="$(BIG_TESTS_INSTANCE)" \
 	FAST_TESTS="$(FAST_TESTS)" \
+	NIGHTLY_TESTS="$(NIGHTLY_TESTS)" \
 		./scripts/integer-tests.sh --rust-toolchain $(CARGO_RS_CHECK_TOOLCHAIN) \
 		--cargo-profile "$(CARGO_PROFILE)" --avx512-support "$(AVX512_SUPPORT)" \
 		--unsigned-only --tfhe-package "$(TFHE_SPEC)"
@@ -543,6 +704,7 @@ test_unsigned_integer_ci: install_rs_check_toolchain install_cargo_nextest
 test_signed_integer_ci: install_rs_check_toolchain install_cargo_nextest
 	BIG_TESTS_INSTANCE="$(BIG_TESTS_INSTANCE)" \
 	FAST_TESTS="$(FAST_TESTS)" \
+	NIGHTLY_TESTS="$(NIGHTLY_TESTS)" \
 		./scripts/integer-tests.sh --rust-toolchain $(CARGO_RS_CHECK_TOOLCHAIN) \
 		--cargo-profile "$(CARGO_PROFILE)" --avx512-support "$(AVX512_SUPPORT)" \
 		--signed-only --tfhe-package "$(TFHE_SPEC)"
@@ -551,22 +713,25 @@ test_signed_integer_ci: install_rs_check_toolchain install_cargo_nextest
 test_integer_multi_bit_ci: install_rs_check_toolchain install_cargo_nextest
 	BIG_TESTS_INSTANCE="$(BIG_TESTS_INSTANCE)" \
 	FAST_TESTS="$(FAST_TESTS)" \
+	NIGHTLY_TESTS="$(NIGHTLY_TESTS)" \
 		./scripts/integer-tests.sh --rust-toolchain $(CARGO_RS_CHECK_TOOLCHAIN) \
 		--cargo-profile "$(CARGO_PROFILE)" --multi-bit --avx512-support "$(AVX512_SUPPORT)" \
 		--tfhe-package "$(TFHE_SPEC)"
 
-.PHONY: test_unsigned_integer_multi_bit_ci # Run the tests for nsigned integer ci running only multibit tests
+.PHONY: test_unsigned_integer_multi_bit_ci # Run the tests for unsigned integer ci running only multibit tests
 test_unsigned_integer_multi_bit_ci: install_rs_check_toolchain install_cargo_nextest
 	BIG_TESTS_INSTANCE="$(BIG_TESTS_INSTANCE)" \
 	FAST_TESTS="$(FAST_TESTS)" \
+	NIGHTLY_TESTS="$(NIGHTLY_TESTS)" \
 		./scripts/integer-tests.sh --rust-toolchain $(CARGO_RS_CHECK_TOOLCHAIN) \
 		--cargo-profile "$(CARGO_PROFILE)" --multi-bit --avx512-support "$(AVX512_SUPPORT)" \
 		--unsigned-only --tfhe-package "$(TFHE_SPEC)"
 
-.PHONY: test_signed_integer_multi_bit_ci # Run the tests for nsigned integer ci running only multibit tests
+.PHONY: test_signed_integer_multi_bit_ci # Run the tests for signed integer ci running only multibit tests
 test_signed_integer_multi_bit_ci: install_rs_check_toolchain install_cargo_nextest
 	BIG_TESTS_INSTANCE="$(BIG_TESTS_INSTANCE)" \
 	FAST_TESTS="$(FAST_TESTS)" \
+	NIGHTLY_TESTS="$(NIGHTLY_TESTS)" \
 		./scripts/integer-tests.sh --rust-toolchain $(CARGO_RS_CHECK_TOOLCHAIN) \
 		--cargo-profile "$(CARGO_PROFILE)" --multi-bit --avx512-support "$(AVX512_SUPPORT)" \
 		--signed-only --tfhe-package "$(TFHE_SPEC)"
@@ -658,16 +823,23 @@ test_zk_pok: install_rs_build_toolchain
 .PHONY: test_versionable # Run tests for tfhe-versionable subcrate
 test_versionable: install_rs_build_toolchain
 	RUSTFLAGS="$(RUSTFLAGS)" cargo $(CARGO_RS_BUILD_TOOLCHAIN) test --profile $(CARGO_PROFILE) \
-		-p tfhe-versionable
+		--all-targets -p tfhe-versionable
 
+# The backward compat data repo holds historical binary data but also rust code to generate and load them.
+# Here we use the "patch" functionality of Cargo to make sure the repo used for the data is the same as the one used for the code.
 .PHONY: test_backward_compatibility_ci
 test_backward_compatibility_ci: install_rs_build_toolchain
 	TFHE_BACKWARD_COMPAT_DATA_DIR="$(BACKWARD_COMPAT_DATA_DIR)" RUSTFLAGS="$(RUSTFLAGS)" cargo $(CARGO_RS_BUILD_TOOLCHAIN) test --profile $(CARGO_PROFILE) \
-		--features=$(TARGET_ARCH_FEATURE),shortint,integer -p $(TFHE_SPEC) test_backward_compatibility -- --nocapture
+		--config "patch.'$(BACKWARD_COMPAT_DATA_URL)'.$(BACKWARD_COMPAT_DATA_PROJECT).path=\"tfhe/$(BACKWARD_COMPAT_DATA_DIR)\"" \
+		--features=$(TARGET_ARCH_FEATURE),shortint,integer,zk-pok -p $(TFHE_SPEC) test_backward_compatibility -- --nocapture
 
 .PHONY: test_backward_compatibility # Same as test_backward_compatibility_ci but tries to clone the data repo first if needed
 test_backward_compatibility: tfhe/$(BACKWARD_COMPAT_DATA_DIR) test_backward_compatibility_ci
 
+.PHONY: backward_compat_branch # Prints the required backward compatibility branch
+backward_compat_branch:
+	@echo "$(BACKWARD_COMPAT_DATA_BRANCH)"
+
 .PHONY: doc # Build rust doc
 doc: install_rs_check_toolchain
 	@# Even though we are not in docs.rs, this allows to "just" build the doc
@@ -730,7 +902,7 @@ check_compile_tests_benches_gpu: install_rs_build_toolchain
 	mkdir -p "$(TFHECUDA_BUILD)" && \
 		cd "$(TFHECUDA_BUILD)" && \
 		cmake .. -DCMAKE_BUILD_TYPE=Debug -DTFHE_CUDA_BACKEND_BUILD_TESTS=ON -DTFHE_CUDA_BACKEND_BUILD_BENCHMARKS=ON && \
-		make -j "$(CPU_COUNT)"
+		"$(MAKE)" -j "$(CPU_COUNT)"
 
 .PHONY: build_nodejs_test_docker # Build a docker image with tools to run nodejs tests for wasm API
 build_nodejs_test_docker:
@@ -750,18 +922,50 @@ test_nodejs_wasm_api_in_docker: build_nodejs_test_docker
 
 .PHONY: test_nodejs_wasm_api # Run tests for the nodejs on wasm API
 test_nodejs_wasm_api: build_node_js_api
-	cd tfhe && node --test js_on_wasm_tests
+	cd tfhe/js_on_wasm_tests && npm install && npm run test
 
-.PHONY: test_web_js_api_parallel # Run tests for the web wasm api
-test_web_js_api_parallel: build_web_js_api_parallel
-	$(MAKE) -C tfhe/web_wasm_parallel_tests test
 
-.PHONY: test_web_js_api_parallel_ci # Run tests for the web wasm api
-test_web_js_api_parallel_ci: build_web_js_api_parallel
+# This is an internal target, not meant to be called on its own.
+run_web_js_api_parallel: build_web_js_api_parallel setup_venv
+	cd $(WEB_SERVER_DIR) && npm install && npm run build
+	source venv/bin/activate && \
+	python ci/webdriver.py \
+	--browser-path $(browser_path) \
+	--driver-path $(driver_path) \
+	--browser-kind  $(browser_kind) \
+	--server-cmd "npm run server" \
+	--server-workdir "$(WEB_SERVER_DIR)" \
+	--id-pattern $(filter)
+
+test_web_js_api_parallel_chrome: browser_path = "$(WEB_RUNNER_DIR)/chrome/chrome-linux64/chrome"
+test_web_js_api_parallel_chrome: driver_path = "$(WEB_RUNNER_DIR)/chrome/chromedriver-linux64/chromedriver"
+test_web_js_api_parallel_chrome: browser_kind = chrome
+test_web_js_api_parallel_chrome: filter = Test
+
+.PHONY: test_web_js_api_parallel_chrome # Run tests for the web wasm api on Chrome
+test_web_js_api_parallel_chrome: run_web_js_api_parallel
+
+.PHONY: test_web_js_api_parallel_chrome_ci # Run tests for the web wasm api on Chrome
+test_web_js_api_parallel_chrome_ci: setup_venv
 	source ~/.nvm/nvm.sh && \
 	nvm install $(NODE_VERSION) && \
 	nvm use $(NODE_VERSION) && \
-	$(MAKE) -C tfhe/web_wasm_parallel_tests test-ci
+	$(MAKE) test_web_js_api_parallel_chrome
+
+test_web_js_api_parallel_firefox: browser_path = "$(WEB_RUNNER_DIR)/firefox/firefox/firefox"
+test_web_js_api_parallel_firefox: driver_path = "$(WEB_RUNNER_DIR)/firefox/geckodriver"
+test_web_js_api_parallel_firefox: browser_kind = firefox
+test_web_js_api_parallel_firefox: filter = Test
+
+.PHONY: test_web_js_api_parallel_firefox # Run tests for the web wasm api on Firefox
+test_web_js_api_parallel_firefox: run_web_js_api_parallel
+
+.PHONY: test_web_js_api_parallel_firefox_ci # Run tests for the web wasm api on Firefox
+test_web_js_api_parallel_firefox_ci: setup_venv
+	source ~/.nvm/nvm.sh && \
+	nvm install $(NODE_VERSION) && \
+	nvm use $(NODE_VERSION) && \
+	$(MAKE) test_web_js_api_parallel_firefox
 
 .PHONY: no_tfhe_typo # Check we did not invert the h and f in tfhe
 no_tfhe_typo:
@@ -779,6 +983,11 @@ dieharder_csprng: install_dieharder build_concrete_csprng
 # Benchmarks
 #
 
+.PHONY: print_doc_bench_parameters # Print parameters used in doc benchmarks
+print_doc_bench_parameters:
+	RUSTFLAGS="" cargo run --example print_doc_bench_parameters \
+	--features=$(TARGET_ARCH_FEATURE),shortint,internal-keycache -p tfhe
+
 .PHONY: bench_integer # Run benchmarks for unsigned integer
 bench_integer: install_rs_check_toolchain
 	RUSTFLAGS="$(RUSTFLAGS)" __TFHE_RS_BENCH_OP_FLAVOR=$(BENCH_OP_FLAVOR) __TFHE_RS_FAST_BENCH=$(FAST_BENCH) \
@@ -800,6 +1009,18 @@ bench_integer_gpu: install_rs_check_toolchain
 	--bench integer-bench \
 	--features=$(TARGET_ARCH_FEATURE),integer,gpu,internal-keycache,nightly-avx512 -p $(TFHE_SPEC) --
 
+.PHONY: bench_integer_compression # Run benchmarks for unsigned integer compression
+bench_integer_compression: install_rs_check_toolchain
+	RUSTFLAGS="$(RUSTFLAGS)" cargo $(CARGO_RS_CHECK_TOOLCHAIN) bench \
+	--bench	glwe_packing_compression-integer-bench \
+	--features=$(TARGET_ARCH_FEATURE),integer,internal-keycache,nightly-avx512 -p $(TFHE_SPEC) --
+
+.PHONY: bench_integer_compression_gpu
+bench_integer_compression_gpu: install_rs_check_toolchain
+	RUSTFLAGS="$(RUSTFLAGS)" cargo $(CARGO_RS_CHECK_TOOLCHAIN) bench \
+	--bench	glwe_packing_compression-integer-bench \
+	--features=$(TARGET_ARCH_FEATURE),integer,internal-keycache,gpu -p $(TFHE_SPEC) --
+
 .PHONY: bench_integer_multi_bit # Run benchmarks for unsigned integer using multi-bit parameters
 bench_integer_multi_bit: install_rs_check_toolchain
 	RUSTFLAGS="$(RUSTFLAGS)" __TFHE_RS_BENCH_TYPE=MULTI_BIT \
@@ -824,6 +1045,14 @@ bench_integer_multi_bit_gpu: install_rs_check_toolchain
 	--bench integer-bench \
 	--features=$(TARGET_ARCH_FEATURE),integer,gpu,internal-keycache,nightly-avx512 -p $(TFHE_SPEC) --
 
+.PHONY: bench_unsigned_integer_multi_bit_gpu # Run benchmarks for unsigned integer on GPU backend using multi-bit parameters
+bench_unsigned_integer_multi_bit_gpu: install_rs_check_toolchain
+	RUSTFLAGS="$(RUSTFLAGS)" __TFHE_RS_BENCH_TYPE=MULTI_BIT \
+	__TFHE_RS_BENCH_OP_FLAVOR=$(BENCH_OP_FLAVOR) __TFHE_RS_FAST_BENCH=$(FAST_BENCH) \
+	cargo $(CARGO_RS_CHECK_TOOLCHAIN) bench \
+	--bench integer-bench \
+	--features=$(TARGET_ARCH_FEATURE),integer,gpu,internal-keycache,nightly-avx512 -p $(TFHE_SPEC) -- ::unsigned
+
 .PHONY: bench_integer_zk # Run benchmarks for integer encryption with ZK proofs
 bench_integer_zk: install_rs_check_toolchain
 	RUSTFLAGS="$(RUSTFLAGS)" \
@@ -839,16 +1068,12 @@ bench_shortint: install_rs_check_toolchain
 	--bench shortint-bench \
 	--features=$(TARGET_ARCH_FEATURE),shortint,internal-keycache,nightly-avx512 -p $(TFHE_SPEC)
 
-.PHONY: bench_oprf # Run benchmarks for shortint
-bench_oprf: install_rs_check_toolchain
+.PHONY: bench_shortint_oprf # Run benchmarks for shortint
+bench_shortint_oprf: install_rs_check_toolchain
 	RUSTFLAGS="$(RUSTFLAGS)" \
 	cargo $(CARGO_RS_CHECK_TOOLCHAIN) bench \
 	--bench oprf-shortint-bench \
 	--features=$(TARGET_ARCH_FEATURE),shortint,internal-keycache,nightly-avx512 -p $(TFHE_SPEC)
-	RUSTFLAGS="$(RUSTFLAGS)" \
-	cargo $(CARGO_RS_CHECK_TOOLCHAIN) bench \
-	--bench oprf-integer-bench \
-	--features=$(TARGET_ARCH_FEATURE),integer,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
@@ -878,7 +1103,7 @@ bench_pbs128: install_rs_check_toolchain
 
 .PHONY: bench_pbs_gpu # Run benchmarks for PBS on GPU backend
 bench_pbs_gpu: install_rs_check_toolchain
-	RUSTFLAGS="$(RUSTFLAGS)" cargo $(CARGO_RS_CHECK_TOOLCHAIN) bench \
+	RUSTFLAGS="$(RUSTFLAGS)" __TFHE_RS_FAST_BENCH=$(FAST_BENCH) cargo $(CARGO_RS_CHECK_TOOLCHAIN) bench \
 	--bench pbs-bench \
 	--features=$(TARGET_ARCH_FEATURE),boolean,shortint,gpu,internal-keycache,nightly-avx512 -p $(TFHE_SPEC)
 
@@ -894,15 +1119,35 @@ bench_ks_gpu: install_rs_check_toolchain
 	--bench ks-bench \
 	--features=$(TARGET_ARCH_FEATURE),boolean,shortint,gpu,internal-keycache,nightly-avx512 -p $(TFHE_SPEC)
 
-.PHONY: bench_web_js_api_parallel # Run benchmarks for the web wasm api
-bench_web_js_api_parallel: build_web_js_api_parallel
-	$(MAKE) -C tfhe/web_wasm_parallel_tests bench
+bench_web_js_api_parallel_chrome: browser_path = "$(WEB_RUNNER_DIR)/chrome/chrome-linux64/chrome"
+bench_web_js_api_parallel_chrome: driver_path = "$(WEB_RUNNER_DIR)/chrome/chromedriver-linux64/chromedriver"
+bench_web_js_api_parallel_chrome: browser_kind = chrome
+bench_web_js_api_parallel_chrome: filter = Bench
 
-.PHONY: bench_web_js_api_parallel_ci # Run benchmarks for the web wasm api
-bench_web_js_api_parallel_ci: build_web_js_api_parallel
+.PHONY: bench_web_js_api_parallel_chrome # Run benchmarks for the web wasm api
+bench_web_js_api_parallel_chrome: run_web_js_api_parallel
+
+.PHONY: bench_web_js_api_parallel_chrome_ci # Run benchmarks for the web wasm api
+bench_web_js_api_parallel_chrome_ci: setup_venv
 	source ~/.nvm/nvm.sh && \
-	nvm use node && \
-	$(MAKE) -C tfhe/web_wasm_parallel_tests bench-ci
+	nvm install $(NODE_VERSION) && \
+	nvm use $(NODE_VERSION) && \
+	$(MAKE) bench_web_js_api_parallel_chrome
+
+bench_web_js_api_parallel_firefox: browser_path = "$(WEB_RUNNER_DIR)/firefox/firefox/firefox"
+bench_web_js_api_parallel_firefox: driver_path = "$(WEB_RUNNER_DIR)/firefox/geckodriver"
+bench_web_js_api_parallel_firefox: browser_kind = firefox
+bench_web_js_api_parallel_firefox: filter = Bench
+
+.PHONY: bench_web_js_api_parallel_firefox # Run benchmarks for the web wasm api
+bench_web_js_api_parallel_firefox: run_web_js_api_parallel
+
+.PHONY: bench_web_js_api_parallel_firefox_ci # Run benchmarks for the web wasm api
+bench_web_js_api_parallel_firefox_ci: setup_venv
+	source ~/.nvm/nvm.sh && \
+	nvm install $(NODE_VERSION) && \
+	nvm use $(NODE_VERSION) && \
+	$(MAKE) bench_web_js_api_parallel_firefox
 
 #
 # Utility tools
@@ -950,7 +1195,7 @@ parse_wasm_benchmarks: install_rs_check_toolchain
 	RUSTFLAGS="$(RUSTFLAGS)" cargo $(CARGO_RS_CHECK_TOOLCHAIN) run --profile $(CARGO_PROFILE) \
 	--example wasm_benchmarks_parser \
 	--features=$(TARGET_ARCH_FEATURE),shortint,internal-keycache \
-	-- web_wasm_parallel_tests/test/benchmark_results
+	-- wasm_benchmark_results.json
 
 .PHONY: write_params_to_file # Gather all crypto parameters into a file with a Sage readable format.
 write_params_to_file: install_rs_check_toolchain
@@ -960,7 +1205,7 @@ write_params_to_file: install_rs_check_toolchain
 
 .PHONY: clone_backward_compat_data # Clone the data repo needed for backward compatibility tests
 clone_backward_compat_data:
-	./scripts/clone_backward_compat_data.sh $(BACKWARD_COMPAT_DATA_URL) tfhe/$(BACKWARD_COMPAT_DATA_DIR)
+	./scripts/clone_backward_compat_data.sh $(BACKWARD_COMPAT_DATA_URL) $(BACKWARD_COMPAT_DATA_BRANCH) tfhe/$(BACKWARD_COMPAT_DATA_DIR)
 
 tfhe/$(BACKWARD_COMPAT_DATA_DIR): clone_backward_compat_data
 
@@ -989,14 +1234,14 @@ sha256_bool: install_rs_check_toolchain
 	--features=$(TARGET_ARCH_FEATURE),boolean
 
 .PHONY: pcc # pcc stands for pre commit checks (except GPU)
-pcc: no_tfhe_typo no_dbg_log check_fmt lint_doc check_md_docs_are_tested check_intra_md_links \
-clippy_all check_compile_tests
+pcc: no_tfhe_typo no_dbg_log check_fmt check_typos lint_doc check_md_docs_are_tested check_intra_md_links \
+clippy_all tfhe_lints check_compile_tests
 
 .PHONY: pcc_gpu # pcc stands for pre commit checks for GPU compilation
 pcc_gpu: clippy_gpu clippy_cuda_backend check_compile_tests_benches_gpu
 
 .PHONY: fpcc # pcc stands for pre commit checks, the f stands for fast
-fpcc: no_tfhe_typo no_dbg_log check_fmt lint_doc check_md_docs_are_tested clippy_fast \
+fpcc: no_tfhe_typo no_dbg_log check_fmt check_typos lint_doc check_md_docs_are_tested clippy_fast \
 check_compile_tests
 
 .PHONY: conformance # Automatically fix problems that can be fixed

README.md

@@ -159,7 +159,7 @@ To run this code, use the following command:
 > Note that when running code that uses `TFHE-rs`, it is highly recommended
 to run in release mode with cargo's `--release` flag to have the best performances possible.
 
-*Find an example with more explanations in [this part of the documentation](https://docs.zama.ai/tfhe-rs/getting-started/quick_start)*
+*Find an example with more explanations in [this part of the documentation](https://docs.zama.ai/tfhe-rs/get-started/quick_start)*
 
 <p align="right">
   <a href="#about" > ↑ Back to top </a> 
@@ -208,7 +208,7 @@ When a new update is published in the Lattice Estimator, we update parameters ac
 
 ### Security model
 
-The default parameters for the TFHE-rs library are chosen considering the IND-CPA security model, and are selected with a bootstrapping failure probability fixed at p_error = $2^{-40}$. In particular, it is assumed that the results of decrypted computations are not shared by the secret key owner with any third parties, as such an action can lead to leakage of the secret encryption key. If you are designing an application where decryptions must be shared, you will need to craft custom encryption parameters which are chosen in consideration of the IND-CPA^D security model [1]. 
+The default parameters for the TFHE-rs library are chosen considering the IND-CPA security model, and are selected with a bootstrapping failure probability fixed at p_error = $2^{-64}$. In particular, it is assumed that the results of decrypted computations are not shared by the secret key owner with any third parties, as such an action can lead to leakage of the secret encryption key. If you are designing an application where decryptions must be shared, you will need to craft custom encryption parameters which are chosen in consideration of the IND-CPA^D security model [1]. 
 
 [1] Li, Baiyu, et al. "Securing approximate homomorphic encryption using differential privacy." Annual International Cryptology Conference. Cham: Springer Nature Switzerland, 2022. https://eprint.iacr.org/2022/816.pdf
 

_typos.toml

@@ -0,0 +1,15 @@
+[default]
+extend-ignore-identifiers-re = [
+    # Related to serialized object
+    "ser",
+    "unser",
+    # Used when dumping tfhe-rs parameters set into Sage format
+    "ND.*",
+    # Related to FHE strings example handling "banana"
+    "ba",
+    "enc_ba",
+    # Example with string replacing "hello" with "herlo"
+    "herlo",
+    # Example in trivium
+    "C9217BA0D762ACA1"
+]

apps/trivium/benches/kreyvium_byte.rs

@@ -4,9 +4,8 @@ use tfhe::{generate_keys, ConfigBuilder, FheUint64, FheUint8};
 use tfhe_trivium::{KreyviumStreamByte, TransCiphering};
 
 pub fn kreyvium_byte_gen(c: &mut Criterion) {
-    let config = ConfigBuilder::default()
-        .enable_function_evaluation()
-        .build();
+    let config = ConfigBuilder::default().build();
+
     let (client_key, server_key) = generate_keys(config);
 
     let key_string = "0053A6F94C9FF24598EB000000000000".to_string();
@@ -33,9 +32,8 @@ pub fn kreyvium_byte_gen(c: &mut Criterion) {
 }
 
 pub fn kreyvium_byte_trans(c: &mut Criterion) {
-    let config = ConfigBuilder::default()
-        .enable_function_evaluation()
-        .build();
+    let config = ConfigBuilder::default().build();
+
     let (client_key, server_key) = generate_keys(config);
 
     let key_string = "0053A6F94C9FF24598EB000000000000".to_string();
@@ -63,9 +61,8 @@ pub fn kreyvium_byte_trans(c: &mut Criterion) {
 }
 
 pub fn kreyvium_byte_warmup(c: &mut Criterion) {
-    let config = ConfigBuilder::default()
-        .enable_function_evaluation()
-        .build();
+    let config = ConfigBuilder::default().build();
+
     let (client_key, server_key) = generate_keys(config);
 
     let key_string = "0053A6F94C9FF24598EB000000000000".to_string();

apps/trivium/src/kreyvium/kreyvium.rs

@@ -148,10 +148,9 @@ where
 
     /// Computes one turn of the stream, updating registers and outputting the new bit.
     pub fn next_bool(&mut self) -> T {
-        match &self.fhe_key {
-            Some(sk) => set_server_key(sk.clone()),
-            None => (),
-        };
+        if let Some(sk) = &self.fhe_key {
+            set_server_key(sk.clone());
+        }
 
         let [o, a, b, c] = self.get_output_and_values(0);
 
@@ -226,18 +225,12 @@ where
     /// Computes 64 turns of the stream, outputting the 64 bits all at once in a
     /// Vec (first value is oldest, last is newest)
     pub fn next_64(&mut self) -> Vec<T> {
-        match &self.fhe_key {
-            Some(sk) => {
-                rayon::broadcast(|_| set_server_key(sk.clone()));
-            }
-            None => (),
+        if let Some(sk) = &self.fhe_key {
+            rayon::broadcast(|_| set_server_key(sk.clone()));
         }
         let mut values = self.get_64_output_and_values();
-        match &self.fhe_key {
-            Some(_) => {
-                rayon::broadcast(|_| unset_server_key());
-            }
-            None => (),
+        if self.fhe_key.is_some() {
+            rayon::broadcast(|_| unset_server_key());
         }
 
         let mut ret = Vec::<T>::with_capacity(64);

apps/trivium/src/kreyvium/kreyvium_byte.rs

@@ -119,7 +119,7 @@ impl KreyviumStreamByte<FheUint8> {
         }
 
         // Key and iv are stored in reverse in their shift registers
-        let mut key = key_bytes.map(|b| b.map(|x| (x as u8).reverse_bits() as u64));
+        let mut key = key_bytes.map(|b| b.reverse_bits());
         let mut iv = iv_bytes.map(|x| FheUint8::encrypt_trivial(x.reverse_bits()));
         key.reverse();
         iv.reverse();
@@ -237,18 +237,12 @@ where
     /// Computes 64 turns of the stream, outputting the 64 bits (in 8 bytes) all at once in a
     /// Vec (first value is oldest, last is newest)
     pub fn next_64(&mut self) -> Vec<T> {
-        match &self.fhe_key {
-            Some(sk) => {
-                rayon::broadcast(|_| set_server_key(sk.clone()));
-            }
-            None => (),
+        if let Some(sk) = &self.fhe_key {
+            rayon::broadcast(|_| set_server_key(sk.clone()));
         }
         let values = self.get_64_output_and_values();
-        match &self.fhe_key {
-            Some(_) => {
-                rayon::broadcast(|_| unset_server_key());
-            }
-            None => (),
+        if self.fhe_key.is_some() {
+            rayon::broadcast(|_| unset_server_key());
         }
 
         let mut bytes = Vec::<T>::with_capacity(8);

apps/trivium/src/kreyvium/test.rs

@@ -299,9 +299,8 @@ fn kreyvium_test_clear_byte() {
 
 #[test]
 fn kreyvium_test_byte_long() {
-    let config = ConfigBuilder::default()
-        .enable_function_evaluation()
-        .build();
+    let config = ConfigBuilder::default().build();
+
     let (client_key, server_key) = generate_keys(config);
 
     let key_string = "0053A6F94C9FF24598EB000000000000".to_string();
@@ -338,9 +337,8 @@ fn kreyvium_test_byte_long() {
 
 #[test]
 fn kreyvium_test_fhe_byte_transciphering_long() {
-    let config = ConfigBuilder::default()
-        .enable_function_evaluation()
-        .build();
+    let config = ConfigBuilder::default().build();
+
     let (client_key, server_key) = generate_keys(config);
 
     let key_string = "0053A6F94C9FF24598EB000000000000".to_string();

apps/trivium/src/lib.rs

@@ -1,3 +1,5 @@
+#![allow(clippy::too_long_first_doc_paragraph)]
+
 mod static_deque;
 
 mod kreyvium;

apps/trivium/src/trivium/trivium_bool.rs

@@ -120,10 +120,9 @@ where
 
     /// Computes one turn of the stream, updating registers and outputting the new bit.
     pub fn next_bool(&mut self) -> T {
-        match &self.fhe_key {
-            Some(sk) => set_server_key(sk.clone()),
-            None => (),
-        };
+        if let Some(sk) = &self.fhe_key {
+            set_server_key(sk.clone());
+        }
 
         let [o, a, b, c] = self.get_output_and_values(0);
 
@@ -196,18 +195,12 @@ where
     /// Computes 64 turns of the stream, outputting the 64 bits all at once in a
     /// Vec (first value is oldest, last is newest)
     pub fn next_64(&mut self) -> Vec<T> {
-        match &self.fhe_key {
-            Some(sk) => {
-                rayon::broadcast(|_| set_server_key(sk.clone()));
-            }
-            None => (),
+        if let Some(sk) = &self.fhe_key {
+            rayon::broadcast(|_| set_server_key(sk.clone()));
         }
         let mut values = self.get_64_output_and_values();
-        match &self.fhe_key {
-            Some(_) => {
-                rayon::broadcast(|_| unset_server_key());
-            }
-            None => (),
+        if self.fhe_key.is_some() {
+            rayon::broadcast(|_| unset_server_key());
         }
 
         let mut ret = Vec::<T>::with_capacity(64);

apps/trivium/src/trivium/trivium_byte.rs

@@ -187,18 +187,12 @@ where
     /// Computes 64 turns of the stream, outputting the 64 bits (in 8 bytes) all at once in a
     /// Vec (first value is oldest, last is newest)
     pub fn next_64(&mut self) -> Vec<T> {
-        match &self.fhe_key {
-            Some(sk) => {
-                rayon::broadcast(|_| set_server_key(sk.clone()));
-            }
-            None => (),
+        if let Some(sk) = &self.fhe_key {
+            rayon::broadcast(|_| set_server_key(sk.clone()));
         }
         let values = self.get_64_output_and_values();
-        match &self.fhe_key {
-            Some(_) => {
-                rayon::broadcast(|_| unset_server_key());
-            }
-            None => (),
+        if self.fhe_key.is_some() {
+            rayon::broadcast(|_| unset_server_key());
         }
 
         let mut bytes = Vec::<T>::with_capacity(8);

backends/tfhe-cuda-backend/Cargo.toml

@@ -1,6 +1,6 @@
 [package]
 name = "tfhe-cuda-backend"
-version = "0.3.0"
+version = "0.4.0"
 edition = "2021"
 authors = ["Zama team"]
 license = "BSD-3-Clause-Clear"

backends/tfhe-cuda-backend/build.rs

@@ -8,6 +8,18 @@ fn main() {
         }
     }
 
+    // This is a workaround to the current nightly toolchain (2024-06-27 which started with
+    // toolchain 2024-05-05) build issue
+    // Essentially if cbindgen is running, a wrong argument ends up forwarded to the cuda backend
+    // "make" command during macro expansions for TFHE-rs C API, crashing make for make < 4.4 and
+    // thus crashing the build
+    // On the other hand, this speeds up C API build greatly given we don't have macro expansions
+    // in the CUDA backend so this skips the second compilation of TFHE-rs for macro inspection by
+    // cbindgen
+    if std::env::var("_CBINDGEN_IS_RUNNING").is_ok() {
+        return;
+    }
+
     println!("Build tfhe-cuda-backend");
     println!("cargo::rerun-if-changed=cuda/include");
     println!("cargo::rerun-if-changed=cuda/src");

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

@@ -67,9 +67,21 @@ endif()
 
 add_compile_definitions(CUDA_ARCH=${CUDA_ARCH})
 
+# Check if the DEBUG flag is defined
+if(CMAKE_BUILD_TYPE STREQUAL "Debug")
+  # Debug mode
+  message("Compiling in Debug mode")
+  add_definitions(-DDEBUG)
+  set(OPTIMIZATION_FLAGS "${OPTIMIZATION_FLAGS} -O0 -G -g")
+else()
+  # Release mode
+  message("Compiling in Release mode")
+  set(OPTIMIZATION_FLAGS "${OPTIMIZATION_FLAGS} -O3")
+endif()
+
 # in production, should use -arch=sm_70 --ptxas-options=-v to see register spills -lineinfo for better debugging
 set(CMAKE_CUDA_FLAGS
-    "${CMAKE_CUDA_FLAGS} -ccbin ${CMAKE_CXX_COMPILER} -O3 \
+    "${CMAKE_CUDA_FLAGS} -ccbin ${CMAKE_CXX_COMPILER} ${OPTIMIZATION_FLAGS}\
   -std=c++17 --no-exceptions  --expt-relaxed-constexpr -rdc=true \
   --use_fast_math -Xcompiler -fPIC")
 

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

@@ -1,6 +1,7 @@
 #ifndef CUDA_CIPHERTEXT_H
 #define CUDA_CIPHERTEXT_H
 
+#include "device.h"
 #include <cstdint>
 
 extern "C" {
@@ -14,5 +15,11 @@ void cuda_convert_lwe_ciphertext_vector_to_cpu_64(void *stream,
                                                   void *dest, void *src,
                                                   uint32_t number_of_cts,
                                                   uint32_t lwe_dimension);
+
+void cuda_glwe_sample_extract_64(void *stream, uint32_t gpu_index,
+                                 void *lwe_array_out, void *glwe_array_in,
+                                 uint32_t *nth_array, uint32_t num_nths,
+                                 uint32_t glwe_dimension,
+                                 uint32_t polynomial_size);
 };
 #endif

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

@@ -0,0 +1,163 @@
+#ifndef CUDA_INTEGER_COMPRESSION_H
+#define CUDA_INTEGER_COMPRESSION_H
+
+#include "integer.h"
+
+extern "C" {
+void scratch_cuda_integer_compress_radix_ciphertext_64(
+    void **streams, uint32_t *gpu_indexes, uint32_t gpu_count, int8_t **mem_ptr,
+    uint32_t compression_glwe_dimension, uint32_t compression_polynomial_size,
+    uint32_t lwe_dimension, uint32_t ks_level, uint32_t ks_base_log,
+    uint32_t num_radix_blocks, uint32_t message_modulus, uint32_t carry_modulus,
+    PBS_TYPE pbs_type, uint32_t lwe_per_glwe, uint32_t storage_log_modulus,
+    bool allocate_gpu_memory);
+
+void scratch_cuda_integer_decompress_radix_ciphertext_64(
+    void **streams, uint32_t *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, uint32_t compression_polynomial_size,
+    uint32_t lwe_dimension, uint32_t pbs_level, uint32_t pbs_base_log,
+    uint32_t num_radix_blocks, uint32_t message_modulus, uint32_t carry_modulus,
+    PBS_TYPE pbs_type, uint32_t storage_log_modulus, uint32_t body_count,
+    bool allocate_gpu_memory);
+
+void cuda_integer_compress_radix_ciphertext_64(
+    void **streams, uint32_t *gpu_indexes, uint32_t gpu_count,
+    void *glwe_array_out, void *lwe_array_in, void **fp_ksk, uint32_t num_nths,
+    int8_t *mem_ptr);
+
+void cuda_integer_decompress_radix_ciphertext_64(
+    void **streams, uint32_t *gpu_indexes, uint32_t gpu_count,
+    void *lwe_array_out, void *glwe_in, uint32_t *indexes_array,
+    uint32_t indexes_array_size, void **bsks, int8_t *mem_ptr);
+
+void cleanup_cuda_integer_compress_radix_ciphertext_64(void **streams,
+                                                       uint32_t *gpu_indexes,
+                                                       uint32_t gpu_count,
+                                                       int8_t **mem_ptr_void);
+
+void cleanup_cuda_integer_decompress_radix_ciphertext_64(void **streams,
+                                                         uint32_t *gpu_indexes,
+                                                         uint32_t gpu_count,
+                                                         int8_t **mem_ptr_void);
+}
+
+template <typename Torus> struct int_compression {
+  int_radix_params compression_params;
+  uint32_t storage_log_modulus;
+  uint32_t lwe_per_glwe;
+
+  uint32_t body_count;
+
+  // Compression
+  int8_t *fp_ks_buffer;
+  Torus *tmp_lwe;
+  Torus *tmp_glwe_array_out;
+
+  int_compression(cudaStream_t *streams, uint32_t *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) {
+    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;
+
+      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]);
+
+      scratch_packing_keyswitch_lwe_list_to_glwe_64(
+          streams[0], gpu_indexes[0], &fp_ks_buffer,
+          compression_params.glwe_dimension, compression_params.polynomial_size,
+          num_radix_blocks, true);
+    }
+  }
+  void release(cudaStream_t *streams, uint32_t *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);
+  }
+};
+
+template <typename Torus> struct int_decompression {
+  int_radix_params encryption_params;
+  int_radix_params compression_params;
+
+  uint32_t storage_log_modulus;
+
+  uint32_t num_radix_blocks;
+  uint32_t body_count;
+
+  Torus *tmp_extracted_glwe;
+  Torus *tmp_extracted_lwe;
+  uint32_t *tmp_indexes_array;
+
+  int_radix_lut<Torus> *carry_extract_lut;
+
+  int_decompression(cudaStream_t *streams, uint32_t *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) {
+    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);
+      carry_extract_lut = new int_radix_lut<Torus>(
+          streams, gpu_indexes, gpu_count, encryption_params, 1,
+          num_radix_blocks, allocate_gpu_memory);
+
+      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]);
+
+      // Carry extract LUT
+      auto carry_extract_f = [encryption_params](Torus x) -> Torus {
+        return x / encryption_params.message_modulus;
+      };
+
+      generate_device_accumulator<Torus>(
+          streams[0], gpu_indexes[0],
+          carry_extract_lut->get_lut(gpu_indexes[0], 0),
+          encryption_params.glwe_dimension, encryption_params.polynomial_size,
+          encryption_params.message_modulus, encryption_params.carry_modulus,
+          carry_extract_f);
+
+      carry_extract_lut->broadcast_lut(streams, gpu_indexes, gpu_indexes[0]);
+    }
+  }
+  void release(cudaStream_t *streams, uint32_t *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]);
+
+    carry_extract_lut->release(streams, gpu_indexes, gpu_count);
+    delete carry_extract_lut;
+  }
+};
+#endif

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

@@ -39,16 +39,15 @@ 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_support_cooperative_groups();
-
-bool cuda_check_support_thread_block_clusters();
-
 void cuda_memcpy_async_to_gpu(void *dest, void *src, uint64_t size,
                               cudaStream_t stream, uint32_t gpu_index);
 
 void cuda_memcpy_async_gpu_to_gpu(void *dest, void *src, uint64_t size,
                                   cudaStream_t stream, uint32_t gpu_index);
 
+void cuda_memcpy_gpu_to_gpu(void *dest, void *src, uint64_t size,
+                            uint32_t gpu_index);
+
 void cuda_memcpy_async_to_cpu(void *dest, const void *src, uint64_t size,
                               cudaStream_t stream, uint32_t gpu_index);
 
@@ -62,15 +61,13 @@ void cuda_synchronize_device(uint32_t gpu_index);
 void cuda_drop(void *ptr, uint32_t gpu_index);
 
 void cuda_drop_async(void *ptr, cudaStream_t stream, uint32_t gpu_index);
+}
 
 int cuda_get_max_shared_memory(uint32_t gpu_index);
 
-void cuda_stream_add_callback(cudaStream_t stream, uint32_t gpu_index,
-                              cudaStreamCallback_t callback, void *user_data);
-}
+bool cuda_check_support_cooperative_groups();
 
-void host_free_on_stream_callback(cudaStream_t stream, cudaError_t status,
-                                  void *host_pointer);
+bool cuda_check_support_thread_block_clusters();
 
 template <typename Torus>
 void cuda_set_value_async(cudaStream_t stream, uint32_t gpu_index,

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

@@ -1,14 +1,30 @@
 #ifndef HELPER_MULTI_GPU_H
 #define HELPER_MULTI_GPU_H
 #include <mutex>
+#include <variant>
+#include <vector>
 
 extern std::mutex m;
 extern bool p2p_enabled;
 
 extern "C" {
-int cuda_setup_multi_gpu();
+int32_t cuda_setup_multi_gpu();
 }
 
+// Define a variant type that can be either a vector or a single pointer
+template <typename Torus>
+using LweArrayVariant = std::variant<std::vector<Torus *>, Torus *>;
+
+// Macro to define the visitor logic using std::holds_alternative for vectors
+#define GET_VARIANT_ELEMENT(variant, index)                                    \
+  [&] {                                                                        \
+    if (std::holds_alternative<std::vector<Torus *>>(variant)) {               \
+      return std::get<std::vector<Torus *>>(variant)[index];                   \
+    } else {                                                                   \
+      return std::get<Torus *>(variant);                                       \
+    }                                                                          \
+  }()
+
 int get_active_gpu_count(int num_inputs, int gpu_count);
 
 int get_num_inputs_on_gpu(int total_num_inputs, int gpu_index, int gpu_count);

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

@@ -9,15 +9,28 @@ void cuda_keyswitch_lwe_ciphertext_vector_32(
     void *stream, uint32_t gpu_index, void *lwe_array_out,
     void *lwe_output_indexes, void *lwe_array_in, void *lwe_input_indexes,
     void *ksk, uint32_t lwe_dimension_in, uint32_t lwe_dimension_out,
-    uint32_t base_log, uint32_t level_count, uint32_t num_samples,
-    uint32_t gpu_offset = 0);
+    uint32_t base_log, uint32_t level_count, uint32_t num_samples);
 
 void cuda_keyswitch_lwe_ciphertext_vector_64(
     void *stream, uint32_t gpu_index, void *lwe_array_out,
     void *lwe_output_indexes, void *lwe_array_in, void *lwe_input_indexes,
     void *ksk, uint32_t lwe_dimension_in, uint32_t lwe_dimension_out,
-    uint32_t base_log, uint32_t level_count, uint32_t num_samples,
-    uint32_t gpu_offset = 0);
+    uint32_t base_log, uint32_t level_count, uint32_t num_samples);
+
+void scratch_packing_keyswitch_lwe_list_to_glwe_64(
+    void *stream, uint32_t gpu_index, int8_t **fp_ks_buffer,
+    uint32_t glwe_dimension, uint32_t polynomial_size, uint32_t num_lwes,
+    bool allocate_gpu_memory);
+
+void cuda_packing_keyswitch_lwe_list_to_glwe_64(
+    void *stream, uint32_t gpu_index, void *glwe_array_out, void *lwe_array_in,
+    void *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);
 }
 
 #endif // CNCRT_KS_H_

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

@@ -26,14 +26,12 @@ void cuda_convert_lwe_programmable_bootstrap_key_64(
 void scratch_cuda_programmable_bootstrap_amortized_32(
     void *stream, uint32_t gpu_index, int8_t **pbs_buffer,
     uint32_t glwe_dimension, uint32_t polynomial_size,
-    uint32_t input_lwe_ciphertext_count, uint32_t max_shared_memory,
-    bool allocate_gpu_memory);
+    uint32_t input_lwe_ciphertext_count, bool allocate_gpu_memory);
 
 void 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, uint32_t max_shared_memory,
-    bool allocate_gpu_memory);
+    uint32_t input_lwe_ciphertext_count, bool allocate_gpu_memory);
 
 void cuda_programmable_bootstrap_amortized_lwe_ciphertext_vector_32(
     void *stream, uint32_t gpu_index, void *lwe_array_out,
@@ -41,8 +39,7 @@ void cuda_programmable_bootstrap_amortized_lwe_ciphertext_vector_32(
     void *lwe_array_in, void *lwe_input_indexes, void *bootstrapping_key,
     int8_t *pbs_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_luts, uint32_t lwe_idx,
-    uint32_t max_shared_memory);
+    uint32_t num_samples);
 
 void cuda_programmable_bootstrap_amortized_lwe_ciphertext_vector_64(
     void *stream, uint32_t gpu_index, void *lwe_array_out,
@@ -50,8 +47,7 @@ void cuda_programmable_bootstrap_amortized_lwe_ciphertext_vector_64(
     void *lwe_array_in, void *lwe_input_indexes, void *bootstrapping_key,
     int8_t *pbs_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_luts, uint32_t lwe_idx,
-    uint32_t max_shared_memory, uint32_t gpu_offset = 0);
+    uint32_t num_samples);
 
 void cleanup_cuda_programmable_bootstrap_amortized(void *stream,
                                                    uint32_t gpu_index,
@@ -60,14 +56,12 @@ void cleanup_cuda_programmable_bootstrap_amortized(void *stream,
 void scratch_cuda_programmable_bootstrap_32(
     void *stream, uint32_t gpu_index, int8_t **buffer, uint32_t glwe_dimension,
     uint32_t polynomial_size, uint32_t level_count,
-    uint32_t input_lwe_ciphertext_count, uint32_t max_shared_memory,
-    bool allocate_gpu_memory);
+    uint32_t input_lwe_ciphertext_count, bool allocate_gpu_memory);
 
 void scratch_cuda_programmable_bootstrap_64(
     void *stream, uint32_t gpu_index, int8_t **buffer, uint32_t glwe_dimension,
     uint32_t polynomial_size, uint32_t level_count,
-    uint32_t input_lwe_ciphertext_count, uint32_t max_shared_memory,
-    bool allocate_gpu_memory);
+    uint32_t input_lwe_ciphertext_count, bool allocate_gpu_memory);
 
 void cuda_programmable_bootstrap_lwe_ciphertext_vector_32(
     void *stream, uint32_t gpu_index, void *lwe_array_out,
@@ -75,8 +69,7 @@ void cuda_programmable_bootstrap_lwe_ciphertext_vector_32(
     void *lwe_array_in, void *lwe_input_indexes, void *bootstrapping_key,
     int8_t *buffer, uint32_t lwe_dimension, uint32_t glwe_dimension,
     uint32_t polynomial_size, uint32_t base_log, uint32_t level_count,
-    uint32_t num_samples, uint32_t num_luts, uint32_t lwe_idx,
-    uint32_t max_shared_memory, uint32_t gpu_offset = 0);
+    uint32_t num_samples, uint32_t lut_count, uint32_t lut_stride);
 
 void cuda_programmable_bootstrap_lwe_ciphertext_vector_64(
     void *stream, uint32_t gpu_index, void *lwe_array_out,
@@ -84,44 +77,33 @@ void cuda_programmable_bootstrap_lwe_ciphertext_vector_64(
     void *lwe_array_in, void *lwe_input_indexes, void *bootstrapping_key,
     int8_t *buffer, uint32_t lwe_dimension, uint32_t glwe_dimension,
     uint32_t polynomial_size, uint32_t base_log, uint32_t level_count,
-    uint32_t num_samples, uint32_t num_luts, uint32_t lwe_idx,
-    uint32_t max_shared_memory, uint32_t gpu_offset = 0);
+    uint32_t num_samples, uint32_t lut_count, uint32_t lut_stride);
 
 void cleanup_cuda_programmable_bootstrap(void *stream, uint32_t gpu_index,
                                          int8_t **pbs_buffer);
-
-uint64_t get_buffer_size_programmable_bootstrap_amortized_64(
-    uint32_t glwe_dimension, uint32_t polynomial_size,
-    uint32_t input_lwe_ciphertext_count, uint32_t max_shared_memory);
-
-uint64_t get_buffer_size_programmable_bootstrap_64(
-    uint32_t glwe_dimension, uint32_t polynomial_size, uint32_t level_count,
-    uint32_t input_lwe_ciphertext_count, uint32_t max_shared_memory);
 }
 
 template <typename Torus>
-__host__ __device__ uint64_t
-get_buffer_size_full_sm_programmable_bootstrap_step_one(
+uint64_t get_buffer_size_full_sm_programmable_bootstrap_step_one(
     uint32_t polynomial_size) {
   return sizeof(Torus) * polynomial_size +      // accumulator_rotated
          sizeof(double2) * polynomial_size / 2; // accumulator fft
 }
 template <typename Torus>
-__host__ __device__ uint64_t
-get_buffer_size_full_sm_programmable_bootstrap_step_two(
+uint64_t get_buffer_size_full_sm_programmable_bootstrap_step_two(
     uint32_t polynomial_size) {
   return sizeof(Torus) * polynomial_size +      // accumulator
          sizeof(double2) * polynomial_size / 2; // accumulator fft
 }
 
 template <typename Torus>
-__host__ __device__ uint64_t
+uint64_t
 get_buffer_size_partial_sm_programmable_bootstrap(uint32_t polynomial_size) {
   return sizeof(double2) * polynomial_size / 2; // accumulator fft
 }
 
 template <typename Torus>
-__host__ __device__ uint64_t
+uint64_t
 get_buffer_size_full_sm_programmable_bootstrap_tbc(uint32_t polynomial_size) {
   return sizeof(Torus) * polynomial_size +      // accumulator_rotated
          sizeof(Torus) * polynomial_size +      // accumulator
@@ -129,21 +111,19 @@ get_buffer_size_full_sm_programmable_bootstrap_tbc(uint32_t polynomial_size) {
 }
 
 template <typename Torus>
-__host__ __device__ uint64_t
-get_buffer_size_partial_sm_programmable_bootstrap_tbc(
+uint64_t get_buffer_size_partial_sm_programmable_bootstrap_tbc(
     uint32_t polynomial_size) {
   return sizeof(double2) * polynomial_size / 2; // accumulator fft mask & body
 }
 
 template <typename Torus>
-__host__ __device__ uint64_t
-get_buffer_size_sm_dsm_plus_tbc_classic_programmable_bootstrap(
+uint64_t get_buffer_size_sm_dsm_plus_tbc_classic_programmable_bootstrap(
     uint32_t polynomial_size) {
   return sizeof(double2) * polynomial_size / 2; // tbc
 }
 
 template <typename Torus>
-__host__ __device__ uint64_t
+uint64_t
 get_buffer_size_full_sm_programmable_bootstrap_cg(uint32_t polynomial_size) {
   return sizeof(Torus) * polynomial_size +      // accumulator_rotated
          sizeof(Torus) * polynomial_size +      // accumulator
@@ -151,15 +131,14 @@ get_buffer_size_full_sm_programmable_bootstrap_cg(uint32_t polynomial_size) {
 }
 
 template <typename Torus>
-__host__ __device__ uint64_t
+uint64_t
 get_buffer_size_partial_sm_programmable_bootstrap_cg(uint32_t polynomial_size) {
   return sizeof(double2) * polynomial_size / 2; // accumulator fft mask & body
 }
 
 template <typename Torus>
-__host__ bool
-supports_distributed_shared_memory_on_classic_programmable_bootstrap(
-    uint32_t polynomial_size, uint32_t max_shared_memory);
+bool supports_distributed_shared_memory_on_classic_programmable_bootstrap(
+    uint32_t polynomial_size);
 
 template <typename Torus, PBS_TYPE pbs_type> struct pbs_buffer;
 
@@ -178,7 +157,7 @@ template <typename Torus> struct pbs_buffer<Torus, PBS_TYPE::CLASSICAL> {
 
     this->pbs_variant = pbs_variant;
 
-    auto max_shared_memory = cuda_get_max_shared_memory(gpu_index);
+    auto max_shared_memory = cuda_get_max_shared_memory(0);
 
     if (allocate_gpu_memory) {
       switch (pbs_variant) {
@@ -255,7 +234,7 @@ template <typename Torus> struct pbs_buffer<Torus, PBS_TYPE::CLASSICAL> {
 
         bool supports_dsm =
             supports_distributed_shared_memory_on_classic_programmable_bootstrap<
-                Torus>(polynomial_size, max_shared_memory);
+                Torus>(polynomial_size);
 
         uint64_t full_sm =
             get_buffer_size_full_sm_programmable_bootstrap_tbc<Torus>(
@@ -314,10 +293,10 @@ template <typename Torus> struct pbs_buffer<Torus, PBS_TYPE::CLASSICAL> {
 };
 
 template <typename Torus>
-__host__ __device__ uint64_t get_buffer_size_programmable_bootstrap_cg(
+uint64_t get_buffer_size_programmable_bootstrap_cg(
     uint32_t glwe_dimension, uint32_t polynomial_size, uint32_t level_count,
-    uint32_t input_lwe_ciphertext_count, uint32_t max_shared_memory) {
-
+    uint32_t input_lwe_ciphertext_count) {
+  int max_shared_memory = cuda_get_max_shared_memory(0);
   uint64_t full_sm =
       get_buffer_size_full_sm_programmable_bootstrap_cg<Torus>(polynomial_size);
   uint64_t partial_sm =
@@ -343,8 +322,7 @@ template <typename Torus>
 bool has_support_to_cuda_programmable_bootstrap_cg(uint32_t glwe_dimension,
                                                    uint32_t polynomial_size,
                                                    uint32_t level_count,
-                                                   uint32_t num_samples,
-                                                   uint32_t max_shared_memory);
+                                                   uint32_t num_samples);
 
 template <typename Torus>
 void cuda_programmable_bootstrap_cg_lwe_ciphertext_vector(
@@ -353,8 +331,8 @@ void cuda_programmable_bootstrap_cg_lwe_ciphertext_vector(
     Torus *lwe_array_in, Torus *lwe_input_indexes, double2 *bootstrapping_key,
     pbs_buffer<Torus, CLASSICAL> *buffer, uint32_t lwe_dimension,
     uint32_t glwe_dimension, uint32_t polynomial_size, uint32_t base_log,
-    uint32_t level_count, uint32_t num_samples, uint32_t num_luts,
-    uint32_t lwe_idx, uint32_t max_shared_memory, uint32_t gpu_offset = 0);
+    uint32_t level_count, uint32_t num_samples, uint32_t lut_count,
+    uint32_t lut_stride);
 
 template <typename Torus>
 void cuda_programmable_bootstrap_lwe_ciphertext_vector(
@@ -363,8 +341,8 @@ void cuda_programmable_bootstrap_lwe_ciphertext_vector(
     Torus *lwe_array_in, Torus *lwe_input_indexes, double2 *bootstrapping_key,
     pbs_buffer<Torus, CLASSICAL> *buffer, uint32_t lwe_dimension,
     uint32_t glwe_dimension, uint32_t polynomial_size, uint32_t base_log,
-    uint32_t level_count, uint32_t num_samples, uint32_t num_luts,
-    uint32_t lwe_idx, uint32_t max_shared_memory, uint32_t gpu_offset = 0);
+    uint32_t level_count, uint32_t num_samples, uint32_t lut_count,
+    uint32_t lut_stride);
 
 #if (CUDA_ARCH >= 900)
 template <typename Torus>
@@ -374,43 +352,45 @@ void cuda_programmable_bootstrap_tbc_lwe_ciphertext_vector(
     Torus *lwe_array_in, Torus *lwe_input_indexes, double2 *bootstrapping_key,
     pbs_buffer<Torus, CLASSICAL> *buffer, uint32_t lwe_dimension,
     uint32_t glwe_dimension, uint32_t polynomial_size, uint32_t base_log,
-    uint32_t level_count, uint32_t num_samples, uint32_t num_luts,
-    uint32_t lwe_idx, uint32_t max_shared_memory, uint32_t gpu_offset = 0);
+    uint32_t level_count, uint32_t num_samples, uint32_t lut_count,
+    uint32_t lut_stride);
 
-template <typename Torus, typename STorus>
+template <typename Torus>
 void scratch_cuda_programmable_bootstrap_tbc(
     void *stream, uint32_t gpu_index, pbs_buffer<Torus, CLASSICAL> **pbs_buffer,
     uint32_t glwe_dimension, uint32_t polynomial_size, uint32_t level_count,
-    uint32_t input_lwe_ciphertext_count, uint32_t max_shared_memory,
-    bool allocate_gpu_memory);
+    uint32_t input_lwe_ciphertext_count, bool allocate_gpu_memory);
 #endif
 
-template <typename Torus, typename STorus>
+template <typename Torus>
 void scratch_cuda_programmable_bootstrap_cg(
     void *stream, uint32_t gpu_index, pbs_buffer<Torus, CLASSICAL> **pbs_buffer,
     uint32_t glwe_dimension, uint32_t polynomial_size, uint32_t level_count,
-    uint32_t input_lwe_ciphertext_count, uint32_t max_shared_memory,
-    bool allocate_gpu_memory);
+    uint32_t input_lwe_ciphertext_count, bool allocate_gpu_memory);
 
-template <typename Torus, typename STorus>
+template <typename Torus>
 void scratch_cuda_programmable_bootstrap(
     void *stream, uint32_t gpu_index, pbs_buffer<Torus, CLASSICAL> **buffer,
     uint32_t glwe_dimension, uint32_t polynomial_size, uint32_t level_count,
-    uint32_t input_lwe_ciphertext_count, uint32_t max_shared_memory,
-    bool allocate_gpu_memory);
+    uint32_t input_lwe_ciphertext_count, bool allocate_gpu_memory);
 
 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 max_shared_memory);
+                                                    uint32_t level_count);
 
 #ifdef __CUDACC__
 __device__ inline int get_start_ith_ggsw(int i, uint32_t polynomial_size,
                                          int glwe_dimension,
                                          uint32_t level_count);
 
+template <typename T>
+__device__ const T *get_ith_mask_kth_block(const T *ptr, int i, int k,
+                                           int level, uint32_t polynomial_size,
+                                           int glwe_dimension,
+                                           uint32_t level_count);
+
 template <typename T>
 __device__ T *get_ith_mask_kth_block(T *ptr, int i, int k, int level,
                                      uint32_t polynomial_size,
@@ -422,8 +402,8 @@ __device__ T *get_ith_body_kth_block(T *ptr, int i, int k, int level,
                                      int glwe_dimension, uint32_t level_count);
 
 template <typename T>
-__device__ T *get_multi_bit_ith_lwe_gth_group_kth_block(
-    T *ptr, int g, int i, int k, int level, uint32_t grouping_factor,
+__device__ const T *get_multi_bit_ith_lwe_gth_group_kth_block(
+    const T *ptr, int g, int i, int k, int level, uint32_t grouping_factor,
     uint32_t polynomial_size, uint32_t glwe_dimension, uint32_t level_count);
 
 #endif

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

@@ -8,7 +8,7 @@ extern "C" {
 
 bool has_support_to_cuda_programmable_bootstrap_cg_multi_bit(
     uint32_t glwe_dimension, uint32_t polynomial_size, uint32_t level_count,
-    uint32_t num_samples, uint32_t max_shared_memory);
+    uint32_t num_samples);
 
 void cuda_convert_lwe_multi_bit_programmable_bootstrap_key_64(
     void *stream, uint32_t gpu_index, void *dest, void *src,
@@ -17,10 +17,8 @@ void cuda_convert_lwe_multi_bit_programmable_bootstrap_key_64(
 
 void scratch_cuda_multi_bit_programmable_bootstrap_64(
     void *stream, uint32_t gpu_index, int8_t **pbs_buffer,
-    uint32_t lwe_dimension, uint32_t glwe_dimension, uint32_t polynomial_size,
-    uint32_t level_count, uint32_t grouping_factor,
-    uint32_t input_lwe_ciphertext_count, uint32_t max_shared_memory,
-    bool allocate_gpu_memory, uint32_t chunk_size = 0);
+    uint32_t glwe_dimension, uint32_t polynomial_size, uint32_t level_count,
+    uint32_t input_lwe_ciphertext_count, bool allocate_gpu_memory);
 
 void cuda_multi_bit_programmable_bootstrap_lwe_ciphertext_vector_64(
     void *stream, uint32_t gpu_index, void *lwe_array_out,
@@ -28,9 +26,8 @@ void cuda_multi_bit_programmable_bootstrap_lwe_ciphertext_vector_64(
     void *lwe_array_in, void *lwe_input_indexes, void *bootstrapping_key,
     int8_t *buffer, uint32_t lwe_dimension, uint32_t glwe_dimension,
     uint32_t polynomial_size, uint32_t grouping_factor, uint32_t base_log,
-    uint32_t level_count, uint32_t num_samples, uint32_t num_luts,
-    uint32_t lwe_idx, uint32_t max_shared_memory, uint32_t gpu_offset,
-    uint32_t lwe_chunk_size = 0);
+    uint32_t level_count, uint32_t num_samples, uint32_t lut_count,
+    uint32_t lut_stride);
 
 void cleanup_cuda_multi_bit_programmable_bootstrap(void *stream,
                                                    uint32_t gpu_index,
@@ -38,23 +35,20 @@ void cleanup_cuda_multi_bit_programmable_bootstrap(void *stream,
 }
 
 template <typename Torus>
-__host__ bool
-supports_distributed_shared_memory_on_multibit_programmable_bootstrap(
-    uint32_t polynomial_size, uint32_t max_shared_memory);
+bool supports_distributed_shared_memory_on_multibit_programmable_bootstrap(
+    uint32_t polynomial_size);
 
 template <typename Torus>
 bool has_support_to_cuda_programmable_bootstrap_tbc_multi_bit(
     uint32_t num_samples, uint32_t glwe_dimension, uint32_t polynomial_size,
-    uint32_t level_count, uint32_t max_shared_memory);
+    uint32_t level_count);
 
 #if CUDA_ARCH >= 900
-template <typename Torus, typename STorus>
+template <typename Torus>
 void scratch_cuda_tbc_multi_bit_programmable_bootstrap(
     void *stream, uint32_t gpu_index, pbs_buffer<Torus, MULTI_BIT> **buffer,
-    uint32_t lwe_dimension, uint32_t glwe_dimension, uint32_t polynomial_size,
-    uint32_t level_count, uint32_t grouping_factor,
-    uint32_t input_lwe_ciphertext_count, uint32_t max_shared_memory,
-    bool allocate_gpu_memory, uint32_t lwe_chunk_size);
+    uint32_t glwe_dimension, uint32_t polynomial_size, uint32_t level_count,
+    uint32_t input_lwe_ciphertext_count, bool allocate_gpu_memory);
 
 template <typename Torus>
 void cuda_tbc_multi_bit_programmable_bootstrap_lwe_ciphertext_vector(
@@ -64,24 +58,14 @@ void cuda_tbc_multi_bit_programmable_bootstrap_lwe_ciphertext_vector(
     pbs_buffer<Torus, MULTI_BIT> *pbs_buffer, uint32_t lwe_dimension,
     uint32_t glwe_dimension, uint32_t polynomial_size, uint32_t grouping_factor,
     uint32_t base_log, uint32_t level_count, uint32_t num_samples,
-    uint32_t num_luts, uint32_t lwe_idx, uint32_t max_shared_memory,
-    uint32_t gpu_offset, uint32_t lwe_chunk_size);
+    uint32_t lut_count, uint32_t lut_stride);
 #endif
 
-template <typename Torus, typename STorus>
-void scratch_cuda_cg_multi_bit_programmable_bootstrap(
-    void *stream, uint32_t gpu_index, pbs_buffer<Torus, MULTI_BIT> **pbs_buffer,
-    uint32_t lwe_dimension, uint32_t glwe_dimension, uint32_t polynomial_size,
-    uint32_t level_count, uint32_t grouping_factor,
-    uint32_t input_lwe_ciphertext_count, uint32_t max_shared_memory,
-    bool allocate_gpu_memory, uint32_t lwe_chunk_size = 0);
-
-template <typename Torus, typename STorus>
+template <typename Torus>
 void 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, uint32_t max_shared_memory,
-    bool allocate_gpu_memory, uint32_t lwe_chunk_size = 0);
+    uint32_t input_lwe_ciphertext_count, bool allocate_gpu_memory);
 
 template <typename Torus>
 void cuda_cg_multi_bit_programmable_bootstrap_lwe_ciphertext_vector(
@@ -91,16 +75,13 @@ void cuda_cg_multi_bit_programmable_bootstrap_lwe_ciphertext_vector(
     pbs_buffer<Torus, MULTI_BIT> *pbs_buffer, uint32_t lwe_dimension,
     uint32_t glwe_dimension, uint32_t polynomial_size, uint32_t grouping_factor,
     uint32_t base_log, uint32_t level_count, uint32_t num_samples,
-    uint32_t num_luts, uint32_t lwe_idx, uint32_t max_shared_memory,
-    uint32_t gpu_offset, uint32_t lwe_chunk_size = 0);
+    uint32_t lut_count, uint32_t lut_stride);
 
-template <typename Torus, typename STorus>
+template <typename Torus>
 void scratch_cuda_multi_bit_programmable_bootstrap(
     void *stream, uint32_t gpu_index, pbs_buffer<Torus, MULTI_BIT> **pbs_buffer,
-    uint32_t lwe_dimension, uint32_t glwe_dimension, uint32_t polynomial_size,
-    uint32_t level_count, uint32_t grouping_factor,
-    uint32_t input_lwe_ciphertext_count, uint32_t max_shared_memory,
-    bool allocate_gpu_memory, uint32_t lwe_chunk_size = 0);
+    uint32_t glwe_dimension, uint32_t polynomial_size, uint32_t level_count,
+    uint32_t input_lwe_ciphertext_count, bool allocate_gpu_memory);
 
 template <typename Torus>
 void cuda_multi_bit_programmable_bootstrap_lwe_ciphertext_vector(
@@ -110,44 +91,34 @@ void cuda_multi_bit_programmable_bootstrap_lwe_ciphertext_vector(
     pbs_buffer<Torus, MULTI_BIT> *pbs_buffer, uint32_t lwe_dimension,
     uint32_t glwe_dimension, uint32_t polynomial_size, uint32_t grouping_factor,
     uint32_t base_log, uint32_t level_count, uint32_t num_samples,
-    uint32_t num_luts, uint32_t lwe_idx, uint32_t max_shared_memory,
-    uint32_t gpu_offset, uint32_t lwe_chunk_size = 0);
+    uint32_t lut_count, uint32_t lut_stride);
 
 template <typename Torus>
-__host__ __device__ uint64_t
-get_buffer_size_full_sm_multibit_programmable_bootstrap_keybundle(
+uint64_t get_buffer_size_full_sm_multibit_programmable_bootstrap_keybundle(
     uint32_t polynomial_size);
 template <typename Torus>
-__host__ __device__ uint64_t
-get_buffer_size_full_sm_multibit_programmable_bootstrap_step_one(
+uint64_t get_buffer_size_full_sm_multibit_programmable_bootstrap_step_one(
     uint32_t polynomial_size);
 template <typename Torus>
-__host__ __device__ uint64_t
-get_buffer_size_full_sm_multibit_programmable_bootstrap_step_two(
+uint64_t get_buffer_size_full_sm_multibit_programmable_bootstrap_step_two(
     uint32_t polynomial_size);
 template <typename Torus>
-__host__ __device__ uint64_t
-get_buffer_size_partial_sm_multibit_programmable_bootstrap_step_one(
+uint64_t get_buffer_size_partial_sm_multibit_programmable_bootstrap_step_one(
     uint32_t polynomial_size);
 template <typename Torus>
-__host__ __device__ uint64_t
-get_buffer_size_full_sm_cg_multibit_programmable_bootstrap(
+uint64_t get_buffer_size_full_sm_cg_multibit_programmable_bootstrap(
     uint32_t polynomial_size);
 template <typename Torus>
-__host__ __device__ uint64_t
-get_buffer_size_partial_sm_cg_multibit_programmable_bootstrap(
+uint64_t get_buffer_size_partial_sm_cg_multibit_programmable_bootstrap(
     uint32_t polynomial_size);
 template <typename Torus>
-__host__ __device__ uint64_t
-get_buffer_size_sm_dsm_plus_tbc_multibit_programmable_bootstrap(
+uint64_t get_buffer_size_sm_dsm_plus_tbc_multibit_programmable_bootstrap(
     uint32_t polynomial_size);
 template <typename Torus>
-__host__ __device__ uint64_t
-get_buffer_size_partial_sm_tbc_multibit_programmable_bootstrap(
+uint64_t get_buffer_size_partial_sm_tbc_multibit_programmable_bootstrap(
     uint32_t polynomial_size);
 template <typename Torus>
-__host__ __device__ uint64_t
-get_buffer_size_full_sm_tbc_multibit_programmable_bootstrap(
+uint64_t get_buffer_size_full_sm_tbc_multibit_programmable_bootstrap(
     uint32_t polynomial_size);
 
 template <typename Torus> struct pbs_buffer<Torus, PBS_TYPE::MULTI_BIT> {
@@ -156,7 +127,7 @@ template <typename Torus> struct pbs_buffer<Torus, PBS_TYPE::MULTI_BIT> {
   int8_t *d_mem_acc_step_two = NULL;
   int8_t *d_mem_acc_cg = NULL;
   int8_t *d_mem_acc_tbc = NULL;
-
+  uint32_t lwe_chunk_size;
   double2 *keybundle_fft;
   Torus *global_accumulator;
   double2 *global_accumulator_fft;
@@ -168,6 +139,7 @@ template <typename Torus> struct pbs_buffer<Torus, PBS_TYPE::MULTI_BIT> {
              uint32_t input_lwe_ciphertext_count, uint32_t lwe_chunk_size,
              PBS_VARIANT pbs_variant, bool allocate_gpu_memory) {
     this->pbs_variant = pbs_variant;
+    this->lwe_chunk_size = lwe_chunk_size;
     auto max_shared_memory = cuda_get_max_shared_memory(gpu_index);
 
     // default
@@ -317,8 +289,7 @@ template <typename Torus> struct pbs_buffer<Torus, PBS_TYPE::MULTI_BIT> {
 };
 
 template <typename Torus, class params>
-__host__ uint32_t get_lwe_chunk_size(uint32_t gpu_index, uint32_t max_num_pbs,
-                                     uint32_t polynomial_size,
-                                     uint32_t max_shared_memory);
+uint32_t get_lwe_chunk_size(uint32_t gpu_index, uint32_t max_num_pbs,
+                            uint32_t polynomial_size);
 
 #endif // CUDA_MULTI_BIT_H

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

@@ -1,17 +1,3 @@
-set(SOURCES
-    ${CMAKE_SOURCE_DIR}/${INCLUDE_DIR}/bit_extraction.h
-    ${CMAKE_SOURCE_DIR}/${INCLUDE_DIR}/bitwise_ops.h
-    ${CMAKE_SOURCE_DIR}/${INCLUDE_DIR}/bootstrap.h
-    ${CMAKE_SOURCE_DIR}/${INCLUDE_DIR}/bootstrap_multibit.h
-    ${CMAKE_SOURCE_DIR}/${INCLUDE_DIR}/ciphertext.h
-    ${CMAKE_SOURCE_DIR}/${INCLUDE_DIR}/circuit_bootstrap.h
-    ${CMAKE_SOURCE_DIR}/${INCLUDE_DIR}/device.h
-    ${CMAKE_SOURCE_DIR}/${INCLUDE_DIR}/integer.h
-    ${CMAKE_SOURCE_DIR}/${INCLUDE_DIR}/keyswitch.h
-    ${CMAKE_SOURCE_DIR}/${INCLUDE_DIR}/linear_algebra.h
-    ${CMAKE_SOURCE_DIR}/${INCLUDE_DIR}/shifts.h
-    ${CMAKE_SOURCE_DIR}/${INCLUDE_DIR}/vertical_packing.h
-    ${CMAKE_SOURCE_DIR}/${INCLUDE_DIR}/helper_multi_gpu.h)
 file(GLOB_RECURSE SOURCES "*.cu")
 add_library(tfhe_cuda_backend STATIC ${SOURCES})
 set_target_properties(tfhe_cuda_backend PROPERTIES CUDA_SEPARABLE_COMPILATION ON CUDA_RESOLVE_DEVICE_SYMBOLS ON)

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

@@ -1,4 +1,5 @@
 #include "ciphertext.cuh"
+#include "polynomial/parameters.cuh"
 
 void cuda_convert_lwe_ciphertext_vector_to_gpu_64(void *stream,
                                                   uint32_t gpu_index,
@@ -19,3 +20,58 @@ void cuda_convert_lwe_ciphertext_vector_to_cpu_64(void *stream,
       static_cast<cudaStream_t>(stream), gpu_index, (uint64_t *)dest,
       (uint64_t *)src, number_of_cts, lwe_dimension);
 }
+
+void cuda_glwe_sample_extract_64(void *stream, uint32_t gpu_index,
+                                 void *lwe_array_out, void *glwe_array_in,
+                                 uint32_t *nth_array, uint32_t num_nths,
+                                 uint32_t glwe_dimension,
+                                 uint32_t polynomial_size) {
+
+  switch (polynomial_size) {
+  case 256:
+    host_sample_extract<uint64_t, AmortizedDegree<256>>(
+        static_cast<cudaStream_t>(stream), gpu_index, (uint64_t *)lwe_array_out,
+        (uint64_t *)glwe_array_in, (uint32_t *)nth_array, num_nths,
+        glwe_dimension);
+    break;
+  case 512:
+    host_sample_extract<uint64_t, AmortizedDegree<512>>(
+        static_cast<cudaStream_t>(stream), gpu_index, (uint64_t *)lwe_array_out,
+        (uint64_t *)glwe_array_in, (uint32_t *)nth_array, num_nths,
+        glwe_dimension);
+    break;
+  case 1024:
+    host_sample_extract<uint64_t, AmortizedDegree<1024>>(
+        static_cast<cudaStream_t>(stream), gpu_index, (uint64_t *)lwe_array_out,
+        (uint64_t *)glwe_array_in, (uint32_t *)nth_array, num_nths,
+        glwe_dimension);
+    break;
+  case 2048:
+    host_sample_extract<uint64_t, AmortizedDegree<2048>>(
+        static_cast<cudaStream_t>(stream), gpu_index, (uint64_t *)lwe_array_out,
+        (uint64_t *)glwe_array_in, (uint32_t *)nth_array, num_nths,
+        glwe_dimension);
+    break;
+  case 4096:
+    host_sample_extract<uint64_t, AmortizedDegree<4096>>(
+        static_cast<cudaStream_t>(stream), gpu_index, (uint64_t *)lwe_array_out,
+        (uint64_t *)glwe_array_in, (uint32_t *)nth_array, num_nths,
+        glwe_dimension);
+    break;
+  case 8192:
+    host_sample_extract<uint64_t, AmortizedDegree<8192>>(
+        static_cast<cudaStream_t>(stream), gpu_index, (uint64_t *)lwe_array_out,
+        (uint64_t *)glwe_array_in, (uint32_t *)nth_array, num_nths,
+        glwe_dimension);
+    break;
+  case 16384:
+    host_sample_extract<uint64_t, AmortizedDegree<16384>>(
+        static_cast<cudaStream_t>(stream), gpu_index, (uint64_t *)lwe_array_out,
+        (uint64_t *)glwe_array_in, (uint32_t *)nth_array, num_nths,
+        glwe_dimension);
+    break;
+  default:
+    PANIC("Cuda error: unsupported polynomial size. Supported "
+          "N's are powers of two in the interval [256..16384].")
+  }
+}

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

@@ -3,6 +3,7 @@
 
 #include "ciphertext.h"
 #include "device.h"
+#include "polynomial/functions.cuh"
 #include <cstdint>
 
 template <typename T>
@@ -25,4 +26,40 @@ void cuda_convert_lwe_ciphertext_vector_to_cpu(cudaStream_t stream,
   cuda_memcpy_async_to_cpu(dest, src, size, stream, gpu_index);
 }
 
+template <typename Torus, class params>
+__global__ void sample_extract(Torus *lwe_array_out, Torus *glwe_array_in,
+                               uint32_t *nth_array, uint32_t glwe_dimension) {
+
+  const int input_id = blockIdx.x;
+
+  const int glwe_input_size = (glwe_dimension + 1) * params::degree;
+  const int lwe_output_size = glwe_dimension * params::degree + 1;
+
+  auto lwe_out = lwe_array_out + input_id * lwe_output_size;
+
+  // We assume each GLWE will store the first polynomial_size inputs
+  uint32_t lwe_per_glwe = params::degree;
+  auto glwe_in = glwe_array_in + (input_id / lwe_per_glwe) * glwe_input_size;
+
+  // nth is ensured to be in [0, lwe_per_glwe)
+  auto nth = nth_array[input_id] % lwe_per_glwe;
+
+  sample_extract_mask<Torus, params>(lwe_out, glwe_in, glwe_dimension, nth);
+  sample_extract_body<Torus, params>(lwe_out, glwe_in, glwe_dimension, nth);
+}
+
+template <typename Torus, class params>
+__host__ void host_sample_extract(cudaStream_t stream, uint32_t gpu_index,
+                                  Torus *lwe_array_out, Torus *glwe_array_in,
+                                  uint32_t *nth_array, uint32_t num_nths,
+                                  uint32_t glwe_dimension) {
+  cudaSetDevice(gpu_index);
+
+  dim3 grid(num_nths);
+  dim3 thds(params::degree / params::opt);
+  sample_extract<Torus, params><<<grid, thds, 0, stream>>>(
+      lwe_array_out, glwe_array_in, nth_array, glwe_dimension);
+  check_cuda_error(cudaGetLastError());
+}
+
 #endif

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

@@ -9,16 +9,14 @@ void cuda_keyswitch_lwe_ciphertext_vector_32(
     void *stream, uint32_t gpu_index, void *lwe_array_out,
     void *lwe_output_indexes, void *lwe_array_in, void *lwe_input_indexes,
     void *ksk, uint32_t lwe_dimension_in, uint32_t lwe_dimension_out,
-    uint32_t base_log, uint32_t level_count, uint32_t num_samples,
-    uint32_t gpu_offset) {
-  cuda_keyswitch_lwe_ciphertext_vector(
+    uint32_t base_log, uint32_t level_count, uint32_t num_samples) {
+  host_keyswitch_lwe_ciphertext_vector<uint32_t>(
       static_cast<cudaStream_t>(stream), gpu_index,
       static_cast<uint32_t *>(lwe_array_out),
       static_cast<uint32_t *>(lwe_output_indexes),
       static_cast<uint32_t *>(lwe_array_in),
       static_cast<uint32_t *>(lwe_input_indexes), static_cast<uint32_t *>(ksk),
-      lwe_dimension_in, lwe_dimension_out, base_log, level_count, num_samples,
-      gpu_offset);
+      lwe_dimension_in, lwe_dimension_out, base_log, level_count, num_samples);
 }
 
 /* Perform keyswitch on a batch of 64 bits input LWE ciphertexts.
@@ -41,14 +39,44 @@ void cuda_keyswitch_lwe_ciphertext_vector_64(
     void *stream, uint32_t gpu_index, void *lwe_array_out,
     void *lwe_output_indexes, void *lwe_array_in, void *lwe_input_indexes,
     void *ksk, uint32_t lwe_dimension_in, uint32_t lwe_dimension_out,
-    uint32_t base_log, uint32_t level_count, uint32_t num_samples,
-    uint32_t gpu_offset) {
-  cuda_keyswitch_lwe_ciphertext_vector(
+    uint32_t base_log, uint32_t level_count, uint32_t num_samples) {
+  host_keyswitch_lwe_ciphertext_vector<uint64_t>(
       static_cast<cudaStream_t>(stream), gpu_index,
       static_cast<uint64_t *>(lwe_array_out),
       static_cast<uint64_t *>(lwe_output_indexes),
       static_cast<uint64_t *>(lwe_array_in),
       static_cast<uint64_t *>(lwe_input_indexes), static_cast<uint64_t *>(ksk),
-      lwe_dimension_in, lwe_dimension_out, base_log, level_count, num_samples,
-      gpu_offset);
+      lwe_dimension_in, lwe_dimension_out, base_log, level_count, num_samples);
+}
+
+void scratch_packing_keyswitch_lwe_list_to_glwe_64(
+    void *stream, uint32_t gpu_index, int8_t **fp_ks_buffer,
+    uint32_t glwe_dimension, uint32_t polynomial_size, uint32_t num_lwes,
+    bool allocate_gpu_memory) {
+  scratch_packing_keyswitch_lwe_list_to_glwe<uint64_t>(
+      static_cast<cudaStream_t>(stream), gpu_index, fp_ks_buffer,
+      glwe_dimension, polynomial_size, num_lwes, allocate_gpu_memory);
+}
+/* Perform functional packing keyswitch on a batch of 64 bits input LWE
+ * ciphertexts.
+ */
+void cuda_packing_keyswitch_lwe_list_to_glwe_64(
+    void *stream, uint32_t gpu_index, void *glwe_array_out, void *lwe_array_in,
+    void *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) {
+
+  host_packing_keyswitch_lwe_list_to_glwe<uint64_t>(
+      static_cast<cudaStream_t>(stream), gpu_index,
+      static_cast<uint64_t *>(glwe_array_out),
+      static_cast<uint64_t *>(lwe_array_in),
+      static_cast<uint64_t *>(fp_ksk_array), fp_ks_buffer, input_lwe_dimension,
+      output_glwe_dimension, output_polynomial_size, base_log, level_count,
+      num_lwes);
+}
+
+void cleanup_packing_keyswitch_lwe_list_to_glwe(void *stream,
+                                                uint32_t gpu_index,
+                                                int8_t **fp_ks_buffer) {
+  cuda_drop_async(*fp_ks_buffer, static_cast<cudaStream_t>(stream), gpu_index);
 }

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

@@ -7,6 +7,7 @@
 #include "polynomial/functions.cuh"
 #include "polynomial/polynomial_math.cuh"
 #include "torus.cuh"
+#include "utils/helper.cuh"
 #include "utils/kernel_dimensions.cuh"
 #include <thread>
 #include <vector>
@@ -38,26 +39,25 @@ __device__ Torus *get_ith_block(Torus *ksk, int i, int level,
 // threads in y are used to paralelize the lwe_dimension_in loop.
 // shared memory is used to store intermediate results of the reduction.
 template <typename Torus>
-__global__ void keyswitch(Torus *lwe_array_out, Torus *lwe_output_indexes,
-                          Torus *lwe_array_in, Torus *lwe_input_indexes,
-                          Torus *ksk, uint32_t lwe_dimension_in,
-                          uint32_t lwe_dimension_out, uint32_t base_log,
-                          uint32_t level_count, int gpu_offset) {
+__global__ void
+keyswitch(Torus *lwe_array_out, const Torus *__restrict__ lwe_output_indexes,
+          const Torus *__restrict__ lwe_array_in,
+          const Torus *__restrict__ lwe_input_indexes,
+          const Torus *__restrict__ ksk, uint32_t lwe_dimension_in,
+          uint32_t lwe_dimension_out, uint32_t base_log, uint32_t level_count) {
   const int tid = threadIdx.x + blockIdx.x * blockDim.x;
   const int shmem_index = threadIdx.x + threadIdx.y * blockDim.x;
 
   extern __shared__ int8_t sharedmem[];
   Torus *lwe_acc_out = (Torus *)sharedmem;
-  auto block_lwe_array_out =
-      get_chunk(lwe_array_out, lwe_output_indexes[blockIdx.y + gpu_offset],
-                lwe_dimension_out + 1);
+  auto block_lwe_array_out = get_chunk(
+      lwe_array_out, lwe_output_indexes[blockIdx.y], lwe_dimension_out + 1);
 
   if (tid <= lwe_dimension_out) {
 
     Torus local_lwe_out = 0;
-    auto block_lwe_array_in =
-        get_chunk(lwe_array_in, lwe_input_indexes[blockIdx.y + gpu_offset],
-                  lwe_dimension_in + 1);
+    auto block_lwe_array_in = get_chunk(
+        lwe_array_in, lwe_input_indexes[blockIdx.y], lwe_dimension_in + 1);
 
     if (tid == lwe_dimension_out && threadIdx.y == 0) {
       local_lwe_out = block_lwe_array_in[lwe_dimension_in];
@@ -99,12 +99,11 @@ __global__ void keyswitch(Torus *lwe_array_out, Torus *lwe_output_indexes,
 }
 
 template <typename Torus>
-__host__ void cuda_keyswitch_lwe_ciphertext_vector(
+__host__ void host_keyswitch_lwe_ciphertext_vector(
     cudaStream_t stream, uint32_t gpu_index, Torus *lwe_array_out,
     Torus *lwe_output_indexes, Torus *lwe_array_in, Torus *lwe_input_indexes,
     Torus *ksk, uint32_t lwe_dimension_in, uint32_t lwe_dimension_out,
-    uint32_t base_log, uint32_t level_count, uint32_t num_samples,
-    uint32_t gpu_offset = 0) {
+    uint32_t base_log, uint32_t level_count, uint32_t num_samples) {
 
   cudaSetDevice(gpu_index);
 
@@ -120,42 +119,196 @@ __host__ void cuda_keyswitch_lwe_ciphertext_vector(
 
   keyswitch<Torus><<<grid, threads, shared_mem, stream>>>(
       lwe_array_out, lwe_output_indexes, lwe_array_in, lwe_input_indexes, ksk,
-      lwe_dimension_in, lwe_dimension_out, base_log, level_count, gpu_offset);
+      lwe_dimension_in, lwe_dimension_out, base_log, level_count);
   check_cuda_error(cudaGetLastError());
 }
 
 template <typename Torus>
-void execute_keyswitch(cudaStream_t *streams, uint32_t *gpu_indexes,
-                       uint32_t gpu_count, Torus *lwe_array_out,
-                       Torus *lwe_output_indexes, Torus *lwe_array_in,
-                       Torus *lwe_input_indexes, Torus **ksks,
-                       uint32_t lwe_dimension_in, uint32_t lwe_dimension_out,
-                       uint32_t base_log, uint32_t level_count,
-                       uint32_t num_samples, bool sync_streams = true) {
+void execute_keyswitch_async(cudaStream_t *streams, uint32_t *gpu_indexes,
+                             uint32_t gpu_count,
+                             const LweArrayVariant<Torus> &lwe_array_out,
+                             const LweArrayVariant<Torus> &lwe_output_indexes,
+                             const LweArrayVariant<Torus> &lwe_array_in,
+                             const LweArrayVariant<Torus> &lwe_input_indexes,
+                             Torus **ksks, uint32_t lwe_dimension_in,
+                             uint32_t lwe_dimension_out, uint32_t base_log,
+                             uint32_t level_count, uint32_t num_samples) {
 
   /// If the number of radix blocks is lower than the number of GPUs, not all
   /// GPUs will be active and there will be 1 input per GPU
-  auto active_gpu_count = get_active_gpu_count(num_samples, gpu_count);
-  int num_samples_on_gpu_0 = get_num_inputs_on_gpu(num_samples, 0, gpu_count);
-  if (sync_streams)
-    cuda_synchronize_stream(streams[0], gpu_indexes[0]);
-#pragma omp parallel for num_threads(active_gpu_count)
-  for (uint i = 0; i < active_gpu_count; i++) {
+  for (uint i = 0; i < gpu_count; i++) {
     int num_samples_on_gpu = get_num_inputs_on_gpu(num_samples, i, gpu_count);
-    int gpu_offset = get_gpu_offset(num_samples, i, gpu_count);
+
+    Torus *current_lwe_array_out = GET_VARIANT_ELEMENT(lwe_array_out, i);
+    Torus *current_lwe_output_indexes =
+        GET_VARIANT_ELEMENT(lwe_output_indexes, i);
+    Torus *current_lwe_array_in = GET_VARIANT_ELEMENT(lwe_array_in, i);
+    Torus *current_lwe_input_indexes =
+        GET_VARIANT_ELEMENT(lwe_input_indexes, i);
 
     // Compute Keyswitch
-    cuda_keyswitch_lwe_ciphertext_vector<Torus>(
-        streams[i], gpu_indexes[i], lwe_array_out, lwe_output_indexes,
-        lwe_array_in, lwe_input_indexes, ksks[i], lwe_dimension_in,
-        lwe_dimension_out, base_log, level_count, num_samples_on_gpu,
-        gpu_offset);
+    host_keyswitch_lwe_ciphertext_vector<Torus>(
+        streams[i], gpu_indexes[i], current_lwe_array_out,
+        current_lwe_output_indexes, current_lwe_array_in,
+        current_lwe_input_indexes, ksks[i], lwe_dimension_in, lwe_dimension_out,
+        base_log, level_count, num_samples_on_gpu);
   }
+}
 
-  if (sync_streams)
-    for (uint i = 0; i < active_gpu_count; i++) {
-      cuda_synchronize_stream(streams[i], gpu_indexes[i]);
+template <typename Torus>
+__host__ void scratch_packing_keyswitch_lwe_list_to_glwe(
+    cudaStream_t stream, uint32_t gpu_index, int8_t **fp_ks_buffer,
+    uint32_t glwe_dimension, uint32_t polynomial_size, uint32_t num_lwes,
+    bool allocate_gpu_memory) {
+  cudaSetDevice(gpu_index);
+
+  int glwe_accumulator_size = (glwe_dimension + 1) * polynomial_size;
+
+  if (allocate_gpu_memory)
+    *fp_ks_buffer = (int8_t *)cuda_malloc_async(
+        2 * num_lwes * glwe_accumulator_size * sizeof(Torus), stream,
+        gpu_index);
+}
+
+// public functional packing keyswitch for a single LWE ciphertext
+//
+// Assumes there are (glwe_dimension+1) * polynomial_size threads split through
+// different thread blocks at the x-axis to work on that input.
+template <typename Torus>
+__device__ void packing_keyswitch_lwe_ciphertext_into_glwe_ciphertext(
+    Torus *glwe_out, Torus *lwe_in, Torus *fp_ksk, uint32_t lwe_dimension_in,
+    uint32_t glwe_dimension, uint32_t polynomial_size, uint32_t base_log,
+    uint32_t level_count) {
+
+  const int tid = threadIdx.x + blockIdx.x * blockDim.x;
+  size_t glwe_size = (glwe_dimension + 1);
+
+  if (tid < glwe_size * polynomial_size) {
+    const int local_index = threadIdx.x;
+    // the output_glwe is split in polynomials and each x-block takes one of
+    // them
+    size_t poly_id = blockIdx.x;
+    size_t coef_per_block = blockDim.x;
+
+    // number of coefficients inside fp-ksk block for each lwe_input coefficient
+    size_t ksk_block_size = glwe_size * polynomial_size * level_count;
+
+    // initialize accumulator to 0
+    glwe_out[tid] = SEL(0, lwe_in[lwe_dimension_in],
+                        tid == glwe_dimension * polynomial_size);
+
+    // Iterate through all lwe elements
+    for (int i = 0; i < lwe_dimension_in; i++) {
+      // Round and prepare decomposition
+      Torus a_i = round_to_closest_multiple(lwe_in[i], base_log, level_count);
+
+      Torus state = a_i >> (sizeof(Torus) * 8 - base_log * level_count);
+      Torus mod_b_mask = (1ll << base_log) - 1ll;
+
+      // block of key for current lwe coefficient (cur_input_lwe[i])
+      auto ksk_block = &fp_ksk[i * ksk_block_size];
+      for (int j = 0; j < level_count; j++) {
+        auto ksk_glwe = &ksk_block[j * glwe_size * polynomial_size];
+        // Iterate through each level and multiply by the ksk piece
+        auto ksk_glwe_chunk = &ksk_glwe[poly_id * coef_per_block];
+        Torus decomposed = decompose_one<Torus>(state, mod_b_mask, base_log);
+        glwe_out[tid] -= decomposed * ksk_glwe_chunk[local_index];
+      }
     }
+  }
+}
+
+// public functional packing keyswitch for a batch of LWE ciphertexts
+//
+// Selects the input each thread is working on using the y-block index.
+//
+// Assumes there are (glwe_dimension+1) * polynomial_size threads split through
+// different thread blocks at the x-axis to work on that input.
+template <typename Torus>
+__global__ void
+packing_keyswitch_lwe_list_to_glwe(Torus *glwe_array_out, Torus *lwe_array_in,
+                                   Torus *fp_ksk, uint32_t lwe_dimension_in,
+                                   uint32_t glwe_dimension,
+                                   uint32_t polynomial_size, uint32_t base_log,
+                                   uint32_t level_count, Torus *d_mem) {
+  const int tid = threadIdx.x + blockIdx.x * blockDim.x;
+
+  const int glwe_accumulator_size = (glwe_dimension + 1) * polynomial_size;
+  const int lwe_size = (lwe_dimension_in + 1);
+
+  const int input_id = blockIdx.y;
+  const int degree = input_id;
+
+  // Select an input
+  auto lwe_in = lwe_array_in + input_id * lwe_size;
+  auto ks_glwe_out = d_mem + input_id * glwe_accumulator_size;
+  auto glwe_out = glwe_array_out + input_id * glwe_accumulator_size;
+  // KS LWE to GLWE
+  packing_keyswitch_lwe_ciphertext_into_glwe_ciphertext(
+      ks_glwe_out, lwe_in, fp_ksk, lwe_dimension_in, glwe_dimension,
+      polynomial_size, base_log, level_count);
+
+  // P * x ^degree
+  auto in_poly = ks_glwe_out + (tid / polynomial_size) * polynomial_size;
+  auto out_result = glwe_out + (tid / polynomial_size) * polynomial_size;
+  polynomial_accumulate_monic_monomial_mul(out_result, in_poly, degree,
+                                           tid % polynomial_size,
+                                           polynomial_size, 1, true);
+}
+
+/// To-do: Rewrite this kernel for efficiency
+template <typename Torus>
+__global__ void accumulate_glwes(Torus *glwe_out, Torus *glwe_array_in,
+                                 uint32_t glwe_dimension,
+                                 uint32_t polynomial_size, uint32_t num_lwes) {
+  const int tid = threadIdx.x + blockIdx.x * blockDim.x;
+  if (tid < (glwe_dimension + 1) * polynomial_size) {
+    glwe_out[tid] = glwe_array_in[tid];
+
+    // Accumulate
+    for (int i = 1; i < num_lwes; i++) {
+      auto glwe_in = glwe_array_in + i * (glwe_dimension + 1) * polynomial_size;
+      glwe_out[tid] += glwe_in[tid];
+    }
+  }
+}
+
+template <typename Torus>
+__host__ void host_packing_keyswitch_lwe_list_to_glwe(
+    cudaStream_t stream, uint32_t gpu_index, Torus *glwe_out,
+    Torus *lwe_array_in, Torus *fp_ksk_array, int8_t *fp_ks_buffer,
+    uint32_t lwe_dimension_in, uint32_t glwe_dimension,
+    uint32_t polynomial_size, uint32_t base_log, uint32_t level_count,
+    uint32_t num_lwes) {
+
+  if (num_lwes > polynomial_size)
+    PANIC("Cuda error: too many LWEs to pack. The number of LWEs should be "
+          "smaller than "
+          "polynomial_size.")
+
+  cudaSetDevice(gpu_index);
+  int glwe_accumulator_size = (glwe_dimension + 1) * polynomial_size;
+
+  int num_blocks = 0, num_threads = 0;
+  getNumBlocksAndThreads(glwe_accumulator_size, 128, num_blocks, num_threads);
+
+  dim3 grid(num_blocks, num_lwes);
+  dim3 threads(num_threads);
+
+  auto d_mem = (Torus *)fp_ks_buffer;
+  auto d_tmp_glwe_array_out = d_mem + num_lwes * glwe_accumulator_size;
+
+  // individually keyswitch each lwe
+  packing_keyswitch_lwe_list_to_glwe<<<grid, threads, 0, stream>>>(
+      d_tmp_glwe_array_out, lwe_array_in, fp_ksk_array, lwe_dimension_in,
+      glwe_dimension, polynomial_size, base_log, level_count, d_mem);
+  check_cuda_error(cudaGetLastError());
+
+  // accumulate to a single glwe
+  accumulate_glwes<<<num_blocks, threads, 0, stream>>>(
+      glwe_out, d_tmp_glwe_array_out, glwe_dimension, polynomial_size,
+      num_lwes);
+  check_cuda_error(cudaGetLastError());
 }
 
 #endif

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

@@ -1,9 +1,16 @@
 #ifndef CNCRT_TORUS_CUH
 #define CNCRT_TORUS_CUH
 
+#include "polynomial/parameters.cuh"
 #include "types/int128.cuh"
+#include "utils/kernel_dimensions.cuh"
 #include <limits>
 
+template <typename T>
+__host__ __device__ __forceinline__ constexpr double get_two_pow_torus_bits() {
+  return (sizeof(T) == 4) ? 4294967296.0 : 18446744073709551616.0;
+}
+
 template <typename T>
 __device__ inline void typecast_double_to_torus(double x, T &r) {
   r = T(x);
@@ -26,49 +33,63 @@ __device__ inline void typecast_double_to_torus<uint64_t>(double x,
   r = lll;
 }
 
+template <typename T>
+__device__ inline void typecast_double_round_to_torus(double x, T &r) {
+  constexpr double mx = get_two_pow_torus_bits<T>();
+  // floor must be used here because round has an issue with rounding .5,
+  // as it rounds away from zero.
+  double frac = x - floor(x);
+  frac *= mx;
+  typecast_double_to_torus(round(frac), r);
+}
+
 template <typename T>
 __device__ inline T round_to_closest_multiple(T x, uint32_t base_log,
                                               uint32_t level_count) {
-  T shift = sizeof(T) * 8 - level_count * base_log;
-  T mask = 1ll << (shift - 1);
-  T b = (x & mask) >> (shift - 1);
+  const T non_rep_bit_count = sizeof(T) * 8 - level_count * base_log;
+  const T shift = non_rep_bit_count - 1;
   T res = x >> shift;
-  res += b;
-  res <<= shift;
-  return res;
+  res += 1;
+  res &= (T)(-2);
+  return res << shift;
 }
 
 template <typename T>
-__device__ __forceinline__ void rescale_torus_element(T element, T &output,
-                                                      uint32_t log_shift) {
-  output =
-      round((double)element / (double(std::numeric_limits<T>::max()) + 1.0) *
-            (double)log_shift);
+__device__ __forceinline__ void modulus_switch(T input, T &output,
+                                               uint32_t log_modulus) {
+  constexpr uint32_t BITS = sizeof(T) * 8;
+  output = input + (((T)1) << (BITS - log_modulus - 1));
+  output >>= (BITS - log_modulus);
 }
 
 template <typename T>
-__device__ __forceinline__ T rescale_torus_element(T element,
-                                                   uint32_t log_shift) {
-  return round((double)element / (double(std::numeric_limits<T>::max()) + 1.0) *
-               (double)log_shift);
+__device__ __forceinline__ T modulus_switch(T input, uint32_t log_modulus) {
+  T output;
+  modulus_switch(input, output, log_modulus);
+  return output;
 }
 
-template <>
-__device__ __forceinline__ void
-rescale_torus_element<uint32_t>(uint32_t element, uint32_t &output,
-                                uint32_t log_shift) {
-  output =
-      round(__uint2double_rn(element) /
-            (__uint2double_rn(std::numeric_limits<uint32_t>::max()) + 1.0) *
-            __uint2double_rn(log_shift));
+template <typename Torus>
+__global__ void modulus_switch_inplace(Torus *array, int size,
+                                       uint32_t log_modulus) {
+  const int tid = threadIdx.x + blockIdx.x * blockDim.x;
+  if (tid < size) {
+    array[tid] = modulus_switch(array[tid], log_modulus);
+  }
 }
 
-template <>
-__device__ __forceinline__ void
-rescale_torus_element<uint64_t>(uint64_t element, uint64_t &output,
-                                uint32_t log_shift) {
-  output = round(__ull2double_rn(element) /
-                 (__ull2double_rn(std::numeric_limits<uint64_t>::max()) + 1.0) *
-                 __uint2double_rn(log_shift));
+template <typename Torus>
+__host__ void host_modulus_switch_inplace(cudaStream_t stream,
+                                          uint32_t gpu_index, Torus *array,
+                                          int size, uint32_t log_modulus) {
+  cudaSetDevice(gpu_index);
+
+  int num_threads = 0, num_blocks = 0;
+  getNumBlocksAndThreads(size, 1024, num_blocks, num_threads);
+
+  modulus_switch_inplace<<<num_blocks, num_threads, 0, stream>>>(array, size,
+                                                                 log_modulus);
+  check_cuda_error(cudaGetLastError());
 }
+
 #endif // CNCRT_TORUS_H

backends/tfhe-cuda-backend/cuda/src/device.cu

@@ -137,6 +137,30 @@ void cuda_memcpy_async_gpu_to_gpu(void *dest, void *src, uint64_t size,
   }
 }
 
+/// Copy memory within a GPU
+void cuda_memcpy_gpu_to_gpu(void *dest, void *src, uint64_t size,
+                            uint32_t gpu_index) {
+  if (size == 0)
+    return;
+  cudaPointerAttributes attr_dest;
+  check_cuda_error(cudaPointerGetAttributes(&attr_dest, dest));
+  if (attr_dest.type != cudaMemoryTypeDevice) {
+    PANIC("Cuda error: invalid dest device pointer in copy from GPU to GPU.")
+  }
+  cudaPointerAttributes attr_src;
+  check_cuda_error(cudaPointerGetAttributes(&attr_src, src));
+  if (attr_src.type != cudaMemoryTypeDevice) {
+    PANIC("Cuda error: invalid src device pointer in copy from GPU to GPU.")
+  }
+  check_cuda_error(cudaSetDevice(gpu_index));
+  if (attr_src.device == attr_dest.device) {
+    check_cuda_error(cudaMemcpy(dest, src, size, cudaMemcpyDeviceToDevice));
+  } else {
+    check_cuda_error(
+        cudaMemcpyPeer(dest, attr_dest.device, src, attr_src.device, size));
+  }
+}
+
 /// Synchronizes device
 void cuda_synchronize_device(uint32_t gpu_index) {
   check_cuda_error(cudaSetDevice(gpu_index));
@@ -166,19 +190,21 @@ __global__ void cuda_set_value_kernel(Torus *array, Torus value, Torus n) {
 template <typename Torus>
 void cuda_set_value_async(cudaStream_t stream, uint32_t gpu_index,
                           Torus *d_array, Torus value, Torus n) {
-  cudaPointerAttributes attr;
-  check_cuda_error(cudaPointerGetAttributes(&attr, d_array));
-  if (attr.type != cudaMemoryTypeDevice) {
-    PANIC("Cuda error: invalid dest device pointer in cuda set value.")
-  }
-  check_cuda_error(cudaSetDevice(gpu_index));
-  int block_size = 256;
-  int num_blocks = (n + block_size - 1) / block_size;
+  if (n > 0) {
+    cudaPointerAttributes attr;
+    check_cuda_error(cudaPointerGetAttributes(&attr, d_array));
+    if (attr.type != cudaMemoryTypeDevice) {
+      PANIC("Cuda error: invalid dest device pointer in cuda set value.")
+    }
+    check_cuda_error(cudaSetDevice(gpu_index));
+    int block_size = 256;
+    int num_blocks = (n + block_size - 1) / block_size;
 
-  // Launch the kernel
-  cuda_set_value_kernel<<<num_blocks, block_size, 0, stream>>>(d_array, value,
-                                                               n);
-  check_cuda_error(cudaGetLastError());
+    // Launch the kernel
+    cuda_set_value_kernel<Torus>
+        <<<num_blocks, block_size, 0, stream>>>(d_array, value, n);
+    check_cuda_error(cudaGetLastError());
+  }
 }
 
 /// Explicitly instantiate cuda_set_value_async for 32 and 64 bits
@@ -241,22 +267,18 @@ void cuda_drop_async(void *ptr, cudaStream_t stream, uint32_t gpu_index) {
 
 /// Get the maximum size for the shared memory
 int cuda_get_max_shared_memory(uint32_t gpu_index) {
-  check_cuda_error(cudaSetDevice(gpu_index));
   int max_shared_memory = 0;
   cudaDeviceGetAttribute(&max_shared_memory, cudaDevAttrMaxSharedMemoryPerBlock,
                          gpu_index);
   check_cuda_error(cudaGetLastError());
+#if CUDA_ARCH == 900
+  max_shared_memory = 226000;
+#elif CUDA_ARCH == 890
+  max_shared_memory = 127000;
+#elif CUDA_ARCH == 800
+  max_shared_memory = 163000;
+#elif CUDA_ARCH == 700
+  max_shared_memory = 95000;
+#endif
   return max_shared_memory;
 }
-
-void cuda_stream_add_callback(cudaStream_t stream, uint32_t gpu_index,
-                              cudaStreamCallback_t callback, void *user_data) {
-
-  check_cuda_error(cudaSetDevice(gpu_index));
-  check_cuda_error(cudaStreamAddCallback(stream, callback, user_data, 0));
-}
-
-void host_free_on_stream_callback(cudaStream_t stream, cudaError_t status,
-                                  void *host_pointer) {
-  free(host_pointer);
-}

backends/tfhe-cuda-backend/cuda/src/fft/bnsmfft.cuh

@@ -6,6 +6,7 @@
 #include "twiddles.cuh"
 #include "types/complex/operations.cuh"
 
+using Index = unsigned;
 /*
  * Direct negacyclic FFT:
  *   - before the FFT the N real coefficients are stored into a
@@ -31,293 +32,81 @@ template <class params> __device__ void NSMFFT_direct(double2 *A) {
    *  full loop, which should increase performance
    */
 
-  size_t tid = threadIdx.x;
-  size_t twid_id;
-  size_t i1, i2;
-  double2 u, v, w;
+  __syncthreads();
+  constexpr Index BUTTERFLY_DEPTH = params::opt >> 1;
+  constexpr Index LOG2_DEGREE = params::log2_degree;
+  constexpr Index HALF_DEGREE = params::degree >> 1;
+  constexpr Index STRIDE = params::degree / params::opt;
+
+  Index tid = threadIdx.x;
+  double2 u[BUTTERFLY_DEPTH], v[BUTTERFLY_DEPTH], w;
+
+  // load into registers
+#pragma unroll
+  for (Index i = 0; i < BUTTERFLY_DEPTH; ++i) {
+    u[i] = A[tid];
+    v[i] = A[tid + HALF_DEGREE];
+
+    tid += STRIDE;
+  }
+
   // level 1
   // we don't make actual complex multiplication on level1 since we have only
   // one twiddle, it's real and image parts are equal, so we can multiply
   // it with simpler operations
 #pragma unroll
-  for (size_t i = 0; i < params::opt / 2; ++i) {
-    i1 = tid;
-    i2 = tid + params::degree / 2;
-
-    u = A[i1];
-    v = A[i2] * (double2){0.707106781186547461715008466854,
-                          0.707106781186547461715008466854};
-
-    A[i1] += v;
-    A[i2] = u - v;
-
-    tid += params::degree / params::opt;
+  for (Index i = 0; i < BUTTERFLY_DEPTH; ++i) {
+    w = v[i] * (double2){0.707106781186547461715008466854,
+                         0.707106781186547461715008466854};
+    v[i] = u[i] - w;
+    u[i] = u[i] + w;
   }
-  __syncthreads();
 
-  // level 2
-  // from this level there are more than one twiddles and none of them has equal
-  // real and imag parts, so complete complex multiplication is needed
-  // for each level params::degree / 2^level represents number of coefficients
-  // inside divided chunk of specific level
-  //
-  tid = threadIdx.x;
-#pragma unroll
-  for (size_t i = 0; i < params::opt / 2; ++i) {
-    twid_id = tid / (params::degree / 4);
-    i1 = 2 * (params::degree / 4) * twid_id + (tid & (params::degree / 4 - 1));
-    i2 = i1 + params::degree / 4;
+  Index twiddle_shift = 1;
+  for (Index l = LOG2_DEGREE - 1; l >= 1; --l) {
+    Index lane_mask = 1 << (l - 1);
+    Index thread_mask = (1 << l) - 1;
+    twiddle_shift <<= 1;
 
-    w = negtwiddles[twid_id + 2];
-    u = A[i1];
-    v = A[i2] * w;
-
-    A[i1] += v;
-    A[i2] = u - v;
-
-    tid += params::degree / params::opt;
-  }
-  __syncthreads();
-
-  // level 3
-  tid = threadIdx.x;
-#pragma unroll
-  for (size_t i = 0; i < params::opt / 2; ++i) {
-    twid_id = tid / (params::degree / 8);
-    i1 = 2 * (params::degree / 8) * twid_id + (tid & (params::degree / 8 - 1));
-    i2 = i1 + params::degree / 8;
-
-    w = negtwiddles[twid_id + 4];
-    u = A[i1];
-    v = A[i2] * w;
-
-    A[i1] += v;
-    A[i2] = u - v;
-
-    tid += params::degree / params::opt;
-  }
-  __syncthreads();
-
-  // level 4
-  tid = threadIdx.x;
-#pragma unroll
-  for (size_t i = 0; i < params::opt / 2; ++i) {
-    twid_id = tid / (params::degree / 16);
-    i1 =
-        2 * (params::degree / 16) * twid_id + (tid & (params::degree / 16 - 1));
-    i2 = i1 + params::degree / 16;
-
-    w = negtwiddles[twid_id + 8];
-    u = A[i1];
-    v = A[i2] * w;
-
-    A[i1] += v;
-    A[i2] = u - v;
-
-    tid += params::degree / params::opt;
-  }
-  __syncthreads();
-
-  // level 5
-  tid = threadIdx.x;
-#pragma unroll
-  for (size_t i = 0; i < params::opt / 2; ++i) {
-    twid_id = tid / (params::degree / 32);
-    i1 =
-        2 * (params::degree / 32) * twid_id + (tid & (params::degree / 32 - 1));
-    i2 = i1 + params::degree / 32;
-
-    w = negtwiddles[twid_id + 16];
-    u = A[i1];
-    v = A[i2] * w;
-
-    A[i1] += v;
-    A[i2] = u - v;
-
-    tid += params::degree / params::opt;
-  }
-  __syncthreads();
-
-  // level 6
-  tid = threadIdx.x;
-#pragma unroll
-  for (size_t i = 0; i < params::opt / 2; ++i) {
-    twid_id = tid / (params::degree / 64);
-    i1 =
-        2 * (params::degree / 64) * twid_id + (tid & (params::degree / 64 - 1));
-    i2 = i1 + params::degree / 64;
-
-    w = negtwiddles[twid_id + 32];
-    u = A[i1];
-    v = A[i2] * w;
-
-    A[i1] += v;
-    A[i2] = u - v;
-
-    tid += params::degree / params::opt;
-  }
-  __syncthreads();
-
-  // level 7
-  tid = threadIdx.x;
-#pragma unroll
-  for (size_t i = 0; i < params::opt / 2; ++i) {
-    twid_id = tid / (params::degree / 128);
-    i1 = 2 * (params::degree / 128) * twid_id +
-         (tid & (params::degree / 128 - 1));
-    i2 = i1 + params::degree / 128;
-
-    w = negtwiddles[twid_id + 64];
-    u = A[i1];
-    v = A[i2] * w;
-
-    A[i1] += v;
-    A[i2] = u - v;
-
-    tid += params::degree / params::opt;
-  }
-  __syncthreads();
-
-  // from level 8, we need to check size of params degree, because we support
-  // minimum actual polynomial size = 256,  when compressed size is halfed and
-  // minimum supported compressed size is 128, so we always need first 7
-  // levels of butterfly operation, since butterfly levels are hardcoded
-  // we need to check if polynomial size is big enough to require specific level
-  // of butterfly.
-  if constexpr (params::degree >= 256) {
-    // level 8
     tid = threadIdx.x;
+    __syncthreads();
 #pragma unroll
-    for (size_t i = 0; i < params::opt / 2; ++i) {
-      twid_id = tid / (params::degree / 256);
-      i1 = 2 * (params::degree / 256) * twid_id +
-           (tid & (params::degree / 256 - 1));
-      i2 = i1 + params::degree / 256;
-
-      w = negtwiddles[twid_id + 128];
-      u = A[i1];
-      v = A[i2] * w;
-
-      A[i1] += v;
-      A[i2] = u - v;
-
-      tid += params::degree / params::opt;
+    for (Index i = 0; i < BUTTERFLY_DEPTH; i++) {
+      Index rank = tid & thread_mask;
+      bool u_stays_in_register = rank < lane_mask;
+      A[tid] = (u_stays_in_register) ? v[i] : u[i];
+      tid = tid + STRIDE;
     }
     __syncthreads();
-  }
 
-  if constexpr (params::degree >= 512) {
-    // level 9
     tid = threadIdx.x;
 #pragma unroll
-    for (size_t i = 0; i < params::opt / 2; ++i) {
-      twid_id = tid / (params::degree / 512);
-      i1 = 2 * (params::degree / 512) * twid_id +
-           (tid & (params::degree / 512 - 1));
-      i2 = i1 + params::degree / 512;
+    for (Index i = 0; i < BUTTERFLY_DEPTH; i++) {
+      Index rank = tid & thread_mask;
+      bool u_stays_in_register = rank < lane_mask;
+      w = A[tid ^ lane_mask];
+      u[i] = (u_stays_in_register) ? u[i] : w;
+      v[i] = (u_stays_in_register) ? w : v[i];
+      w = negtwiddles[tid / lane_mask + twiddle_shift];
 
-      w = negtwiddles[twid_id + 256];
-      u = A[i1];
-      v = A[i2] * w;
+      w *= v[i];
 
-      A[i1] += v;
-      A[i2] = u - v;
-
-      tid += params::degree / params::opt;
+      v[i] = u[i] - w;
+      u[i] = u[i] + w;
+      tid = tid + STRIDE;
     }
-    __syncthreads();
   }
+  __syncthreads();
 
-  if constexpr (params::degree >= 1024) {
-    // level 10
-    tid = threadIdx.x;
+  // store registers in SM
+  tid = threadIdx.x;
 #pragma unroll
-    for (size_t i = 0; i < params::opt / 2; ++i) {
-      twid_id = tid / (params::degree / 1024);
-      i1 = 2 * (params::degree / 1024) * twid_id +
-           (tid & (params::degree / 1024 - 1));
-      i2 = i1 + params::degree / 1024;
-
-      w = negtwiddles[twid_id + 512];
-      u = A[i1];
-      v = A[i2] * w;
-
-      A[i1] += v;
-      A[i2] = u - v;
-
-      tid += params::degree / params::opt;
-    }
-    __syncthreads();
-  }
-
-  if constexpr (params::degree >= 2048) {
-    // level 11
-    tid = threadIdx.x;
-#pragma unroll
-    for (size_t i = 0; i < params::opt / 2; ++i) {
-      twid_id = tid / (params::degree / 2048);
-      i1 = 2 * (params::degree / 2048) * twid_id +
-           (tid & (params::degree / 2048 - 1));
-      i2 = i1 + params::degree / 2048;
-
-      w = negtwiddles[twid_id + 1024];
-      u = A[i1];
-      v = A[i2] * w;
-
-      A[i1] += v;
-      A[i2] = u - v;
-
-      tid += params::degree / params::opt;
-    }
-    __syncthreads();
-  }
-
-  if constexpr (params::degree >= 4096) {
-    // level 12
-    tid = threadIdx.x;
-#pragma unroll
-    for (size_t i = 0; i < params::opt / 2; ++i) {
-      twid_id = tid / (params::degree / 4096);
-      i1 = 2 * (params::degree / 4096) * twid_id +
-           (tid & (params::degree / 4096 - 1));
-      i2 = i1 + params::degree / 4096;
-
-      w = negtwiddles[twid_id + 2048];
-      u = A[i1];
-      v = A[i2] * w;
-
-      A[i1] += v;
-      A[i2] = u - v;
-
-      tid += params::degree / params::opt;
-    }
-    __syncthreads();
-  }
-
-  // compressed size = 8192 is actual polynomial size = 16384.
-  // from this size, twiddles can't fit in constant memory,
-  // so from here, butterfly operation access device memory.
-  if constexpr (params::degree >= 8192) {
-    // level 13
-    tid = threadIdx.x;
-#pragma unroll
-    for (size_t i = 0; i < params::opt / 2; ++i) {
-      twid_id = tid / (params::degree / 8192);
-      i1 = 2 * (params::degree / 8192) * twid_id +
-           (tid & (params::degree / 8192 - 1));
-      i2 = i1 + params::degree / 8192;
-
-      w = negtwiddles13[twid_id];
-      u = A[i1];
-      v = A[i2] * w;
-
-      A[i1] += v;
-      A[i2] = u - v;
-
-      tid += params::degree / params::opt;
-    }
-    __syncthreads();
+  for (Index i = 0; i < BUTTERFLY_DEPTH; i++) {
+    A[tid * 2] = u[i];
+    A[tid * 2 + 1] = v[i];
+    tid = tid + STRIDE;
   }
+  __syncthreads();
 }
 
 /*
@@ -332,288 +121,82 @@ template <class params> __device__ void NSMFFT_inverse(double2 *A) {
    *  full loop, which should increase performance
    */
 
+  __syncthreads();
+  constexpr Index BUTTERFLY_DEPTH = params::opt >> 1;
+  constexpr Index LOG2_DEGREE = params::log2_degree;
+  constexpr Index DEGREE = params::degree;
+  constexpr Index HALF_DEGREE = params::degree >> 1;
+  constexpr Index STRIDE = params::degree / params::opt;
+
   size_t tid = threadIdx.x;
-  size_t twid_id;
-  size_t i1, i2;
-  double2 u, w;
+  double2 u[BUTTERFLY_DEPTH], v[BUTTERFLY_DEPTH], w;
 
-  // divide input by compressed polynomial size
-  tid = threadIdx.x;
-  for (size_t i = 0; i < params::opt; ++i) {
-    A[tid] /= params::degree;
-    tid += params::degree / params::opt;
-  }
-  __syncthreads();
-
-  // none of the twiddles have equal real and imag part, so
-  // complete complex multiplication has to be done
-  // here we have more than one twiddle
-  // mapping in backward fft is reversed
-  // butterfly operation is started from last level
-
-  // compressed size = 8192 is actual polynomial size = 16384.
-  // twiddles for this size can't fit in constant memory so
-  // butterfly operation for this level access device memory to fetch
-  // twiddles
-  if constexpr (params::degree >= 8192) {
-    // level 13
-    tid = threadIdx.x;
+  // load into registers and divide by compressed polynomial size
 #pragma unroll
-    for (size_t i = 0; i < params::opt / 2; ++i) {
-      twid_id = tid / (params::degree / 8192);
-      i1 = 2 * (params::degree / 8192) * twid_id +
-           (tid & (params::degree / 8192 - 1));
-      i2 = i1 + params::degree / 8192;
+  for (Index i = 0; i < BUTTERFLY_DEPTH; ++i) {
+    u[i] = A[2 * tid];
+    v[i] = A[2 * tid + 1];
 
-      w = negtwiddles13[twid_id];
-      u = A[i1] - A[i2];
+    u[i] /= DEGREE;
+    v[i] /= DEGREE;
 
-      A[i1] += A[i2];
-      A[i2] = u * conjugate(w);
+    tid += STRIDE;
+  }
 
-      tid += params::degree / params::opt;
+  Index twiddle_shift = DEGREE;
+  for (Index l = 1; l <= LOG2_DEGREE - 1; ++l) {
+    Index lane_mask = 1 << (l - 1);
+    Index thread_mask = (1 << l) - 1;
+    tid = threadIdx.x;
+    twiddle_shift >>= 1;
+
+    // at this point registers are ready for the  butterfly
+    tid = threadIdx.x;
+    __syncthreads();
+#pragma unroll
+    for (Index i = 0; i < BUTTERFLY_DEPTH; ++i) {
+      w = (u[i] - v[i]);
+      u[i] += v[i];
+      v[i] = w * conjugate(negtwiddles[tid / lane_mask + twiddle_shift]);
+
+      // keep one of the register for next iteration and store another one in sm
+      Index rank = tid & thread_mask;
+      bool u_stays_in_register = rank < lane_mask;
+      A[tid] = (u_stays_in_register) ? v[i] : u[i];
+
+      tid = tid + STRIDE;
     }
     __syncthreads();
-  }
 
-  if constexpr (params::degree >= 4096) {
-    // level 12
+    // prepare registers for next butterfly iteration
     tid = threadIdx.x;
 #pragma unroll
-    for (size_t i = 0; i < params::opt / 2; ++i) {
-      twid_id = tid / (params::degree / 4096);
-      i1 = 2 * (params::degree / 4096) * twid_id +
-           (tid & (params::degree / 4096 - 1));
-      i2 = i1 + params::degree / 4096;
+    for (Index i = 0; i < BUTTERFLY_DEPTH; ++i) {
+      Index rank = tid & thread_mask;
+      bool u_stays_in_register = rank < lane_mask;
+      w = A[tid ^ lane_mask];
+      u[i] = (u_stays_in_register) ? u[i] : w;
+      v[i] = (u_stays_in_register) ? w : v[i];
 
-      w = negtwiddles[twid_id + 2048];
-      u = A[i1] - A[i2];
-
-      A[i1] += A[i2];
-      A[i2] = u * conjugate(w);
-
-      tid += params::degree / params::opt;
+      tid = tid + STRIDE;
     }
-    __syncthreads();
   }
 
-  if constexpr (params::degree >= 2048) {
-    // level 11
-    tid = threadIdx.x;
-#pragma unroll
-    for (size_t i = 0; i < params::opt / 2; ++i) {
-      twid_id = tid / (params::degree / 2048);
-      i1 = 2 * (params::degree / 2048) * twid_id +
-           (tid & (params::degree / 2048 - 1));
-      i2 = i1 + params::degree / 2048;
-
-      w = negtwiddles[twid_id + 1024];
-      u = A[i1] - A[i2];
-
-      A[i1] += A[i2];
-      A[i2] = u * conjugate(w);
-
-      tid += params::degree / params::opt;
-    }
-    __syncthreads();
-  }
-
-  if constexpr (params::degree >= 1024) {
-    // level 10
-    tid = threadIdx.x;
-#pragma unroll
-    for (size_t i = 0; i < params::opt / 2; ++i) {
-      twid_id = tid / (params::degree / 1024);
-      i1 = 2 * (params::degree / 1024) * twid_id +
-           (tid & (params::degree / 1024 - 1));
-      i2 = i1 + params::degree / 1024;
-
-      w = negtwiddles[twid_id + 512];
-      u = A[i1] - A[i2];
-
-      A[i1] += A[i2];
-      A[i2] = u * conjugate(w);
-
-      tid += params::degree / params::opt;
-    }
-    __syncthreads();
-  }
-
-  if constexpr (params::degree >= 512) {
-    // level 9
-    tid = threadIdx.x;
-#pragma unroll
-    for (size_t i = 0; i < params::opt / 2; ++i) {
-      twid_id = tid / (params::degree / 512);
-      i1 = 2 * (params::degree / 512) * twid_id +
-           (tid & (params::degree / 512 - 1));
-      i2 = i1 + params::degree / 512;
-
-      w = negtwiddles[twid_id + 256];
-      u = A[i1] - A[i2];
-
-      A[i1] += A[i2];
-      A[i2] = u * conjugate(w);
-
-      tid += params::degree / params::opt;
-    }
-    __syncthreads();
-  }
-
-  if constexpr (params::degree >= 256) {
-    // level 8
-    tid = threadIdx.x;
-#pragma unroll
-    for (size_t i = 0; i < params::opt / 2; ++i) {
-      twid_id = tid / (params::degree / 256);
-      i1 = 2 * (params::degree / 256) * twid_id +
-           (tid & (params::degree / 256 - 1));
-      i2 = i1 + params::degree / 256;
-
-      w = negtwiddles[twid_id + 128];
-      u = A[i1] - A[i2];
-
-      A[i1] += A[i2];
-      A[i2] = u * conjugate(w);
-
-      tid += params::degree / params::opt;
-    }
-    __syncthreads();
-  }
-
-  // below level 8, we don't need to check size of params degree, because we
-  // support minimum actual polynomial size = 256,  when compressed size is
-  // halfed and minimum supported compressed size is 128, so we always need
-  // last 7 levels of butterfly operation, since butterfly levels are hardcoded
-  // we don't need to check if polynomial size is big enough to require
-  // specific level of butterfly.
-  // level 7
-  tid = threadIdx.x;
-#pragma unroll
-  for (size_t i = 0; i < params::opt / 2; ++i) {
-    twid_id = tid / (params::degree / 128);
-    i1 = 2 * (params::degree / 128) * twid_id +
-         (tid & (params::degree / 128 - 1));
-    i2 = i1 + params::degree / 128;
-
-    w = negtwiddles[twid_id + 64];
-    u = A[i1] - A[i2];
-
-    A[i1] += A[i2];
-    A[i2] = u * conjugate(w);
-
-    tid += params::degree / params::opt;
+  // last iteration
+  for (Index i = 0; i < BUTTERFLY_DEPTH; ++i) {
+    w = (u[i] - v[i]);
+    u[i] = u[i] + v[i];
+    v[i] = w * (double2){0.707106781186547461715008466854,
+                         -0.707106781186547461715008466854};
   }
   __syncthreads();
-
-  // level 6
+  // store registers in SM
   tid = threadIdx.x;
 #pragma unroll
-  for (size_t i = 0; i < params::opt / 2; ++i) {
-    twid_id = tid / (params::degree / 64);
-    i1 =
-        2 * (params::degree / 64) * twid_id + (tid & (params::degree / 64 - 1));
-    i2 = i1 + params::degree / 64;
-
-    w = negtwiddles[twid_id + 32];
-    u = A[i1] - A[i2];
-
-    A[i1] += A[i2];
-    A[i2] = u * conjugate(w);
-
-    tid += params::degree / params::opt;
-  }
-  __syncthreads();
-
-  // level 5
-  tid = threadIdx.x;
-#pragma unroll
-  for (size_t i = 0; i < params::opt / 2; ++i) {
-    twid_id = tid / (params::degree / 32);
-    i1 =
-        2 * (params::degree / 32) * twid_id + (tid & (params::degree / 32 - 1));
-    i2 = i1 + params::degree / 32;
-
-    w = negtwiddles[twid_id + 16];
-    u = A[i1] - A[i2];
-
-    A[i1] += A[i2];
-    A[i2] = u * conjugate(w);
-
-    tid += params::degree / params::opt;
-  }
-  __syncthreads();
-
-  // level 4
-  tid = threadIdx.x;
-#pragma unroll
-  for (size_t i = 0; i < params::opt / 2; ++i) {
-    twid_id = tid / (params::degree / 16);
-    i1 =
-        2 * (params::degree / 16) * twid_id + (tid & (params::degree / 16 - 1));
-    i2 = i1 + params::degree / 16;
-
-    w = negtwiddles[twid_id + 8];
-    u = A[i1] - A[i2];
-
-    A[i1] += A[i2];
-    A[i2] = u * conjugate(w);
-
-    tid += params::degree / params::opt;
-  }
-  __syncthreads();
-
-  // level 3
-  tid = threadIdx.x;
-#pragma unroll
-  for (size_t i = 0; i < params::opt / 2; ++i) {
-    twid_id = tid / (params::degree / 8);
-    i1 = 2 * (params::degree / 8) * twid_id + (tid & (params::degree / 8 - 1));
-    i2 = i1 + params::degree / 8;
-
-    w = negtwiddles[twid_id + 4];
-    u = A[i1] - A[i2];
-
-    A[i1] += A[i2];
-    A[i2] = u * conjugate(w);
-
-    tid += params::degree / params::opt;
-  }
-  __syncthreads();
-
-  // level 2
-  tid = threadIdx.x;
-#pragma unroll
-  for (size_t i = 0; i < params::opt / 2; ++i) {
-    twid_id = tid / (params::degree / 4);
-    i1 = 2 * (params::degree / 4) * twid_id + (tid & (params::degree / 4 - 1));
-    i2 = i1 + params::degree / 4;
-
-    w = negtwiddles[twid_id + 2];
-    u = A[i1] - A[i2];
-
-    A[i1] += A[i2];
-    A[i2] = u * conjugate(w);
-
-    tid += params::degree / params::opt;
-  }
-  __syncthreads();
-
-  // level 1
-  tid = threadIdx.x;
-#pragma unroll
-  for (size_t i = 0; i < params::opt / 2; ++i) {
-    twid_id = tid / (params::degree / 2);
-    i1 = 2 * (params::degree / 2) * twid_id + (tid & (params::degree / 2 - 1));
-    i2 = i1 + params::degree / 2;
-
-    w = negtwiddles[twid_id + 1];
-    u = A[i1] - A[i2];
-
-    A[i1] += A[i2];
-    A[i2] = u * conjugate(w);
-
-    tid += params::degree / params::opt;
+  for (Index i = 0; i < BUTTERFLY_DEPTH; i++) {
+    A[tid] = u[i];
+    A[tid + HALF_DEGREE] = v[i];
+    tid = tid + STRIDE;
   }
   __syncthreads();
 }

backends/tfhe-cuda-backend/cuda/src/fft/twiddles.cu

@@ -1,6 +1,6 @@
 #include "cuComplex.h"
 
-__constant__ double2 negtwiddles[4096] = {
+__device__ double2 negtwiddles[8192] = {
     {0, 0},
     {0.707106781186547461715008466854, 0.707106781186547572737310929369},
     {0.92387953251128673848313610506, 0.382683432365089781779232680492},
@@ -4096,9 +4096,7 @@ __constant__ double2 negtwiddles[4096] = {
     {0.70791982920081630847874976098, 0.706292797233758484765075991163},
     {-0.706292797233758484765075991163, 0.70791982920081630847874976098},
     {0.00115048533711384847431913325266, 0.99999933819152553304832053982},
-    {-0.99999933819152553304832053982, 0.00115048533711384847431913325266}};
-
-__device__ double2 negtwiddles13[4096] = {
+    {-0.99999933819152553304832053982, 0.00115048533711384847431913325266},
     {0.999999981616429334252416083473, 0.000191747597310703291528452552051},
     {-0.000191747597310703291528452552051, 0.999999981616429334252416083473},
     {0.706971182161065359039753275283, 0.707242354213734603085583785287},

backends/tfhe-cuda-backend/cuda/src/fft/twiddles.cuh

@@ -2,12 +2,7 @@
 #define GPU_BOOTSTRAP_TWIDDLES_CUH
 
 /*
- * 'negtwiddles' are stored in constant memory for faster access times
- * because of it's limited size, only twiddles for up to 2^12 polynomial size
- * can be stored there, twiddles for 2^13 are stored in device memory
- * 'negtwiddles13'
+ * 'negtwiddles' are stored in device memory to profit caching
  */
-
-extern __constant__ double2 negtwiddles[4096];
-extern __device__ double2 negtwiddles13[4096];
+extern __device__ double2 negtwiddles[8192];
 #endif

backends/tfhe-cuda-backend/cuda/src/integer/addition.cu

@@ -0,0 +1,49 @@
+#include "integer/addition.cuh"
+
+void scratch_cuda_signed_overflowing_add_or_sub_radix_ciphertext_kb_64(
+    void **streams, uint32_t *gpu_indexes, uint32_t gpu_count, int8_t **mem_ptr,
+    uint32_t glwe_dimension, uint32_t polynomial_size,
+    uint32_t big_lwe_dimension, uint32_t small_lwe_dimension, uint32_t ks_level,
+    uint32_t ks_base_log, uint32_t pbs_level, uint32_t pbs_base_log,
+    uint32_t grouping_factor, uint32_t num_blocks, int8_t signed_operation,
+    uint32_t message_modulus, uint32_t carry_modulus, PBS_TYPE pbs_type,
+    bool allocate_gpu_memory) {
+
+  SIGNED_OPERATION op = (signed_operation == 1) ? SIGNED_OPERATION::ADDITION
+                                                : SIGNED_OPERATION::SUBTRACTION;
+  int_radix_params params(pbs_type, glwe_dimension, polynomial_size,
+                          big_lwe_dimension, small_lwe_dimension, ks_level,
+                          ks_base_log, pbs_level, pbs_base_log, grouping_factor,
+                          message_modulus, carry_modulus);
+
+  scratch_cuda_integer_signed_overflowing_add_or_sub_kb<uint64_t>(
+      (cudaStream_t *)(streams), gpu_indexes, gpu_count,
+      (int_signed_overflowing_add_or_sub_memory<uint64_t> **)mem_ptr,
+      num_blocks, op, params, allocate_gpu_memory);
+}
+
+void cuda_signed_overflowing_add_or_sub_radix_ciphertext_kb_64(
+    void **streams, uint32_t *gpu_indexes, uint32_t gpu_count, void *lhs,
+    void *rhs, void *overflowed, int8_t signed_operation, int8_t *mem_ptr,
+    void **bsks, void **ksks, uint32_t num_blocks) {
+
+  auto mem = (int_signed_overflowing_add_or_sub_memory<uint64_t> *)mem_ptr;
+  SIGNED_OPERATION op = (signed_operation == 1) ? SIGNED_OPERATION::ADDITION
+                                                : SIGNED_OPERATION::SUBTRACTION;
+
+  host_integer_signed_overflowing_add_or_sub_kb<uint64_t>(
+      (cudaStream_t *)(streams), gpu_indexes, gpu_count,
+      static_cast<uint64_t *>(lhs), static_cast<uint64_t *>(rhs),
+      static_cast<uint64_t *>(overflowed), op, bsks, (uint64_t **)(ksks), mem,
+      num_blocks);
+}
+
+void cleanup_signed_overflowing_add_or_sub(void **streams,
+                                           uint32_t *gpu_indexes,
+                                           uint32_t gpu_count,
+                                           int8_t **mem_ptr_void) {
+  int_signed_overflowing_add_or_sub_memory<uint64_t> *mem_ptr =
+      (int_signed_overflowing_add_or_sub_memory<uint64_t> *)(*mem_ptr_void);
+
+  mem_ptr->release((cudaStream_t *)(streams), gpu_indexes, gpu_count);
+}

backends/tfhe-cuda-backend/cuda/src/integer/addition.cuh

@@ -0,0 +1,137 @@
+#ifndef TFHE_RS_ADDITION_CUH
+#define TFHE_RS_ADDITION_CUH
+
+#include "crypto/keyswitch.cuh"
+#include "device.h"
+#include "integer.h"
+#include "integer/comparison.cuh"
+#include "integer/integer.cuh"
+#include "integer/negation.cuh"
+#include "integer/scalar_shifts.cuh"
+#include "linear_algebra.h"
+#include "programmable_bootstrap.h"
+#include "utils/helper.cuh"
+#include "utils/kernel_dimensions.cuh"
+#include <fstream>
+#include <iostream>
+#include <sstream>
+#include <string>
+#include <vector>
+
+template <typename Torus>
+void host_resolve_signed_overflow(
+    cudaStream_t *streams, uint32_t *gpu_indexes, uint32_t gpu_count,
+    Torus *result, Torus *last_block_inner_propagation,
+    Torus *last_block_input_carry, Torus *last_block_output_carry,
+    int_resolve_signed_overflow_memory<Torus> *mem, void **bsks, Torus **ksks) {
+
+  auto x = mem->x;
+
+  Torus *d_clears =
+      (Torus *)cuda_malloc_async(sizeof(Torus), streams[0], gpu_indexes[0]);
+
+  cuda_set_value_async<Torus>(streams[0], gpu_indexes[0], d_clears, 2, 1);
+
+  // replace with host function call
+  cuda_mult_lwe_ciphertext_vector_cleartext_vector_64(
+      streams[0], gpu_indexes[0], x, last_block_output_carry, d_clears,
+      mem->params.big_lwe_dimension, 1);
+
+  host_addition<Torus>(streams[0], gpu_indexes[0], last_block_inner_propagation,
+                       last_block_inner_propagation, x,
+                       mem->params.big_lwe_dimension, 1);
+  host_addition<Torus>(streams[0], gpu_indexes[0], last_block_inner_propagation,
+                       last_block_inner_propagation, last_block_input_carry,
+                       mem->params.big_lwe_dimension, 1);
+
+  host_apply_univariate_lut_kb<Torus>(streams, gpu_indexes, gpu_count, result,
+                                      last_block_inner_propagation,
+                                      mem->resolve_overflow_lut, ksks, bsks, 1);
+
+  cuda_drop_async(d_clears, streams[0], gpu_indexes[0]);
+}
+
+template <typename Torus>
+__host__ void scratch_cuda_integer_signed_overflowing_add_or_sub_kb(
+    cudaStream_t *streams, uint32_t *gpu_indexes, uint32_t gpu_count,
+    int_signed_overflowing_add_or_sub_memory<Torus> **mem_ptr,
+    uint32_t num_blocks, SIGNED_OPERATION op, int_radix_params params,
+    bool allocate_gpu_memory) {
+
+  *mem_ptr = new int_signed_overflowing_add_or_sub_memory<Torus>(
+      streams, gpu_indexes, gpu_count, params, num_blocks, op,
+      allocate_gpu_memory);
+}
+
+/*
+ * Addition - signed_operation = 1
+ * Subtraction - signed_operation = -1
+ */
+template <typename Torus>
+__host__ void host_integer_signed_overflowing_add_or_sub_kb(
+    cudaStream_t *streams, uint32_t *gpu_indexes, uint32_t gpu_count,
+    Torus *lhs, Torus *rhs, Torus *overflowed, SIGNED_OPERATION op, void **bsks,
+    uint64_t **ksks,
+    int_signed_overflowing_add_or_sub_memory<uint64_t> *mem_ptr,
+    uint32_t num_blocks) {
+
+  auto radix_params = mem_ptr->params;
+
+  uint32_t big_lwe_dimension = radix_params.big_lwe_dimension;
+  uint32_t big_lwe_size = big_lwe_dimension + 1;
+  uint32_t big_lwe_size_bytes = big_lwe_size * sizeof(Torus);
+
+  assert(radix_params.message_modulus >= 4 && radix_params.carry_modulus >= 4);
+
+  auto result = mem_ptr->result;
+  auto neg_rhs = mem_ptr->neg_rhs;
+  auto input_carries = mem_ptr->input_carries;
+  auto output_carry = mem_ptr->output_carry;
+  auto last_block_inner_propagation = mem_ptr->last_block_inner_propagation;
+
+  cuda_memcpy_async_gpu_to_gpu(result, lhs, num_blocks * big_lwe_size_bytes,
+                               streams[0], gpu_indexes[0]);
+
+  // phase 1
+  if (op == SIGNED_OPERATION::ADDITION) {
+    host_addition<Torus>(streams[0], gpu_indexes[0], result, lhs, rhs,
+                         big_lwe_dimension, num_blocks);
+  } else {
+    host_integer_radix_negation<Torus>(
+        streams, gpu_indexes, gpu_count, neg_rhs, rhs, big_lwe_dimension,
+        num_blocks, radix_params.message_modulus, radix_params.carry_modulus);
+    host_addition<Torus>(streams[0], gpu_indexes[0], result, lhs, neg_rhs,
+                         big_lwe_dimension, num_blocks);
+  }
+
+  // phase 2
+  for (uint j = 0; j < gpu_count; j++) {
+    cuda_synchronize_stream(streams[j], gpu_indexes[j]);
+  }
+
+  host_propagate_single_carry<Torus>(
+      mem_ptr->sub_streams_1, gpu_indexes, gpu_count, result, output_carry,
+      input_carries, mem_ptr->scp_mem, bsks, ksks, num_blocks);
+  host_generate_last_block_inner_propagation<Torus>(
+      mem_ptr->sub_streams_2, gpu_indexes, gpu_count,
+      last_block_inner_propagation, &lhs[(num_blocks - 1) * big_lwe_size],
+      &rhs[(num_blocks - 1) * big_lwe_size], mem_ptr->las_block_prop_mem, bsks,
+      ksks);
+
+  for (uint j = 0; j < mem_ptr->active_gpu_count; j++) {
+    cuda_synchronize_stream(mem_ptr->sub_streams_1[j], gpu_indexes[j]);
+    cuda_synchronize_stream(mem_ptr->sub_streams_2[j], gpu_indexes[j]);
+  }
+
+  // phase 3
+  auto input_carry = &input_carries[(num_blocks - 1) * big_lwe_size];
+
+  host_resolve_signed_overflow<Torus>(
+      streams, gpu_indexes, gpu_count, overflowed, last_block_inner_propagation,
+      input_carry, output_carry, mem_ptr->resolve_overflow_mem, bsks, ksks);
+
+  cuda_memcpy_async_gpu_to_gpu(lhs, result, num_blocks * big_lwe_size_bytes,
+                               streams[0], gpu_indexes[0]);
+}
+
+#endif // TFHE_RS_ADDITION_CUH

backends/tfhe-cuda-backend/cuda/src/integer/cmux.cu

@@ -17,7 +17,7 @@ void scratch_cuda_integer_radix_cmux_kb_64(
   std::function<uint64_t(uint64_t)> predicate_lut_f =
       [](uint64_t x) -> uint64_t { return x == 1; };
 
-  scratch_cuda_integer_radix_cmux_kb(
+  scratch_cuda_integer_radix_cmux_kb<uint64_t>(
       (cudaStream_t *)(streams), gpu_indexes, gpu_count,
       (int_cmux_buffer<uint64_t> **)mem_ptr, predicate_lut_f,
       lwe_ciphertext_count, params, allocate_gpu_memory);

backends/tfhe-cuda-backend/cuda/src/integer/cmux.cuh

@@ -2,7 +2,6 @@
 #define CUDA_INTEGER_CMUX_CUH
 
 #include "integer.cuh"
-#include <omp.h>
 
 template <typename Torus>
 __host__ void zero_out_if(cudaStream_t *streams, uint32_t *gpu_indexes,
@@ -28,10 +27,11 @@ __host__ void zero_out_if(cudaStream_t *streams, uint32_t *gpu_indexes,
     auto lwe_array_out_block = tmp_lwe_array_input + i * big_lwe_size;
     auto lwe_array_input_block = lwe_array_input + i * big_lwe_size;
 
-    device_pack_bivariate_blocks<<<num_blocks, num_threads, 0, streams[0]>>>(
-        lwe_array_out_block, predicate->lwe_indexes_in, lwe_array_input_block,
-        lwe_condition, predicate->lwe_indexes_in, params.big_lwe_dimension,
-        params.message_modulus, 1);
+    device_pack_bivariate_blocks<Torus>
+        <<<num_blocks, num_threads, 0, streams[0]>>>(
+            lwe_array_out_block, predicate->lwe_indexes_in,
+            lwe_array_input_block, lwe_condition, predicate->lwe_indexes_in,
+            params.big_lwe_dimension, params.message_modulus, 1);
     check_cuda_error(cudaGetLastError());
   }
 
@@ -57,27 +57,20 @@ __host__ void host_integer_radix_cmux_kb(
     cuda_synchronize_stream(streams[j], gpu_indexes[j]);
   }
 
-#pragma omp parallel sections
-  {
-    // Both sections may be executed in parallel
-#pragma omp section
-    {
-      auto mem_true = mem_ptr->zero_if_true_buffer;
-      zero_out_if(true_streams, gpu_indexes, gpu_count, mem_ptr->tmp_true_ct,
-                  lwe_array_true, lwe_condition, mem_true,
-                  mem_ptr->inverted_predicate_lut, bsks, ksks,
-                  num_radix_blocks);
-    }
-#pragma omp section
-    {
-      auto mem_false = mem_ptr->zero_if_false_buffer;
-      zero_out_if(false_streams, gpu_indexes, gpu_count, mem_ptr->tmp_false_ct,
-                  lwe_array_false, lwe_condition, mem_false,
-                  mem_ptr->predicate_lut, bsks, ksks, num_radix_blocks);
-    }
-  }
-  for (uint j = 0; j < gpu_count; j++) {
+  auto mem_true = mem_ptr->zero_if_true_buffer;
+  zero_out_if<Torus>(true_streams, gpu_indexes, gpu_count, mem_ptr->tmp_true_ct,
+                     lwe_array_true, lwe_condition, mem_true,
+                     mem_ptr->inverted_predicate_lut, bsks, ksks,
+                     num_radix_blocks);
+  auto mem_false = mem_ptr->zero_if_false_buffer;
+  zero_out_if<Torus>(false_streams, gpu_indexes, gpu_count,
+                     mem_ptr->tmp_false_ct, lwe_array_false, lwe_condition,
+                     mem_false, mem_ptr->predicate_lut, bsks, ksks,
+                     num_radix_blocks);
+  for (uint j = 0; j < mem_ptr->zero_if_true_buffer->active_gpu_count; j++) {
     cuda_synchronize_stream(true_streams[j], gpu_indexes[j]);
+  }
+  for (uint j = 0; j < mem_ptr->zero_if_false_buffer->active_gpu_count; j++) {
     cuda_synchronize_stream(false_streams[j], gpu_indexes[j]);
   }
 
@@ -85,9 +78,9 @@ __host__ void host_integer_radix_cmux_kb(
   // will be 0 If the condition was false, true_ct will be 0 and false_ct will
   // have kept its value
   auto added_cts = mem_ptr->tmp_true_ct;
-  host_addition(streams[0], gpu_indexes[0], added_cts, mem_ptr->tmp_true_ct,
-                mem_ptr->tmp_false_ct, params.big_lwe_dimension,
-                num_radix_blocks);
+  host_addition<Torus>(streams[0], gpu_indexes[0], added_cts,
+                       mem_ptr->tmp_true_ct, mem_ptr->tmp_false_ct,
+                       params.big_lwe_dimension, num_radix_blocks);
 
   integer_radix_apply_univariate_lookup_table_kb<Torus>(
       streams, gpu_indexes, gpu_count, lwe_array_out, added_cts, bsks, ksks,

backends/tfhe-cuda-backend/cuda/src/integer/comparison.cuh

@@ -43,7 +43,7 @@ __host__ void accumulate_all_blocks(cudaStream_t stream, uint32_t gpu_index,
   int num_entries = (lwe_dimension + 1);
   getNumBlocksAndThreads(num_entries, 512, num_blocks, num_threads);
   // Add all blocks and store in sum
-  device_accumulate_all_blocks<<<num_blocks, num_threads, 0, stream>>>(
+  device_accumulate_all_blocks<Torus><<<num_blocks, num_threads, 0, stream>>>(
       output, input, lwe_dimension, num_radix_blocks);
   check_cuda_error(cudaGetLastError());
 }
@@ -62,7 +62,6 @@ __host__ void are_all_comparisons_block_true(
     int_comparison_buffer<Torus> *mem_ptr, void **bsks, Torus **ksks,
     uint32_t num_radix_blocks) {
 
-  cudaSetDevice(gpu_indexes[0]);
   auto params = mem_ptr->params;
   auto big_lwe_dimension = params.big_lwe_dimension;
   auto glwe_dimension = params.glwe_dimension;
@@ -75,7 +74,7 @@ __host__ void are_all_comparisons_block_true(
   auto tmp_out = are_all_block_true_buffer->tmp_out;
 
   uint32_t total_modulus = message_modulus * carry_modulus;
-  uint32_t max_value = total_modulus - 1;
+  uint32_t max_value = (total_modulus - 1) / (message_modulus - 1);
 
   cuda_memcpy_async_gpu_to_gpu(tmp_out, lwe_array_in,
                                num_radix_blocks * (big_lwe_dimension + 1) *
@@ -96,8 +95,9 @@ __host__ void are_all_comparisons_block_true(
     auto is_equal_to_num_blocks_map =
         &are_all_block_true_buffer->is_equal_to_lut_map;
     for (int i = 0; i < num_chunks; i++) {
-      accumulate_all_blocks(streams[0], gpu_indexes[0], accumulator,
-                            input_blocks, big_lwe_dimension, chunk_length);
+      accumulate_all_blocks<Torus>(streams[0], gpu_indexes[0], accumulator,
+                                   input_blocks, big_lwe_dimension,
+                                   chunk_length);
 
       accumulator += (big_lwe_dimension + 1);
       remaining_blocks -= (chunk_length - 1);
@@ -121,9 +121,8 @@ __host__ void are_all_comparisons_block_true(
             new int_radix_lut<Torus>(streams, gpu_indexes, gpu_count, params,
                                      max_value, num_radix_blocks, true);
 
-        auto is_equal_to_num_blocks_lut_f = [max_value,
-                                             chunk_length](Torus x) -> Torus {
-          return (x & max_value) == chunk_length;
+        auto is_equal_to_num_blocks_lut_f = [chunk_length](Torus x) -> Torus {
+          return x == chunk_length;
         };
         generate_device_accumulator<Torus>(
             streams[0], gpu_indexes[0], new_lut->get_lut(gpu_indexes[0], 0),
@@ -165,7 +164,6 @@ __host__ void is_at_least_one_comparisons_block_true(
     int_comparison_buffer<Torus> *mem_ptr, void **bsks, Torus **ksks,
     uint32_t num_radix_blocks) {
 
-  cudaSetDevice(gpu_indexes[0]);
   auto params = mem_ptr->params;
   auto big_lwe_dimension = params.big_lwe_dimension;
   auto message_modulus = params.message_modulus;
@@ -174,7 +172,7 @@ __host__ void is_at_least_one_comparisons_block_true(
   auto buffer = mem_ptr->eq_buffer->are_all_block_true_buffer;
 
   uint32_t total_modulus = message_modulus * carry_modulus;
-  uint32_t max_value = total_modulus - 1;
+  uint32_t max_value = (total_modulus - 1) / (message_modulus - 1);
 
   cuda_memcpy_async_gpu_to_gpu(mem_ptr->tmp_lwe_array_out, lwe_array_in,
                                num_radix_blocks * (big_lwe_dimension + 1) *
@@ -192,8 +190,9 @@ __host__ void is_at_least_one_comparisons_block_true(
     auto input_blocks = mem_ptr->tmp_lwe_array_out;
     auto accumulator = buffer->tmp_block_accumulated;
     for (int i = 0; i < num_chunks; i++) {
-      accumulate_all_blocks(streams[0], gpu_indexes[0], accumulator,
-                            input_blocks, big_lwe_dimension, chunk_length);
+      accumulate_all_blocks<Torus>(streams[0], gpu_indexes[0], accumulator,
+                                   input_blocks, big_lwe_dimension,
+                                   chunk_length);
 
       accumulator += (big_lwe_dimension + 1);
       remaining_blocks -= (chunk_length - 1);
@@ -245,7 +244,6 @@ __host__ void host_compare_with_zero_equality(
     int_comparison_buffer<Torus> *mem_ptr, void **bsks, Torus **ksks,
     int32_t num_radix_blocks, int_radix_lut<Torus> *zero_comparison) {
 
-  cudaSetDevice(gpu_indexes[0]);
   auto params = mem_ptr->params;
   auto big_lwe_dimension = params.big_lwe_dimension;
   auto message_modulus = params.message_modulus;
@@ -281,8 +279,8 @@ __host__ void host_compare_with_zero_equality(
       uint32_t chunk_size =
           std::min(remainder_blocks, num_elements_to_fill_carry);
 
-      accumulate_all_blocks(streams[0], gpu_indexes[0], sum_i, chunk,
-                            big_lwe_dimension, chunk_size);
+      accumulate_all_blocks<Torus>(streams[0], gpu_indexes[0], sum_i, chunk,
+                                   big_lwe_dimension, chunk_size);
 
       num_sum_blocks++;
       remainder_blocks -= (chunk_size - 1);
@@ -296,8 +294,9 @@ __host__ void host_compare_with_zero_equality(
   integer_radix_apply_univariate_lookup_table_kb<Torus>(
       streams, gpu_indexes, gpu_count, sum, sum, bsks, ksks, num_sum_blocks,
       zero_comparison);
-  are_all_comparisons_block_true(streams, gpu_indexes, gpu_count, lwe_array_out,
-                                 sum, mem_ptr, bsks, ksks, num_sum_blocks);
+  are_all_comparisons_block_true<Torus>(streams, gpu_indexes, gpu_count,
+                                        lwe_array_out, sum, mem_ptr, bsks, ksks,
+                                        num_sum_blocks);
 }
 
 template <typename Torus>
@@ -311,7 +310,7 @@ __host__ void host_integer_radix_equality_check_kb(
 
   // Applies the LUT for the comparison operation
   auto comparisons = mem_ptr->tmp_block_comparisons;
-  integer_radix_apply_bivariate_lookup_table_kb(
+  integer_radix_apply_bivariate_lookup_table_kb<Torus>(
       streams, gpu_indexes, gpu_count, comparisons, lwe_array_1, lwe_array_2,
       bsks, ksks, num_radix_blocks, eq_buffer->operator_lut,
       eq_buffer->operator_lut->params.message_modulus);
@@ -320,9 +319,9 @@ __host__ void host_integer_radix_equality_check_kb(
   //
   // It returns a block encrypting 1 if all input blocks are 1
   // otherwise the block encrypts 0
-  are_all_comparisons_block_true(streams, gpu_indexes, gpu_count, lwe_array_out,
-                                 comparisons, mem_ptr, bsks, ksks,
-                                 num_radix_blocks);
+  are_all_comparisons_block_true<Torus>(streams, gpu_indexes, gpu_count,
+                                        lwe_array_out, comparisons, mem_ptr,
+                                        bsks, ksks, num_radix_blocks);
 }
 
 template <typename Torus>
@@ -353,19 +352,20 @@ compare_radix_blocks_kb(cudaStream_t *streams, uint32_t *gpu_indexes,
 
   // Subtract
   // Here we need the true lwe sub, not the one that comes from shortint.
-  host_subtraction(streams[0], gpu_indexes[0], lwe_array_out, lwe_array_left,
-                   lwe_array_right, big_lwe_dimension, num_radix_blocks);
+  host_subtraction<Torus>(streams[0], gpu_indexes[0], lwe_array_out,
+                          lwe_array_left, lwe_array_right, big_lwe_dimension,
+                          num_radix_blocks);
 
   // Apply LUT to compare to 0
   auto is_non_zero_lut = mem_ptr->eq_buffer->is_non_zero_lut;
-  integer_radix_apply_univariate_lookup_table_kb(
+  integer_radix_apply_univariate_lookup_table_kb<Torus>(
       streams, gpu_indexes, gpu_count, lwe_array_out, lwe_array_out, bsks, ksks,
       num_radix_blocks, is_non_zero_lut);
 
   // Add one
   // Here Lhs can have the following values: (-1) % (message modulus * carry
   // modulus), 0, 1 So the output values after the addition will be: 0, 1, 2
-  host_integer_radix_add_scalar_one_inplace(
+  host_integer_radix_add_scalar_one_inplace<Torus>(
       streams, gpu_indexes, gpu_count, lwe_array_out, big_lwe_dimension,
       num_radix_blocks, message_modulus, carry_modulus);
 }
@@ -407,8 +407,8 @@ tree_sign_reduction(cudaStream_t *streams, uint32_t *gpu_indexes,
 
   auto inner_tree_leaf = tree_buffer->tree_inner_leaf_lut;
   while (partial_block_count > 2) {
-    pack_blocks(streams[0], gpu_indexes[0], y, x, big_lwe_dimension,
-                partial_block_count, 4);
+    pack_blocks<Torus>(streams[0], gpu_indexes[0], y, x, big_lwe_dimension,
+                       partial_block_count, 4);
 
     integer_radix_apply_univariate_lookup_table_kb<Torus>(
         streams, gpu_indexes, gpu_count, x, y, bsks, ksks,
@@ -434,8 +434,8 @@ tree_sign_reduction(cudaStream_t *streams, uint32_t *gpu_indexes,
   std::function<Torus(Torus)> f;
 
   if (partial_block_count == 2) {
-    pack_blocks(streams[0], gpu_indexes[0], y, x, big_lwe_dimension,
-                partial_block_count, 4);
+    pack_blocks<Torus>(streams[0], gpu_indexes[0], y, x, big_lwe_dimension,
+                       partial_block_count, 4);
 
     f = [block_selector_f, sign_handler_f](Torus x) -> Torus {
       int msb = (x >> 2) & 3;
@@ -455,9 +455,9 @@ tree_sign_reduction(cudaStream_t *streams, uint32_t *gpu_indexes,
   last_lut->broadcast_lut(streams, gpu_indexes, gpu_indexes[0]);
 
   // Last leaf
-  integer_radix_apply_univariate_lookup_table_kb(streams, gpu_indexes,
-                                                 gpu_count, lwe_array_out, y,
-                                                 bsks, ksks, 1, last_lut);
+  integer_radix_apply_univariate_lookup_table_kb<Torus>(
+      streams, gpu_indexes, gpu_count, lwe_array_out, y, bsks, ksks, 1,
+      last_lut);
 }
 
 template <typename Torus>
@@ -489,19 +489,21 @@ __host__ void host_integer_radix_difference_check_kb(
     if (mem_ptr->is_signed) {
       packed_num_radix_blocks -= 2;
     }
-    pack_blocks(streams[0], gpu_indexes[0], packed_left, lwe_array_left,
-                big_lwe_dimension, packed_num_radix_blocks, message_modulus);
-    pack_blocks(streams[0], gpu_indexes[0], packed_right, lwe_array_right,
-                big_lwe_dimension, packed_num_radix_blocks, message_modulus);
+    pack_blocks<Torus>(streams[0], gpu_indexes[0], packed_left, lwe_array_left,
+                       big_lwe_dimension, packed_num_radix_blocks,
+                       message_modulus);
+    pack_blocks<Torus>(streams[0], gpu_indexes[0], packed_right,
+                       lwe_array_right, big_lwe_dimension,
+                       packed_num_radix_blocks, message_modulus);
     // From this point we have half number of blocks
     packed_num_radix_blocks /= 2;
 
     // Clean noise
     auto identity_lut = mem_ptr->identity_lut;
-    integer_radix_apply_univariate_lookup_table_kb(
+    integer_radix_apply_univariate_lookup_table_kb<Torus>(
         streams, gpu_indexes, gpu_count, packed_left, packed_left, bsks, ksks,
         packed_num_radix_blocks, identity_lut);
-    integer_radix_apply_univariate_lookup_table_kb(
+    integer_radix_apply_univariate_lookup_table_kb<Torus>(
         streams, gpu_indexes, gpu_count, packed_right, packed_right, bsks, ksks,
         packed_num_radix_blocks, identity_lut);
 
@@ -518,16 +520,17 @@ __host__ void host_integer_radix_difference_check_kb(
   if (!mem_ptr->is_signed) {
     // Compare packed blocks, or simply the total number of radix blocks in the
     // inputs
-    compare_radix_blocks_kb(streams, gpu_indexes, gpu_count, comparisons, lhs,
-                            rhs, mem_ptr, bsks, ksks, packed_num_radix_blocks);
+    compare_radix_blocks_kb<Torus>(streams, gpu_indexes, gpu_count, comparisons,
+                                   lhs, rhs, mem_ptr, bsks, ksks,
+                                   packed_num_radix_blocks);
     num_comparisons = packed_num_radix_blocks;
   } else {
     // Packing is possible
     if (carry_modulus >= message_modulus) {
       // Compare (num_radix_blocks - 2) / 2 packed blocks
-      compare_radix_blocks_kb(streams, gpu_indexes, gpu_count, comparisons, lhs,
-                              rhs, mem_ptr, bsks, ksks,
-                              packed_num_radix_blocks);
+      compare_radix_blocks_kb<Torus>(streams, gpu_indexes, gpu_count,
+                                     comparisons, lhs, rhs, mem_ptr, bsks, ksks,
+                                     packed_num_radix_blocks);
 
       // Compare the last block before the sign block separately
       auto identity_lut = mem_ptr->identity_lut;
@@ -536,21 +539,21 @@ __host__ void host_integer_radix_difference_check_kb(
       Torus *last_right_block_before_sign_block =
           diff_buffer->tmp_packed_right +
           packed_num_radix_blocks * big_lwe_size;
-      integer_radix_apply_univariate_lookup_table_kb(
+      integer_radix_apply_univariate_lookup_table_kb<Torus>(
           streams, gpu_indexes, gpu_count, last_left_block_before_sign_block,
           lwe_array_left + (num_radix_blocks - 2) * big_lwe_size, bsks, ksks, 1,
           identity_lut);
-      integer_radix_apply_univariate_lookup_table_kb(
+      integer_radix_apply_univariate_lookup_table_kb<Torus>(
           streams, gpu_indexes, gpu_count, last_right_block_before_sign_block,
           lwe_array_right + (num_radix_blocks - 2) * big_lwe_size, bsks, ksks,
           1, identity_lut);
-      compare_radix_blocks_kb(
+      compare_radix_blocks_kb<Torus>(
           streams, gpu_indexes, gpu_count,
           comparisons + packed_num_radix_blocks * big_lwe_size,
           last_left_block_before_sign_block, last_right_block_before_sign_block,
           mem_ptr, bsks, ksks, 1);
       // Compare the sign block separately
-      integer_radix_apply_bivariate_lookup_table_kb(
+      integer_radix_apply_bivariate_lookup_table_kb<Torus>(
           streams, gpu_indexes, gpu_count,
           comparisons + (packed_num_radix_blocks + 1) * big_lwe_size,
           lwe_array_left + (num_radix_blocks - 1) * big_lwe_size,
@@ -559,11 +562,11 @@ __host__ void host_integer_radix_difference_check_kb(
       num_comparisons = packed_num_radix_blocks + 2;
 
     } else {
-      compare_radix_blocks_kb(streams, gpu_indexes, gpu_count, comparisons,
-                              lwe_array_left, lwe_array_right, mem_ptr, bsks,
-                              ksks, num_radix_blocks - 1);
+      compare_radix_blocks_kb<Torus>(
+          streams, gpu_indexes, gpu_count, comparisons, lwe_array_left,
+          lwe_array_right, mem_ptr, bsks, ksks, num_radix_blocks - 1);
       // Compare the sign block separately
-      integer_radix_apply_bivariate_lookup_table_kb(
+      integer_radix_apply_bivariate_lookup_table_kb<Torus>(
           streams, gpu_indexes, gpu_count,
           comparisons + (num_radix_blocks - 1) * big_lwe_size,
           lwe_array_left + (num_radix_blocks - 1) * big_lwe_size,
@@ -576,9 +579,9 @@ __host__ void host_integer_radix_difference_check_kb(
   // Reduces a vec containing radix blocks that encrypts a sign
   // (inferior, equal, superior) to one single radix block containing the
   // final sign
-  tree_sign_reduction(streams, gpu_indexes, gpu_count, lwe_array_out,
-                      comparisons, mem_ptr->diff_buffer->tree_buffer,
-                      reduction_lut_f, bsks, ksks, num_comparisons);
+  tree_sign_reduction<Torus>(streams, gpu_indexes, gpu_count, lwe_array_out,
+                             comparisons, mem_ptr->diff_buffer->tree_buffer,
+                             reduction_lut_f, bsks, ksks, num_comparisons);
 }
 
 template <typename Torus>
@@ -602,16 +605,16 @@ host_integer_radix_maxmin_kb(cudaStream_t *streams, uint32_t *gpu_indexes,
                              Torus **ksks, uint32_t total_num_radix_blocks) {
 
   // Compute the sign
-  host_integer_radix_difference_check_kb(
+  host_integer_radix_difference_check_kb<Torus>(
       streams, gpu_indexes, gpu_count, mem_ptr->tmp_lwe_array_out,
       lwe_array_left, lwe_array_right, mem_ptr, mem_ptr->identity_lut_f, bsks,
       ksks, total_num_radix_blocks);
 
   // Selector
-  host_integer_radix_cmux_kb(streams, gpu_indexes, gpu_count, lwe_array_out,
-                             mem_ptr->tmp_lwe_array_out, lwe_array_left,
-                             lwe_array_right, mem_ptr->cmux_buffer, bsks, ksks,
-                             total_num_radix_blocks);
+  host_integer_radix_cmux_kb<Torus>(
+      streams, gpu_indexes, gpu_count, lwe_array_out,
+      mem_ptr->tmp_lwe_array_out, lwe_array_left, lwe_array_right,
+      mem_ptr->cmux_buffer, bsks, ksks, total_num_radix_blocks);
 }
 
 #endif

backends/tfhe-cuda-backend/cuda/src/integer/compression/compression.cu

@@ -0,0 +1,89 @@
+#include "compression.cuh"
+
+void scratch_cuda_integer_compress_radix_ciphertext_64(
+    void **streams, uint32_t *gpu_indexes, uint32_t gpu_count, int8_t **mem_ptr,
+    uint32_t compression_glwe_dimension, uint32_t compression_polynomial_size,
+    uint32_t lwe_dimension, uint32_t ks_level, uint32_t ks_base_log,
+    uint32_t num_radix_blocks, uint32_t message_modulus, uint32_t carry_modulus,
+    PBS_TYPE pbs_type, uint32_t lwe_per_glwe, uint32_t storage_log_modulus,
+    bool allocate_gpu_memory) {
+
+  int_radix_params compression_params(
+      pbs_type, compression_glwe_dimension, compression_polynomial_size,
+      (compression_glwe_dimension + 1) * compression_polynomial_size,
+      lwe_dimension, ks_level, ks_base_log, 0, 0, 0, message_modulus,
+      carry_modulus);
+
+  scratch_cuda_compress_integer_radix_ciphertext<uint64_t>(
+      (cudaStream_t *)(streams), gpu_indexes, gpu_count,
+      (int_compression<uint64_t> **)mem_ptr, num_radix_blocks,
+      compression_params, lwe_per_glwe, storage_log_modulus,
+      allocate_gpu_memory);
+}
+void scratch_cuda_integer_decompress_radix_ciphertext_64(
+    void **streams, uint32_t *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, uint32_t compression_polynomial_size,
+    uint32_t lwe_dimension, uint32_t pbs_level, uint32_t pbs_base_log,
+    uint32_t num_radix_blocks, uint32_t message_modulus, uint32_t carry_modulus,
+    PBS_TYPE pbs_type, uint32_t storage_log_modulus, uint32_t body_count,
+    bool allocate_gpu_memory) {
+
+  // Decompression doesn't keyswitch, so big and small dimensions are the same
+  int_radix_params encryption_params(
+      pbs_type, encryption_glwe_dimension, encryption_polynomial_size,
+      lwe_dimension, lwe_dimension, 0, 0, pbs_level, pbs_base_log, 0,
+      message_modulus, carry_modulus);
+
+  int_radix_params compression_params(
+      pbs_type, compression_glwe_dimension, compression_polynomial_size,
+      lwe_dimension, compression_glwe_dimension * compression_polynomial_size,
+      0, 0, pbs_level, pbs_base_log, 0, message_modulus, carry_modulus);
+
+  scratch_cuda_integer_decompress_radix_ciphertext<uint64_t>(
+      (cudaStream_t *)(streams), gpu_indexes, gpu_count,
+      (int_decompression<uint64_t> **)mem_ptr, num_radix_blocks, body_count,
+      encryption_params, compression_params, storage_log_modulus,
+      allocate_gpu_memory);
+}
+void cuda_integer_compress_radix_ciphertext_64(
+    void **streams, uint32_t *gpu_indexes, uint32_t gpu_count,
+    void *glwe_array_out, void *lwe_array_in, void **fp_ksk, uint32_t num_nths,
+    int8_t *mem_ptr) {
+
+  host_integer_compress<uint64_t>(
+      (cudaStream_t *)(streams), gpu_indexes, gpu_count,
+      static_cast<uint64_t *>(glwe_array_out),
+      static_cast<uint64_t *>(lwe_array_in), (uint64_t **)(fp_ksk), num_nths,
+      (int_compression<uint64_t> *)mem_ptr);
+}
+void cuda_integer_decompress_radix_ciphertext_64(
+    void **streams, uint32_t *gpu_indexes, uint32_t gpu_count,
+    void *lwe_array_out, void *glwe_in, uint32_t *indexes_array,
+    uint32_t indexes_array_size, void **bsks, int8_t *mem_ptr) {
+
+  host_integer_decompress<uint64_t>(
+      (cudaStream_t *)(streams), gpu_indexes, gpu_count,
+      static_cast<uint64_t *>(lwe_array_out), static_cast<uint64_t *>(glwe_in),
+      indexes_array, indexes_array_size, bsks,
+      (int_decompression<uint64_t> *)mem_ptr);
+}
+
+void cleanup_cuda_integer_compress_radix_ciphertext_64(void **streams,
+                                                       uint32_t *gpu_indexes,
+                                                       uint32_t gpu_count,
+                                                       int8_t **mem_ptr_void) {
+
+  int_compression<uint64_t> *mem_ptr =
+      (int_compression<uint64_t> *)(*mem_ptr_void);
+  mem_ptr->release((cudaStream_t *)(streams), gpu_indexes, gpu_count);
+}
+
+void cleanup_cuda_integer_decompress_radix_ciphertext_64(
+    void **streams, uint32_t *gpu_indexes, uint32_t gpu_count,
+    int8_t **mem_ptr_void) {
+
+  int_decompression<uint64_t> *mem_ptr =
+      (int_decompression<uint64_t> *)(*mem_ptr_void);
+  mem_ptr->release((cudaStream_t *)(streams), gpu_indexes, gpu_count);
+}

backends/tfhe-cuda-backend/cuda/src/integer/compression/compression.cuh

@@ -0,0 +1,388 @@
+#ifndef CUDA_INTEGER_COMPRESSION_CUH
+#define CUDA_INTEGER_COMPRESSION_CUH
+
+#include "ciphertext.h"
+#include "compression.h"
+#include "crypto/keyswitch.cuh"
+#include "device.h"
+#include "integer/integer.cuh"
+#include "linearalgebra/multiplication.cuh"
+#include "polynomial/functions.cuh"
+#include "utils/kernel_dimensions.cuh"
+
+template <typename Torus>
+__global__ void pack(Torus *array_out, Torus *array_in, uint32_t log_modulus,
+                     uint32_t num_coeffs, uint32_t in_len, uint32_t out_len) {
+  auto nbits = sizeof(Torus) * 8;
+  auto tid = threadIdx.x + blockIdx.x * blockDim.x;
+
+  auto glwe_index = tid / out_len;
+  auto i = tid % out_len;
+  auto chunk_array_in = array_in + glwe_index * in_len;
+  auto chunk_array_out = array_out + glwe_index * out_len;
+
+  if (tid < num_coeffs) {
+
+    auto k = nbits * i / log_modulus;
+    auto j = k;
+
+    auto start_shift = i * nbits - j * log_modulus;
+
+    auto value = chunk_array_in[j] >> start_shift;
+    j++;
+
+    while (j * log_modulus < ((i + 1) * nbits) && j < in_len) {
+      auto shift = j * log_modulus - i * nbits;
+      value |= chunk_array_in[j] << shift;
+      j++;
+    }
+
+    chunk_array_out[i] = value;
+  }
+}
+
+template <typename Torus>
+__host__ void host_pack(cudaStream_t stream, uint32_t gpu_index,
+                        Torus *array_out, Torus *array_in, uint32_t num_glwes,
+                        uint32_t num_lwes, int_compression<Torus> *mem_ptr) {
+  if (array_in == array_out)
+    PANIC("Cuda error: Input and output must be different");
+
+  cudaSetDevice(gpu_index);
+  auto compression_params = mem_ptr->compression_params;
+
+  auto log_modulus = mem_ptr->storage_log_modulus;
+  // [0..num_glwes-1) GLWEs
+  auto in_len = (compression_params.glwe_dimension + 1) *
+                compression_params.polynomial_size;
+  auto number_bits_to_pack = in_len * log_modulus;
+  auto nbits = sizeof(Torus) * 8;
+  // number_bits_to_pack.div_ceil(Scalar::BITS)
+  auto out_len = (number_bits_to_pack + nbits - 1) / nbits;
+
+  // Last GLWE
+  auto last_body_count = num_lwes % compression_params.polynomial_size;
+  in_len =
+      compression_params.glwe_dimension * compression_params.polynomial_size +
+      last_body_count;
+  number_bits_to_pack = in_len * log_modulus;
+  auto last_out_len = (number_bits_to_pack + nbits - 1) / nbits;
+
+  auto num_coeffs = (num_glwes - 1) * out_len + last_out_len;
+
+  int num_blocks = 0, num_threads = 0;
+  getNumBlocksAndThreads(num_coeffs, 1024, num_blocks, num_threads);
+
+  dim3 grid(num_blocks);
+  dim3 threads(num_threads);
+  cuda_memset_async(array_out, 0,
+                    num_glwes * (compression_params.glwe_dimension + 1) *
+                        compression_params.polynomial_size * sizeof(Torus),
+                    stream, gpu_index);
+  pack<Torus><<<grid, threads, 0, stream>>>(array_out, array_in, log_modulus,
+                                            num_coeffs, in_len, out_len);
+  check_cuda_error(cudaGetLastError());
+}
+
+template <typename Torus>
+__host__ void host_integer_compress(cudaStream_t *streams,
+                                    uint32_t *gpu_indexes, uint32_t gpu_count,
+                                    Torus *glwe_array_out, Torus *lwe_array_in,
+                                    Torus **fp_ksk, uint32_t num_radix_blocks,
+                                    int_compression<Torus> *mem_ptr) {
+
+  auto compression_params = mem_ptr->compression_params;
+  auto input_lwe_dimension = compression_params.small_lwe_dimension;
+
+  // Shift
+  auto lwe_shifted = mem_ptr->tmp_lwe;
+  host_cleartext_multiplication<Torus>(
+      streams[0], gpu_indexes[0], lwe_shifted, lwe_array_in,
+      (uint64_t)compression_params.message_modulus, input_lwe_dimension,
+      num_radix_blocks);
+
+  uint32_t lwe_in_size = input_lwe_dimension + 1;
+  uint32_t glwe_out_size = (compression_params.glwe_dimension + 1) *
+                           compression_params.polynomial_size;
+  uint32_t num_glwes_for_compression =
+      num_radix_blocks / mem_ptr->lwe_per_glwe + 1;
+
+  // Keyswitch LWEs to GLWE
+  auto tmp_glwe_array_out = mem_ptr->tmp_glwe_array_out;
+  cuda_memset_async(tmp_glwe_array_out, 0,
+                    num_glwes_for_compression *
+                        (compression_params.glwe_dimension + 1) *
+                        compression_params.polynomial_size * sizeof(Torus),
+                    streams[0], gpu_indexes[0]);
+  auto fp_ks_buffer = mem_ptr->fp_ks_buffer;
+  auto rem_lwes = num_radix_blocks;
+
+  auto lwe_subset = lwe_shifted;
+  auto glwe_out = tmp_glwe_array_out;
+  while (rem_lwes > 0) {
+    auto chunk_size = min(rem_lwes, mem_ptr->lwe_per_glwe);
+
+    host_packing_keyswitch_lwe_list_to_glwe<Torus>(
+        streams[0], gpu_indexes[0], glwe_out, lwe_subset, fp_ksk[0],
+        fp_ks_buffer, input_lwe_dimension, compression_params.glwe_dimension,
+        compression_params.polynomial_size, compression_params.ks_base_log,
+        compression_params.ks_level, chunk_size);
+
+    rem_lwes -= chunk_size;
+    lwe_subset += chunk_size * lwe_in_size;
+    glwe_out += glwe_out_size;
+  }
+
+  // Modulus switch
+  host_modulus_switch_inplace<Torus>(
+      streams[0], gpu_indexes[0], tmp_glwe_array_out,
+      num_glwes_for_compression * (compression_params.glwe_dimension + 1) *
+          compression_params.polynomial_size,
+      mem_ptr->storage_log_modulus);
+
+  host_pack<Torus>(streams[0], gpu_indexes[0], glwe_array_out,
+                   tmp_glwe_array_out, num_glwes_for_compression,
+                   num_radix_blocks, mem_ptr);
+}
+
+template <typename Torus>
+__global__ void extract(Torus *glwe_array_out, Torus *array_in, uint32_t index,
+                        uint32_t log_modulus, uint32_t input_len,
+                        uint32_t initial_out_len) {
+  auto nbits = sizeof(Torus) * 8;
+
+  auto i = threadIdx.x + blockIdx.x * blockDim.x;
+  auto chunk_array_in = array_in + index * input_len;
+  if (i < initial_out_len) {
+    // Unpack
+    Torus mask = ((Torus)1 << log_modulus) - 1;
+    auto start = i * log_modulus;
+    auto end = (i + 1) * log_modulus;
+
+    auto start_block = start / nbits;
+    auto start_remainder = start % nbits;
+
+    auto end_block_inclusive = (end - 1) / nbits;
+
+    Torus unpacked_i;
+    if (start_block == end_block_inclusive) {
+      auto single_part = chunk_array_in[start_block] >> start_remainder;
+      unpacked_i = single_part & mask;
+    } else {
+      auto first_part = chunk_array_in[start_block] >> start_remainder;
+      auto second_part = chunk_array_in[start_block + 1]
+                         << (nbits - start_remainder);
+
+      unpacked_i = (first_part | second_part) & mask;
+    }
+
+    // Extract
+    glwe_array_out[i] = unpacked_i << (nbits - log_modulus);
+  }
+}
+
+/// Extracts the glwe_index-nth GLWE ciphertext
+template <typename Torus>
+__host__ void host_extract(cudaStream_t stream, uint32_t gpu_index,
+                           Torus *glwe_array_out, Torus *array_in,
+                           uint32_t glwe_index,
+                           int_decompression<Torus> *mem_ptr) {
+  if (array_in == glwe_array_out)
+    PANIC("Cuda error: Input and output must be different");
+
+  cudaSetDevice(gpu_index);
+
+  auto compression_params = mem_ptr->compression_params;
+
+  auto log_modulus = mem_ptr->storage_log_modulus;
+
+  uint32_t body_count =
+      std::min(mem_ptr->body_count, compression_params.polynomial_size);
+  auto initial_out_len =
+      compression_params.glwe_dimension * compression_params.polynomial_size +
+      body_count;
+
+  auto compressed_glwe_accumulator_size =
+      (compression_params.glwe_dimension + 1) *
+      compression_params.polynomial_size;
+  auto number_bits_to_unpack = compressed_glwe_accumulator_size * log_modulus;
+  auto nbits = sizeof(Torus) * 8;
+  // number_bits_to_unpack.div_ceil(Scalar::BITS)
+  auto input_len = (number_bits_to_unpack + nbits - 1) / nbits;
+
+  // We assure the tail of the glwe is zeroed
+  auto zeroed_slice = glwe_array_out + initial_out_len;
+  cuda_memset_async(zeroed_slice, 0,
+                    (compression_params.polynomial_size - body_count) *
+                        sizeof(Torus),
+                    stream, gpu_index);
+  int num_blocks = 0, num_threads = 0;
+  getNumBlocksAndThreads(initial_out_len, 128, num_blocks, num_threads);
+  dim3 grid(num_blocks);
+  dim3 threads(num_threads);
+  extract<Torus><<<grid, threads, 0, stream>>>(glwe_array_out, array_in,
+                                               glwe_index, log_modulus,
+                                               input_len, initial_out_len);
+  check_cuda_error(cudaGetLastError());
+}
+
+template <typename Torus>
+__host__ void
+host_integer_decompress(cudaStream_t *streams, uint32_t *gpu_indexes,
+                        uint32_t gpu_count, Torus *d_lwe_array_out,
+                        Torus *d_packed_glwe_in, uint32_t *h_indexes_array,
+                        uint32_t indexes_array_size, void **d_bsks,
+                        int_decompression<Torus> *h_mem_ptr) {
+
+  auto d_indexes_array = h_mem_ptr->tmp_indexes_array;
+  cuda_memcpy_async_to_gpu(d_indexes_array, h_indexes_array,
+                           indexes_array_size * sizeof(uint32_t), streams[0],
+                           gpu_indexes[0]);
+
+  auto compression_params = h_mem_ptr->compression_params;
+  auto lwe_per_glwe = compression_params.polynomial_size;
+  if (indexes_array_size > lwe_per_glwe)
+    PANIC("Cuda error: too many LWEs to decompress. The number of LWEs should "
+          "be smaller than "
+          "polynomial_size.")
+
+  auto num_radix_blocks = h_mem_ptr->num_radix_blocks;
+  if (num_radix_blocks != indexes_array_size)
+    PANIC("Cuda error: wrong number of LWEs in decompress: the number of LWEs "
+          "should be the same as indexes_array_size.")
+
+  // the first element is the last index in h_indexes_array that lies in the
+  // related GLWE
+  std::vector<std::pair<int, Torus *>> glwe_vec;
+
+  // Extract all GLWEs
+  Torus glwe_accumulator_size = (compression_params.glwe_dimension + 1) *
+                                compression_params.polynomial_size;
+
+  auto current_glwe_index = h_indexes_array[0] / lwe_per_glwe;
+  auto extracted_glwe = h_mem_ptr->tmp_extracted_glwe;
+  host_extract<Torus>(streams[0], gpu_indexes[0], extracted_glwe,
+                      d_packed_glwe_in, current_glwe_index, h_mem_ptr);
+  glwe_vec.push_back(std::make_pair(0, extracted_glwe));
+  for (int i = 1; i < indexes_array_size; i++) {
+    auto glwe_index = h_indexes_array[i] / lwe_per_glwe;
+    if (glwe_index != current_glwe_index) {
+      extracted_glwe += glwe_accumulator_size;
+      current_glwe_index = glwe_index;
+      // Extracts a new GLWE
+      host_extract<Torus>(streams[0], gpu_indexes[0], extracted_glwe,
+                          d_packed_glwe_in, glwe_index, h_mem_ptr);
+      glwe_vec.push_back(std::make_pair(i, extracted_glwe));
+    } else {
+      // Updates the index
+      glwe_vec.back().first++;
+    }
+  }
+  // Sample extract all LWEs
+  Torus lwe_accumulator_size = compression_params.small_lwe_dimension + 1;
+
+  auto extracted_lwe = h_mem_ptr->tmp_extracted_lwe;
+  uint32_t current_idx = 0;
+  auto d_indexes_array_chunk = d_indexes_array;
+  for (const auto &max_idx_and_glwe : glwe_vec) {
+    uint32_t last_idx = max_idx_and_glwe.first;
+    extracted_glwe = max_idx_and_glwe.second;
+
+    auto num_lwes = last_idx + 1 - current_idx;
+    cuda_glwe_sample_extract_64(streams[0], gpu_indexes[0], extracted_lwe,
+                                extracted_glwe, d_indexes_array_chunk, num_lwes,
+                                compression_params.glwe_dimension,
+                                compression_params.polynomial_size);
+    d_indexes_array_chunk += num_lwes;
+    extracted_lwe += lwe_accumulator_size;
+    current_idx = last_idx;
+  }
+
+  // Reset
+  extracted_lwe = h_mem_ptr->tmp_extracted_lwe;
+
+  // In the case of extracting a single LWE these parameters are dummy
+  uint32_t lut_count = 1;
+  uint32_t lut_stride = 0;
+  /// Apply PBS to apply a LUT, reduce the noise and go from a small LWE
+  /// dimension to a big LWE dimension
+  auto encryption_params = h_mem_ptr->encryption_params;
+  auto lut = h_mem_ptr->carry_extract_lut;
+  auto active_gpu_count = get_active_gpu_count(num_radix_blocks, gpu_count);
+  if (active_gpu_count == 1) {
+    execute_pbs_async<Torus>(
+        streams, gpu_indexes, active_gpu_count, d_lwe_array_out,
+        lut->lwe_indexes_out, lut->lut_vec, lut->lut_indexes_vec, extracted_lwe,
+        lut->lwe_indexes_in, d_bsks, lut->buffer,
+        encryption_params.glwe_dimension,
+        compression_params.small_lwe_dimension,
+        encryption_params.polynomial_size, encryption_params.pbs_base_log,
+        encryption_params.pbs_level, encryption_params.grouping_factor,
+        num_radix_blocks, encryption_params.pbs_type, lut_count, lut_stride);
+  } else {
+    /// For multi GPU execution we create vectors of pointers for inputs and
+    /// outputs
+    std::vector<Torus *> lwe_array_in_vec = lut->lwe_array_in_vec;
+    std::vector<Torus *> lwe_after_pbs_vec = lut->lwe_after_pbs_vec;
+    std::vector<Torus *> lwe_trivial_indexes_vec = lut->lwe_trivial_indexes_vec;
+
+    /// Make sure all data that should be on GPU 0 is indeed there
+    cuda_synchronize_stream(streams[0], gpu_indexes[0]);
+
+    /// With multiple GPUs we push to the vectors on each GPU then when we
+    /// gather data to GPU 0 we can copy back to the original indexing
+    multi_gpu_scatter_lwe_async<Torus>(
+        streams, gpu_indexes, active_gpu_count, lwe_array_in_vec, extracted_lwe,
+        lut->h_lwe_indexes_in, lut->using_trivial_lwe_indexes, num_radix_blocks,
+        compression_params.small_lwe_dimension + 1);
+
+    /// Apply PBS
+    execute_pbs_async<Torus>(
+        streams, gpu_indexes, active_gpu_count, lwe_after_pbs_vec,
+        lwe_trivial_indexes_vec, lut->lut_vec, lut->lut_indexes_vec,
+        lwe_array_in_vec, lwe_trivial_indexes_vec, d_bsks, lut->buffer,
+        encryption_params.glwe_dimension,
+        compression_params.small_lwe_dimension,
+        encryption_params.polynomial_size, encryption_params.pbs_base_log,
+        encryption_params.pbs_level, encryption_params.grouping_factor,
+        num_radix_blocks, encryption_params.pbs_type, lut_count, lut_stride);
+
+    /// Copy data back to GPU 0 and release vecs
+    multi_gpu_gather_lwe_async<Torus>(
+        streams, gpu_indexes, active_gpu_count, d_lwe_array_out,
+        lwe_after_pbs_vec, lut->h_lwe_indexes_out,
+        lut->using_trivial_lwe_indexes, num_radix_blocks,
+        encryption_params.big_lwe_dimension + 1);
+
+    /// Synchronize all GPUs
+    for (uint i = 0; i < active_gpu_count; i++) {
+      cuda_synchronize_stream(streams[i], gpu_indexes[i]);
+    }
+  }
+}
+
+template <typename Torus>
+__host__ void scratch_cuda_compress_integer_radix_ciphertext(
+    cudaStream_t *streams, uint32_t *gpu_indexes, uint32_t gpu_count,
+    int_compression<Torus> **mem_ptr, uint32_t num_radix_blocks,
+    int_radix_params compression_params, uint32_t lwe_per_glwe,
+    uint32_t storage_log_modulus, bool allocate_gpu_memory) {
+
+  *mem_ptr = new int_compression<Torus>(
+      streams, gpu_indexes, gpu_count, compression_params, num_radix_blocks,
+      lwe_per_glwe, storage_log_modulus, allocate_gpu_memory);
+}
+
+template <typename Torus>
+__host__ void scratch_cuda_integer_decompress_radix_ciphertext(
+    cudaStream_t *streams, uint32_t *gpu_indexes, uint32_t gpu_count,
+    int_decompression<Torus> **mem_ptr, uint32_t num_radix_blocks,
+    uint32_t body_count, int_radix_params encryption_params,
+    int_radix_params compression_params, uint32_t storage_log_modulus,
+    bool allocate_gpu_memory) {
+
+  *mem_ptr = new int_decompression<Torus>(
+      streams, gpu_indexes, gpu_count, encryption_params, compression_params,
+      num_radix_blocks, body_count, storage_log_modulus, allocate_gpu_memory);
+}
+#endif

backends/tfhe-cuda-backend/cuda/src/integer/div_rem.cu

@@ -26,54 +26,11 @@ void cuda_integer_div_rem_radix_ciphertext_kb_64(
 
   auto mem = (int_div_rem_memory<uint64_t> *)mem_ptr;
 
-  switch (mem->params.polynomial_size) {
-  case 512:
-    host_integer_div_rem_kb<uint64_t, Degree<512>>(
-        (cudaStream_t *)(streams), gpu_indexes, gpu_count,
-        static_cast<uint64_t *>(quotient), static_cast<uint64_t *>(remainder),
-        static_cast<uint64_t *>(numerator), static_cast<uint64_t *>(divisor),
-        bsks, (uint64_t **)(ksks), mem, num_blocks);
-    break;
-  case 1024:
-
-    host_integer_div_rem_kb<uint64_t, Degree<1024>>(
-        (cudaStream_t *)(streams), gpu_indexes, gpu_count,
-        static_cast<uint64_t *>(quotient), static_cast<uint64_t *>(remainder),
-        static_cast<uint64_t *>(numerator), static_cast<uint64_t *>(divisor),
-        bsks, (uint64_t **)(ksks), mem, num_blocks);
-    break;
-  case 2048:
-    host_integer_div_rem_kb<uint64_t, Degree<2048>>(
-        (cudaStream_t *)(streams), gpu_indexes, gpu_count,
-        static_cast<uint64_t *>(quotient), static_cast<uint64_t *>(remainder),
-        static_cast<uint64_t *>(numerator), static_cast<uint64_t *>(divisor),
-        bsks, (uint64_t **)(ksks), mem, num_blocks);
-    break;
-  case 4096:
-    host_integer_div_rem_kb<uint64_t, Degree<4096>>(
-        (cudaStream_t *)(streams), gpu_indexes, gpu_count,
-        static_cast<uint64_t *>(quotient), static_cast<uint64_t *>(remainder),
-        static_cast<uint64_t *>(numerator), static_cast<uint64_t *>(divisor),
-        bsks, (uint64_t **)(ksks), mem, num_blocks);
-    break;
-  case 8192:
-    host_integer_div_rem_kb<uint64_t, Degree<8192>>(
-        (cudaStream_t *)(streams), gpu_indexes, gpu_count,
-        static_cast<uint64_t *>(quotient), static_cast<uint64_t *>(remainder),
-        static_cast<uint64_t *>(numerator), static_cast<uint64_t *>(divisor),
-        bsks, (uint64_t **)(ksks), mem, num_blocks);
-    break;
-  case 16384:
-    host_integer_div_rem_kb<uint64_t, Degree<16384>>(
-        (cudaStream_t *)(streams), gpu_indexes, gpu_count,
-        static_cast<uint64_t *>(quotient), static_cast<uint64_t *>(remainder),
-        static_cast<uint64_t *>(numerator), static_cast<uint64_t *>(divisor),
-        bsks, (uint64_t **)(ksks), mem, num_blocks);
-    break;
-  default:
-    PANIC("Cuda error (integer div_rem): unsupported polynomial size. "
-          "Only N = 512, 1024, 2048, 4096, 8192, 16384 is supported")
-  }
+  host_integer_div_rem_kb<uint64_t>(
+      (cudaStream_t *)(streams), gpu_indexes, gpu_count,
+      static_cast<uint64_t *>(quotient), static_cast<uint64_t *>(remainder),
+      static_cast<uint64_t *>(numerator), static_cast<uint64_t *>(divisor),
+      bsks, (uint64_t **)(ksks), mem, num_blocks);
 }
 
 void cleanup_cuda_integer_div_rem(void **streams, uint32_t *gpu_indexes,