feat(core_crypto): add glwe keyswitch

.cargo/config.toml

@@ -1,6 +1,2 @@
 [alias]
 xtask = "run --manifest-path ./tasks/Cargo.toml --"
-
-# Accessed by wasm-bindgen when testing for the wasm target
-[target.wasm32-unknown-unknown]
-runner = 'wasm-bindgen-test-runner'

.github/actions/hyperstack_setup/action.yml

@@ -1,53 +0,0 @@
-name: Setup Cuda
-description: Setup Cuda on Hyperstack instance
-
-inputs:
-  cuda-version:
-    description: Version of Cuda to use
-    required: true
-  gcc-version:
-    description: Version of GCC to use
-    required: true
-  cmake-version:
-    description: Version of cmake to use
-    default: 3.29.6
-
-runs:
-  using: "composite"
-  steps:
-    # Mandatory on hyperstack since a bootable volume is not re-usable yet.
-    - name: Install dependencies
-      shell: bash
-      run: |
-        sudo apt update
-        sudo apt install -y checkinstall zlib1g-dev libssl-dev libclang-dev
-        wget https://github.com/Kitware/CMake/releases/download/v${{ inputs.cmake-version }}/cmake-${{ inputs.cmake-version }}.tar.gz
-        tar -zxvf cmake-${{ inputs.cmake-version }}.tar.gz
-        cd cmake-${{ inputs.cmake-version }}
-        ./bootstrap
-        make -j"$(nproc)"
-        sudo make install
-
-    - name: Export CUDA variables
-      shell: bash
-      run: |
-        CUDA_PATH=/usr/local/cuda-${{ inputs.cuda-version }}
-        echo "CUDA_PATH=$CUDA_PATH" >> "${GITHUB_ENV}"
-        echo "$CUDA_PATH/bin" >> "${GITHUB_PATH}"
-        echo "LD_LIBRARY_PATH=$CUDA_PATH/lib:$LD_LIBRARY_PATH" >> "${GITHUB_ENV}"
-        echo "CUDACXX=/usr/local/cuda-${{ inputs.cuda-version }}/bin/nvcc" >> "${GITHUB_ENV}"
-
-    # Specify the correct host compilers
-    - name: Export gcc and g++ variables
-      shell: bash
-      run: |
-        {
-          echo "CC=/usr/bin/gcc-${{ inputs.gcc-version }}";
-          echo "CXX=/usr/bin/g++-${{ inputs.gcc-version }}";
-          echo "CUDAHOSTCXX=/usr/bin/g++-${{ inputs.gcc-version }}";
-          echo "HOME=/home/ubuntu";
-        } >> "${GITHUB_ENV}"
-
-    - name: Check device is detected
-      shell: bash
-      run: nvidia-smi

.github/workflows/aws_tfhe_backward_compat_tests.yml

@@ -26,7 +26,7 @@ jobs:
     steps:
       - name: Start instance
         id: start-instance
-        uses: zama-ai/slab-github-runner@79939325c3c429837c10d6041e4fd8589d328bac
+        uses: zama-ai/slab-github-runner@801df0b8db5ea2b06128b7476c652f5ed5f193a8
         with:
           mode: start
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}
@@ -50,7 +50,7 @@ jobs:
           token: ${{ secrets.FHE_ACTIONS_TOKEN }}
 
       - name: Install latest stable
-        uses: dtolnay/rust-toolchain@a54c7afa936fefeb4456b2dd8068152669aa8203
+        uses: dtolnay/rust-toolchain@7b1c307e0dcbda6122208f10795a713336a9b35a
         with:
           toolchain: stable
 
@@ -100,7 +100,7 @@ jobs:
     steps:
       - name: Stop instance
         id: stop-instance
-        uses: zama-ai/slab-github-runner@79939325c3c429837c10d6041e4fd8589d328bac
+        uses: zama-ai/slab-github-runner@801df0b8db5ea2b06128b7476c652f5ed5f193a8
         with:
           mode: stop
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}

.github/workflows/aws_tfhe_fast_tests.yml

@@ -11,26 +11,16 @@ 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' || github.event_name == 'pull_request_target' }}
+  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:
-  pull_request_target:
 
 jobs:
-  check-user-permission:
-    if: github.event_name == 'pull_request_target'
-    uses: ./.github/workflows/check_triggering_actor.yml
-    secrets:
-      TOKEN: ${{ secrets.GITHUB_TOKEN }}
-
   should-run:
     runs-on: ubuntu-latest
-    needs: check-user-permission
-    if: github.event_name != 'pull_request_target' ||
-      needs.check-user-permission.result == 'success'
     permissions:
       pull-requests: write
     outputs:
@@ -64,24 +54,21 @@ jobs:
         uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
         with:
           fetch-depth: 0
-          token: ${{ secrets.FHE_ACTIONS_TOKEN }}
-          ref: ${{ github.event.pull_request.head.sha }}
 
       - name: Check for file changes
         id: changed-files
-        uses: tj-actions/changed-files@d6e91a2266cdb9d62096cebf1e8546899c6aa18f
+        uses: tj-actions/changed-files@c3a1bb2c992d77180ae65be6ae6c166cf40f857c
         with:
           since_last_remote_commit: true
           files_yaml: |
             dependencies:
               - tfhe/Cargo.toml
-              - tfhe-csprng/**
-              - tfhe-fft/**
+              - concrete-csprng/**
               - tfhe-zk-pok/**
               - utils/tfhe-versionable/**
               - utils/tfhe-versionable-derive/**
             csprng:
-              - tfhe-csprng/**
+              - concrete-csprng/**
             zk_pok:
               - tfhe-zk-pok/**
             versionable:
@@ -144,7 +131,7 @@ jobs:
     steps:
       - name: Start instance
         id: start-instance
-        uses: zama-ai/slab-github-runner@79939325c3c429837c10d6041e4fd8589d328bac
+        uses: zama-ai/slab-github-runner@801df0b8db5ea2b06128b7476c652f5ed5f193a8
         with:
           mode: start
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}
@@ -159,7 +146,7 @@ jobs:
       (github.event_name == 'pull_request' && needs.setup-instance.result != 'skipped')
     needs: [ should-run, setup-instance ]
     concurrency:
-      group: ${{ github.workflow }}_${{ github.head_ref || github.ref }}
+      group: ${{ github.workflow }}_${{ github.ref }}
       cancel-in-progress: true
     runs-on: ${{ needs.setup-instance.outputs.runner-name }}
     steps:
@@ -168,17 +155,16 @@ jobs:
         with:
           persist-credentials: 'false'
           token: ${{ secrets.FHE_ACTIONS_TOKEN }}
-          ref: ${{ github.event.pull_request.head.sha }}
 
       - name: Install latest stable
-        uses: dtolnay/rust-toolchain@a54c7afa936fefeb4456b2dd8068152669aa8203
+        uses: dtolnay/rust-toolchain@7b1c307e0dcbda6122208f10795a713336a9b35a
         with:
           toolchain: stable
 
-      - name: Run tfhe-csprng tests
+      - name: Run concrete-csprng tests
         if: needs.should-run.outputs.csprng_test == 'true'
         run: |
-          make test_tfhe_csprng
+          make test_concrete_csprng
 
       - name: Run tfhe-zk-pok tests
         if: needs.should-run.outputs.zk_pok_test == 'true'
@@ -211,7 +197,7 @@ jobs:
 
       - name: Node cache restoration
         id: node-cache
-        uses: actions/cache/restore@1bd1e32a3bdc45362d1e726936510720a7c30a57 #v4.2.0
+        uses: actions/cache/restore@6849a6489940f00c2f30c0fb92c6274307ccb58a #v4.1.2
         with:
           path: |
             ~/.nvm
@@ -224,7 +210,7 @@ jobs:
           make install_node
 
       - name: Node cache save
-        uses: actions/cache/save@1bd1e32a3bdc45362d1e726936510720a7c30a57 #v4.2.0
+        uses: actions/cache/save@6849a6489940f00c2f30c0fb92c6274307ccb58a #v4.1.2
         if: steps.node-cache.outputs.cache-hit != 'true'
         with:
           path: |
@@ -258,13 +244,9 @@ jobs:
         run: |
           make test_high_level_api
 
-      - name: Run safe serialization tests
+      - name: Run safe deserialization tests
         run: |
-          make test_safe_serialization
-
-      - name: Run zk tests
-        run: |
-          make test_zk
+          make test_safe_deserialization
 
       - name: Slack Notification
         if: ${{ failure() }}
@@ -282,7 +264,7 @@ jobs:
     steps:
       - name: Stop instance
         id: stop-instance
-        uses: zama-ai/slab-github-runner@79939325c3c429837c10d6041e4fd8589d328bac
+        uses: zama-ai/slab-github-runner@801df0b8db5ea2b06128b7476c652f5ed5f193a8
         with:
           mode: stop
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}

.github/workflows/aws_tfhe_integer_tests.yml

@@ -42,24 +42,21 @@ jobs:
         uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
         with:
           fetch-depth: 0
-          token: ${{ secrets.FHE_ACTIONS_TOKEN }}
           persist-credentials: "false"
 
       - name: Check for file changes
         id: changed-files
-        uses: tj-actions/changed-files@d6e91a2266cdb9d62096cebf1e8546899c6aa18f
+        uses: tj-actions/changed-files@c3a1bb2c992d77180ae65be6ae6c166cf40f857c
         with:
           since_last_remote_commit: true
           files_yaml: |
             integer:
               - tfhe/Cargo.toml
-              - tfhe-csprng/**
-              - tfhe-fft/**
+              - concrete-csprng/**
               - tfhe-zk-pok/**
               - tfhe/src/core_crypto/**
               - tfhe/src/shortint/**
               - tfhe/src/integer/**
-              - .github/workflows/aws_tfhe_integer_tests.yml
 
   setup-instance:
     name: Setup instance (unsigned-integer-tests)
@@ -75,7 +72,7 @@ jobs:
     steps:
       - name: Start instance
         id: start-instance
-        uses: zama-ai/slab-github-runner@79939325c3c429837c10d6041e4fd8589d328bac
+        uses: zama-ai/slab-github-runner@801df0b8db5ea2b06128b7476c652f5ed5f193a8
         with:
           mode: start
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}
@@ -99,7 +96,7 @@ jobs:
           token: ${{ secrets.FHE_ACTIONS_TOKEN }}
 
       - name: Install latest stable
-        uses: dtolnay/rust-toolchain@a54c7afa936fefeb4456b2dd8068152669aa8203
+        uses: dtolnay/rust-toolchain@7b1c307e0dcbda6122208f10795a713336a9b35a
         with:
           toolchain: stable
 
@@ -140,7 +137,7 @@ jobs:
     steps:
       - name: Stop instance
         id: stop-instance
-        uses: zama-ai/slab-github-runner@79939325c3c429837c10d6041e4fd8589d328bac
+        uses: zama-ai/slab-github-runner@801df0b8db5ea2b06128b7476c652f5ed5f193a8
         with:
           mode: stop
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}

.github/workflows/aws_tfhe_signed_integer_tests.yml

@@ -42,24 +42,21 @@ jobs:
         uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
         with:
           fetch-depth: 0
-          token: ${{ secrets.FHE_ACTIONS_TOKEN }}
           persist-credentials: "false"
 
       - name: Check for file changes
         id: changed-files
-        uses: tj-actions/changed-files@d6e91a2266cdb9d62096cebf1e8546899c6aa18f
+        uses: tj-actions/changed-files@c3a1bb2c992d77180ae65be6ae6c166cf40f857c
         with:
           since_last_remote_commit: true
           files_yaml: |
             integer:
               - tfhe/Cargo.toml
-              - tfhe-csprng/**
-              - tfhe-fft/**
+              - concrete-csprng/**
               - tfhe-zk-pok/**
               - tfhe/src/core_crypto/**
               - tfhe/src/shortint/**
               - tfhe/src/integer/**
-              - .github/workflows/aws_tfhe_signed_integer_tests.yml
 
   setup-instance:
     name: Setup instance (unsigned-integer-tests)
@@ -75,7 +72,7 @@ jobs:
     steps:
       - name: Start instance
         id: start-instance
-        uses: zama-ai/slab-github-runner@79939325c3c429837c10d6041e4fd8589d328bac
+        uses: zama-ai/slab-github-runner@801df0b8db5ea2b06128b7476c652f5ed5f193a8
         with:
           mode: start
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}
@@ -99,7 +96,7 @@ jobs:
           token: ${{ secrets.FHE_ACTIONS_TOKEN }}
 
       - name: Install latest stable
-        uses: dtolnay/rust-toolchain@a54c7afa936fefeb4456b2dd8068152669aa8203
+        uses: dtolnay/rust-toolchain@7b1c307e0dcbda6122208f10795a713336a9b35a
         with:
           toolchain: stable
 
@@ -144,7 +141,7 @@ jobs:
     steps:
       - name: Stop instance
         id: stop-instance
-        uses: zama-ai/slab-github-runner@79939325c3c429837c10d6041e4fd8589d328bac
+        uses: zama-ai/slab-github-runner@801df0b8db5ea2b06128b7476c652f5ed5f193a8
         with:
           mode: stop
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}

.github/workflows/aws_tfhe_tests.yml

@@ -40,9 +40,6 @@ jobs:
       shortint_test: ${{ env.IS_PULL_REQUEST == 'false' ||
         steps.changed-files.outputs.shortint_any_changed ||
         steps.changed-files.outputs.dependencies_any_changed }}
-      strings_test: ${{ env.IS_PULL_REQUEST == 'false' ||
-        steps.changed-files.outputs.strings_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 }}
@@ -63,21 +60,19 @@ jobs:
         uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
         with:
           fetch-depth: 0
-          token: ${{ secrets.FHE_ACTIONS_TOKEN }}
 
       - name: Check for file changes
         id: changed-files
-        uses: tj-actions/changed-files@d6e91a2266cdb9d62096cebf1e8546899c6aa18f
+        uses: tj-actions/changed-files@c3a1bb2c992d77180ae65be6ae6c166cf40f857c
         with:
           since_last_remote_commit: true
           files_yaml: |
             dependencies:
               - tfhe/Cargo.toml
-              - tfhe-csprng/**
-              - tfhe-fft/**
+              - concrete-csprng/**
               - tfhe-zk-pok/**
             csprng:
-              - tfhe-csprng/**
+              - concrete-csprng/**
             zk_pok:
               - tfhe-zk-pok/**
             core_crypto:
@@ -88,11 +83,6 @@ jobs:
             shortint:
               - tfhe/src/core_crypto/**
               - tfhe/src/shortint/**
-            strings:
-              - tfhe/src/core_crypto/**
-              - tfhe/src/shortint/**
-              - tfhe/src/integer/**
-              - tfhe/src/strings/**
             high_level_api:
               - tfhe/src/**
               - '!tfhe/src/c_api/**'
@@ -122,7 +112,6 @@ jobs:
           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.strings_any_changed == 'true' ||
           steps.changed-files.outputs.high_level_api_any_changed == 'true' ||
           steps.changed-files.outputs.c_api_any_changed == 'true' ||
           steps.changed-files.outputs.examples_any_changed == 'true' ||
@@ -142,7 +131,7 @@ jobs:
     steps:
       - name: Start instance
         id: start-instance
-        uses: zama-ai/slab-github-runner@79939325c3c429837c10d6041e4fd8589d328bac
+        uses: zama-ai/slab-github-runner@801df0b8db5ea2b06128b7476c652f5ed5f193a8
         with:
           mode: start
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}
@@ -168,14 +157,14 @@ jobs:
           token: ${{ secrets.FHE_ACTIONS_TOKEN }}
 
       - name: Install latest stable
-        uses: dtolnay/rust-toolchain@a54c7afa936fefeb4456b2dd8068152669aa8203
+        uses: dtolnay/rust-toolchain@7b1c307e0dcbda6122208f10795a713336a9b35a
         with:
           toolchain: stable
 
-      - name: Run tfhe-csprng tests
+      - name: Run concrete-csprng tests
         if: needs.should-run.outputs.csprng_test == 'true'
         run: |
-          make test_tfhe_csprng
+          make test_concrete_csprng
 
       - name: Run tfhe-zk-pok tests
         if: needs.should-run.outputs.zk_pok_test == 'true'
@@ -212,11 +201,6 @@ jobs:
         run: |
           BIG_TESTS_INSTANCE=TRUE make test_shortint_ci
 
-      - name: Run strings tests
-        if: needs.should-run.outputs.strings_test == 'true'
-        run: |
-          BIG_TESTS_INSTANCE=TRUE make test_strings
-
       - name: Run high-level API tests
         if: needs.should-run.outputs.high_level_api_test == 'true'
         run: |
@@ -250,7 +234,7 @@ jobs:
     steps:
       - name: Stop instance
         id: stop-instance
-        uses: zama-ai/slab-github-runner@79939325c3c429837c10d6041e4fd8589d328bac
+        uses: zama-ai/slab-github-runner@801df0b8db5ea2b06128b7476c652f5ed5f193a8
         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@79939325c3c429837c10d6041e4fd8589d328bac
+        uses: zama-ai/slab-github-runner@801df0b8db5ea2b06128b7476c652f5ed5f193a8
         with:
           mode: start
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}
@@ -51,7 +51,7 @@ jobs:
           token: ${{ secrets.FHE_ACTIONS_TOKEN }}
 
       - name: Install latest stable
-        uses: dtolnay/rust-toolchain@a54c7afa936fefeb4456b2dd8068152669aa8203
+        uses: dtolnay/rust-toolchain@7b1c307e0dcbda6122208f10795a713336a9b35a
         with:
           toolchain: stable
 
@@ -61,7 +61,7 @@ jobs:
 
       - name: Node cache restoration
         id: node-cache
-        uses: actions/cache/restore@1bd1e32a3bdc45362d1e726936510720a7c30a57 #v4.2.0
+        uses: actions/cache/restore@6849a6489940f00c2f30c0fb92c6274307ccb58a #v4.1.2
         with:
           path: |
             ~/.nvm
@@ -74,7 +74,7 @@ jobs:
           make install_node
 
       - name: Node cache save
-        uses: actions/cache/save@1bd1e32a3bdc45362d1e726936510720a7c30a57 #v4.2.0
+        uses: actions/cache/save@6849a6489940f00c2f30c0fb92c6274307ccb58a #v4.1.2
         if: steps.node-cache.outputs.cache-hit != 'true'
         with:
           path: |
@@ -99,10 +99,6 @@ jobs:
         run: |
           make test_web_js_api_parallel_chrome_ci
 
-      - name: Run x86_64/wasm zk compatibility tests
-        run: |
-          make test_zk_wasm_x86_compat_ci
-
       - name: Slack Notification
         if: ${{ failure() }}
         continue-on-error: true
@@ -119,7 +115,7 @@ jobs:
     steps:
       - name: Stop instance
         id: stop-instance
-        uses: zama-ai/slab-github-runner@79939325c3c429837c10d6041e4fd8589d328bac
+        uses: zama-ai/slab-github-runner@801df0b8db5ea2b06128b7476c652f5ed5f193a8
         with:
           mode: stop
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}

.github/workflows/benchmark_boolean.yml

@@ -29,7 +29,7 @@ jobs:
     steps:
       - name: Start instance
         id: start-instance
-        uses: zama-ai/slab-github-runner@79939325c3c429837c10d6041e4fd8589d328bac
+        uses: zama-ai/slab-github-runner@801df0b8db5ea2b06128b7476c652f5ed5f193a8
         with:
           mode: start
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}
@@ -61,8 +61,13 @@ jobs:
             echo "COMMIT_HASH=$(git describe --tags --dirty)";
           } >> "${GITHUB_ENV}"
 
+      - name: Set up home
+        # "Install rust" step require root user to have a HOME directory which is not set.
+        run: |
+          echo "HOME=/home/ubuntu" >> "${GITHUB_ENV}"
+
       - name: Install rust
-        uses: dtolnay/rust-toolchain@a54c7afa936fefeb4456b2dd8068152669aa8203
+        uses: dtolnay/rust-toolchain@7b1c307e0dcbda6122208f10795a713336a9b35a
         with:
           toolchain: nightly
 
@@ -80,7 +85,8 @@ jobs:
           --commit-date "${{ env.COMMIT_DATE }}" \
           --bench-date "${{ env.BENCH_DATE }}" \
           --walk-subdirs \
-          --name-suffix avx512
+          --name-suffix avx512 \
+          --throughput
 
       - name: Measure key sizes
         run: |
@@ -93,7 +99,7 @@ jobs:
           --append-results
 
       - name: Upload parsed results artifact
-        uses: actions/upload-artifact@6f51ac03b9356f520e9adb1b1b7802705f340c2b
+        uses: actions/upload-artifact@b4b15b8c7c6ac21ea08fcf65892d2ee8f75cf882
         with:
           name: ${{ github.sha }}_boolean
           path: ${{ env.RESULTS_FILENAME }}
@@ -127,7 +133,7 @@ jobs:
     steps:
       - name: Stop instance
         id: stop-instance
-        uses: zama-ai/slab-github-runner@79939325c3c429837c10d6041e4fd8589d328bac
+        uses: zama-ai/slab-github-runner@801df0b8db5ea2b06128b7476c652f5ed5f193a8
         with:
           mode: stop
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}

.github/workflows/benchmark_core_crypto.yml

@@ -26,7 +26,7 @@ jobs:
     steps:
       - name: Start instance
         id: start-instance
-        uses: zama-ai/slab-github-runner@79939325c3c429837c10d6041e4fd8589d328bac
+        uses: zama-ai/slab-github-runner@801df0b8db5ea2b06128b7476c652f5ed5f193a8
         with:
           mode: start
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}
@@ -57,8 +57,13 @@ jobs:
             echo "COMMIT_HASH=$(git describe --tags --dirty)";
           } >> "${GITHUB_ENV}"
 
+      - name: Set up home
+        # "Install rust" step require root user to have a HOME directory which is not set.
+        run: |
+          echo "HOME=/home/ubuntu" >> "${GITHUB_ENV}"
+
       - name: Install rust
-        uses: dtolnay/rust-toolchain@a54c7afa936fefeb4456b2dd8068152669aa8203
+        uses: dtolnay/rust-toolchain@7b1c307e0dcbda6122208f10795a713336a9b35a
         with:
           toolchain: nightly
 
@@ -78,10 +83,11 @@ jobs:
           --commit-date "${{ env.COMMIT_DATE }}" \
           --bench-date "${{ env.BENCH_DATE }}" \
           --name-suffix avx512 \
-          --walk-subdirs
+          --walk-subdirs \
+          --throughput
 
       - name: Upload parsed results artifact
-        uses: actions/upload-artifact@6f51ac03b9356f520e9adb1b1b7802705f340c2b
+        uses: actions/upload-artifact@b4b15b8c7c6ac21ea08fcf65892d2ee8f75cf882
         with:
           name: ${{ github.sha }}_core_crypto
           path: ${{ env.RESULTS_FILENAME }}
@@ -115,7 +121,7 @@ jobs:
     steps:
       - name: Stop instance
         id: stop-instance
-        uses: zama-ai/slab-github-runner@79939325c3c429837c10d6041e4fd8589d328bac
+        uses: zama-ai/slab-github-runner@801df0b8db5ea2b06128b7476c652f5ed5f193a8
         with:
           mode: stop
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}

.github/workflows/benchmark_erc20.yml

@@ -29,7 +29,7 @@ jobs:
     steps:
       - name: Start instance
         id: start-instance
-        uses: zama-ai/slab-github-runner@79939325c3c429837c10d6041e4fd8589d328bac
+        uses: zama-ai/slab-github-runner@801df0b8db5ea2b06128b7476c652f5ed5f193a8
         with:
           mode: start
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}
@@ -62,8 +62,13 @@ jobs:
             echo "COMMIT_HASH=$(git describe --tags --dirty)";
           } >> "${GITHUB_ENV}"
 
+      - name: Set up home
+        # "Install rust" step require root user to have a HOME directory which is not set.
+        run: |
+          echo "HOME=/home/ubuntu" >> "${GITHUB_ENV}"
+
       - name: Install rust
-        uses: dtolnay/rust-toolchain@a54c7afa936fefeb4456b2dd8068152669aa8203
+        uses: dtolnay/rust-toolchain@7b1c307e0dcbda6122208f10795a713336a9b35a
         with:
           toolchain: nightly
 
@@ -97,7 +102,7 @@ jobs:
           --append-results
 
       - name: Upload parsed results artifact
-        uses: actions/upload-artifact@6f51ac03b9356f520e9adb1b1b7802705f340c2b
+        uses: actions/upload-artifact@b4b15b8c7c6ac21ea08fcf65892d2ee8f75cf882
         with:
           name: ${{ github.sha }}_erc20
           path: ${{ env.RESULTS_FILENAME }}
@@ -124,7 +129,7 @@ jobs:
     steps:
       - name: Stop instance
         id: stop-instance
-        uses: zama-ai/slab-github-runner@79939325c3c429837c10d6041e4fd8589d328bac
+        uses: zama-ai/slab-github-runner@801df0b8db5ea2b06128b7476c652f5ed5f193a8
         with:
           mode: stop
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}

.github/workflows/benchmark_gpu_4090.yml

@@ -54,7 +54,7 @@ jobs:
           echo "FAST_BENCH=TRUE" >> "${GITHUB_ENV}"
 
       - name: Install rust
-        uses: dtolnay/rust-toolchain@a54c7afa936fefeb4456b2dd8068152669aa8203
+        uses: dtolnay/rust-toolchain@7b1c307e0dcbda6122208f10795a713336a9b35a
         with:
           toolchain: nightly
 
@@ -79,10 +79,11 @@ jobs:
           --branch ${{ github.ref_name }} \
           --commit-date "${{ env.COMMIT_DATE }}" \
           --bench-date "${{ env.BENCH_DATE }}" \
-          --walk-subdirs
+          --walk-subdirs \
+          --throughput
 
       - name: Upload parsed results artifact
-        uses: actions/upload-artifact@6f51ac03b9356f520e9adb1b1b7802705f340c2b
+        uses: actions/upload-artifact@b4b15b8c7c6ac21ea08fcf65892d2ee8f75cf882
         with:
           name: ${{ github.sha }}_integer_multi_bit_gpu_default
           path: ${{ env.RESULTS_FILENAME }}
@@ -116,7 +117,6 @@ jobs:
         uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
         with:
           fetch-depth: 0
-          token: ${{ secrets.FHE_ACTIONS_TOKEN }}
 
       - name: Get benchmark details
         run: |
@@ -127,7 +127,7 @@ jobs:
           } >> "${GITHUB_ENV}"
 
       - name: Install rust
-        uses: dtolnay/rust-toolchain@a54c7afa936fefeb4456b2dd8068152669aa8203
+        uses: dtolnay/rust-toolchain@7b1c307e0dcbda6122208f10795a713336a9b35a
         with:
           toolchain: nightly
 
@@ -154,10 +154,10 @@ jobs:
           --commit-date "${{ env.COMMIT_DATE }}" \
           --bench-date "${{ env.BENCH_DATE }}" \
           --walk-subdirs \
-      
+          --throughput
 
       - name: Upload parsed results artifact
-        uses: actions/upload-artifact@6f51ac03b9356f520e9adb1b1b7802705f340c2b
+        uses: actions/upload-artifact@b4b15b8c7c6ac21ea08fcf65892d2ee8f75cf882
         with:
           name: ${{ github.sha }}_core_crypto
           path: ${{ env.RESULTS_FILENAME }}

.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@79939325c3c429837c10d6041e4fd8589d328bac
+        uses: zama-ai/slab-github-runner@801df0b8db5ea2b06128b7476c652f5ed5f193a8
         with:
           mode: start
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}
@@ -48,19 +48,29 @@ jobs:
           - os: ubuntu-22.04
             cuda: "12.2"
             gcc: 11
+    env:
+      CUDA_PATH: /usr/local/cuda-${{ matrix.cuda }}
+      CMAKE_VERSION: 3.29.6
     steps:
+      # Mandatory on hyperstack since a bootable volume is not re-usable yet.
+      - name: Install dependencies
+        run: |
+          sudo apt update
+          sudo apt install -y checkinstall zlib1g-dev libssl-dev libclang-dev
+          wget https://github.com/Kitware/CMake/releases/download/v${{ env.CMAKE_VERSION }}/cmake-${{ env.CMAKE_VERSION }}.tar.gz
+          tar -zxvf cmake-${{ env.CMAKE_VERSION }}.tar.gz
+          cd cmake-${{ env.CMAKE_VERSION }}
+          ./bootstrap
+          make -j"$(nproc)"
+          sudo make install
+
+
       - name: Checkout tfhe-rs repo with tags
         uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
         with:
           fetch-depth: 0
           token: ${{ secrets.FHE_ACTIONS_TOKEN }}
 
-      - name: Setup Hyperstack dependencies
-        uses: ./.github/actions/hyperstack_setup
-        with:
-          cuda-version: ${{ matrix.cuda }}
-          gcc-version: ${{ matrix.gcc }}
-
       - name: Get benchmark details
         run: |
           {
@@ -75,10 +85,31 @@ jobs:
           echo "HOME=/home/ubuntu" >> "${GITHUB_ENV}"
 
       - name: Install rust
-        uses: dtolnay/rust-toolchain@a54c7afa936fefeb4456b2dd8068152669aa8203
+        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 }}";
+            echo "HOME=/home/ubuntu";
+          } >> "${GITHUB_ENV}"
+
       - name: Run benchmarks with AVX512
         run: |
           make bench_pbs_gpu
@@ -95,10 +126,11 @@ jobs:
           --commit-date "${{ env.COMMIT_DATE }}" \
           --bench-date "${{ env.BENCH_DATE }}" \
           --name-suffix avx512 \
-          --walk-subdirs
+          --walk-subdirs \
+          --throughput
 
       - name: Upload parsed results artifact
-        uses: actions/upload-artifact@6f51ac03b9356f520e9adb1b1b7802705f340c2b
+        uses: actions/upload-artifact@b4b15b8c7c6ac21ea08fcf65892d2ee8f75cf882
         with:
           name: ${{ github.sha }}_core_crypto
           path: ${{ env.RESULTS_FILENAME }}
@@ -137,7 +169,7 @@ jobs:
     steps:
       - name: Stop instance
         id: stop-instance
-        uses: zama-ai/slab-github-runner@79939325c3c429837c10d6041e4fd8589d328bac
+        uses: zama-ai/slab-github-runner@801df0b8db5ea2b06128b7476c652f5ed5f193a8
         with:
           mode: stop
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}

.github/workflows/benchmark_gpu_erc20.yml

@@ -1,44 +1,195 @@
-# Run CUDA ERC20 benchmarks on a Hyperstack VM and return parsed results to Slab CI bot.
-name: Cuda ERC20 benchmarks
+# Run ERC20 benchmarks on an instance with CUDA and return parsed results to Slab CI bot.
+name: ERC20 GPU H100 benchmarks
 
 on:
   workflow_dispatch:
-    inputs:
-      profile:
-        description: "Instance type"
-        required: true
-        type: choice
-        options:
-          - "l40 (n3-L40x1)"
-          - "single-h100 (n3-H100x1)"
-          - "2-h100 (n3-H100x2)"
-          - "4-h100 (n3-H100x4)"
-          - "multi-h100 (n3-H100x8)"
-          - "multi-h100-nvlink (n3-H100x8-NVLink)"
-          - "multi-h100-sxm5 (n3-H100x8-SXM5)"
+  schedule:
+    # Weekly benchmarks will be triggered each Saturday at 5a.m.
+    - cron: '0 5 * * 6'
+
+env:
+  CARGO_TERM_COLOR: always
+  RESULTS_FILENAME: parsed_benchmark_results_${{ github.sha }}.json
+  PARSE_INTEGER_BENCH_CSV_FILE: tfhe_rs_integer_benches_${{ github.sha }}.csv
+  ACTION_RUN_URL: ${{ github.server_url }}/${{ github.repository }}/actions/runs/${{ github.run_id }}
+  RUST_BACKTRACE: "full"
+  RUST_MIN_STACK: "8388608"
+  SLACK_CHANNEL: ${{ secrets.SLACK_CHANNEL }}
+  SLACK_ICON: https://pbs.twimg.com/profile_images/1274014582265298945/OjBKP9kn_400x400.png
+  SLACK_USERNAME: ${{ secrets.BOT_USERNAME }}
+  SLACK_WEBHOOK: ${{ secrets.SLACK_WEBHOOK }}
 
 jobs:
-  parse-inputs:
+  setup-instance:
+    name: Setup instance (cuda-erc20-benchmarks)
     runs-on: ubuntu-latest
+    if:  github.event_name == 'workflow_dispatch' ||
+      (github.event_name == 'schedule' && github.repository == 'zama-ai/tfhe-rs')
     outputs:
-      profile: ${{ steps.parse_profile.outputs.profile }}
-      hardware_name: ${{ steps.parse_hardware_name.outputs.name }}
+      runner-name: ${{ steps.start-instance.outputs.label }}
     steps:
-      - name: Parse profile
-        id: parse_profile
-        run: |
-          echo "profile=$(echo '${{ inputs.profile }}' | sed 's|\(.*\)[[:space:]](.*)|\1|')" >> "${GITHUB_OUTPUT}"
+      - name: Start instance
+        id: start-instance
+        uses: zama-ai/slab-github-runner@801df0b8db5ea2b06128b7476c652f5ed5f193a8
+        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
 
-      - name: Parse hardware name
-        id: parse_hardware_name
+  cuda-erc20-benchmarks:
+    name: Execute GPU integer benchmarks
+    needs: setup-instance
+    runs-on: ${{ needs.setup-instance.outputs.runner-name }}
+    strategy:
+      fail-fast: false
+      # explicit include-based build matrix, of known valid options
+      matrix:
+        include:
+          - os: ubuntu-22.04
+            cuda: "12.2"
+            gcc: 11
+    env:
+      CUDA_PATH: /usr/local/cuda-${{ matrix.cuda }}
+      CMAKE_VERSION: 3.29.6
+    steps:
+      # Mandatory on hyperstack since a bootable volume is not re-usable yet.
+      - name: Install dependencies
         run: |
-          echo "name=$(echo '${{ inputs.profile }}' | sed 's|.*[[:space:]](\(.*\))|\1|')" >> "${GITHUB_OUTPUT}"
+          sudo apt update
+          sudo apt install -y checkinstall zlib1g-dev libssl-dev
+          wget https://github.com/Kitware/CMake/releases/download/v${{ 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
 
-  run-benchmarks:
-    name: Run benchmarks
-    needs: parse-inputs
-    uses: ./.github/workflows/benchmark_gpu_erc20_common.yml
-    with:
-      profile: ${{ needs.parse-inputs.outputs.profile }}
-      hardware_name: ${{ needs.parse-inputs.outputs.hardware_name }}
-    secrets: inherit
+      - name: Checkout tfhe-rs repo with tags
+        uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
+        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: Check device is detected
+        if: ${{ !cancelled() }}
+        run: nvidia-smi
+
+      - name: Run benchmarks
+        run: |
+          make bench_hlapi_erc20_gpu
+
+      - name: Parse results
+        run: |
+          python3 ./ci/benchmark_parser.py target/criterion ${{ env.RESULTS_FILENAME }} \
+          --database tfhe_rs \
+          --hardware "n3-H100x1" \
+          --backend gpu \
+          --project-version "${{ env.COMMIT_HASH }}" \
+          --branch ${{ github.ref_name }} \
+          --commit-date "${{ env.COMMIT_DATE }}" \
+          --bench-date "${{ env.BENCH_DATE }}" \
+          --walk-subdirs \
+          --name-suffix avx512
+
+      - name: Parse PBS counts
+        run: |
+          python3 ./ci/benchmark_parser.py tfhe/erc20_pbs_count.csv ${{ env.RESULTS_FILENAME }} \
+          --object-sizes \
+          --append-results
+
+      - name: Upload parsed results artifact
+        uses: actions/upload-artifact@b4b15b8c7c6ac21ea08fcf65892d2ee8f75cf882
+        with:
+          name: ${{ github.sha }}_erc20
+          path: ${{ env.RESULTS_FILENAME }}
+
+      - name: Checkout Slab repo
+        uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
+        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 }}"
+
+  slack-notify:
+    name: Slack Notification
+    needs: [ setup-instance, cuda-erc20-benchmarks ]
+    runs-on: ubuntu-latest
+    if: ${{ always() && needs.cuda-erc20-benchmarks.result != 'skipped' && failure() }}
+    continue-on-error: true
+    steps:
+      - name: Send message
+        uses: rtCamp/action-slack-notify@c33737706dea87cd7784c687dadc9adf1be59990
+        env:
+          SLACK_COLOR: ${{ needs.cuda-erc20-benchmarks.result }}
+          SLACK_MESSAGE: "Integer GPU benchmarks finished with status: ${{ needs.cuda-erc20-benchmarks.result }}. (${{ env.ACTION_RUN_URL }})"
+
+  teardown-instance:
+    name: Teardown instance (cuda-erc20-benchmarks)
+    if: ${{ always() && needs.setup-instance.result != 'skipped' }}
+    needs: [ setup-instance, cuda-erc20-benchmarks, slack-notify ]
+    runs-on: ubuntu-latest
+    steps:
+      - name: Stop instance
+        id: stop-instance
+        uses: zama-ai/slab-github-runner@801df0b8db5ea2b06128b7476c652f5ed5f193a8
+        with:
+          mode: stop
+          github-token: ${{ secrets.SLAB_ACTION_TOKEN }}
+          slab-url: ${{ secrets.SLAB_BASE_URL }}
+          job-secret: ${{ secrets.JOB_SECRET }}
+          label: ${{ needs.setup-instance.outputs.runner-name }}
+
+      - name: Slack Notification
+        if: ${{ failure() }}
+        continue-on-error: true
+        uses: rtCamp/action-slack-notify@c33737706dea87cd7784c687dadc9adf1be59990
+        env:
+          SLACK_COLOR: ${{ job.status }}
+          SLACK_MESSAGE: "Instance teardown (cuda-erc20-benchmarks) finished with status: ${{ job.status }}. (${{ env.ACTION_RUN_URL }})"

.github/workflows/benchmark_gpu_erc20_weekly.yml

@@ -1,35 +0,0 @@
-# Run CUDA ERC20 benchmarks on multiple Hyperstack VMs and return parsed results to Slab CI bot.
-name: Cuda ERC20 weekly benchmarks
-
-on:
-  schedule:
-    # Weekly benchmarks will be triggered each Saturday at 5a.m.
-    - cron: '0 5 * * 6'
-
-jobs:
-  run-benchmarks-1-h100:
-    name: Run benchmarks (1xH100)
-    if: github.repository == 'zama-ai/tfhe-rs'
-    uses: ./.github/workflows/benchmark_gpu_erc20_common.yml
-    with:
-      profile: single-h100
-      hardware_name: n3-H100x1
-    secrets: inherit
-
-  run-benchmarks-2-h100:
-    name: Run benchmarks (2xH100)
-    if: github.repository == 'zama-ai/tfhe-rs'
-    uses: ./.github/workflows/benchmark_gpu_erc20_common.yml
-    with:
-      profile: 2-h100
-      hardware_name: n3-H100x2
-    secrets: inherit
-
-  run-benchmarks-8-h100:
-    name: Run benchmarks (8xH100)
-    if: github.repository == 'zama-ai/tfhe-rs'
-    uses: ./.github/workflows/benchmark_gpu_erc20_common.yml
-    with:
-      profile: multi-h100
-      hardware_name: n3-H100x8
-    secrets: inherit

.github/workflows/benchmark_gpu_integer.yml

@@ -1,79 +1,201 @@
-# Run CUDA benchmarks on a Hyperstack VM and return parsed results to Slab CI bot.
-name: Cuda benchmarks
+# Run integer benchmarks on an instance with CUDA and return parsed results to Slab CI bot.
+name: Integer GPU benchmarks
 
 on:
   workflow_dispatch:
-    inputs:
-      profile:
-        description: "Instance type"
-        required: true
-        type: choice
-        options:
-          - "l40 (n3-L40x1)"
-          - "single-h100 (n3-H100x1)"
-          - "2-h100 (n3-H100x2)"
-          - "4-h100 (n3-H100x4)"
-          - "multi-h100 (n3-H100x8)"
-          - "multi-h100-nvlink (n3-H100x8-NVLink)"
-          - "multi-h100-sxm5 (n3-H100x8-SXM5)"
-          - "multi-a100-nvlink (n3-A100x8-NVLink)"
-      command:
-        description: "Benchmark command to run"
-        type: choice
-        default: integer_multi_bit
-        options:
-          - integer
-          - integer_multi_bit
-          - integer_compression
-          - pbs
-          - ks
-      op_flavor:
-        description: "Operations set to run"
-        type: choice
-        default: default
-        options:
-          - default
-          - fast_default
-          - unchecked
-      all_precisions:
-        description: "Run all precisions"
-        type: boolean
-        default: false
-      bench_type:
-        description: "Benchmarks type"
-        type: choice
-        default: latency
-        options:
-          - latency
-          - throughput
-          - both
+  push:
+    branches:
+      - main
+
+env:
+  CARGO_TERM_COLOR: always
+  RESULTS_FILENAME: parsed_benchmark_results_${{ github.sha }}.json
+  PARSE_INTEGER_BENCH_CSV_FILE: tfhe_rs_integer_benches_${{ github.sha }}.csv
+  ACTION_RUN_URL: ${{ github.server_url }}/${{ github.repository }}/actions/runs/${{ github.run_id }}
+  RUST_BACKTRACE: "full"
+  RUST_MIN_STACK: "8388608"
+  SLACK_CHANNEL: ${{ secrets.SLACK_CHANNEL }}
+  SLACK_ICON: https://pbs.twimg.com/profile_images/1274014582265298945/OjBKP9kn_400x400.png
+  SLACK_USERNAME: ${{ secrets.BOT_USERNAME }}
+  SLACK_WEBHOOK: ${{ secrets.SLACK_WEBHOOK }}
 
 jobs:
-  parse-inputs:
+  setup-instance:
+    name: Setup instance (cuda-integer-benchmarks)
     runs-on: ubuntu-latest
+    if:  github.event_name == 'workflow_dispatch' ||
+      (github.event_name == 'push' && github.repository == 'zama-ai/tfhe-rs')
     outputs:
-      profile: ${{ steps.parse_profile.outputs.profile }}
-      hardware_name: ${{ steps.parse_hardware_name.outputs.name }}
+      runner-name: ${{ steps.start-instance.outputs.label }}
     steps:
-      - name: Parse profile
-        id: parse_profile
-        run: |
-          echo "profile=$(echo '${{ inputs.profile }}' | sed 's|\(.*\)[[:space:]](.*)|\1|')" >> "${GITHUB_OUTPUT}"
+      - name: Start instance
+        id: start-instance
+        uses: zama-ai/slab-github-runner@801df0b8db5ea2b06128b7476c652f5ed5f193a8
+        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
 
-      - name: Parse hardware name
-        id: parse_hardware_name
+  cuda-integer-benchmarks:
+    name: Execute GPU integer benchmarks
+    needs: setup-instance
+    runs-on: ${{ needs.setup-instance.outputs.runner-name }}
+    strategy:
+      fail-fast: false
+      # explicit include-based build matrix, of known valid options
+      matrix:
+        include:
+          - os: ubuntu-22.04
+            cuda: "12.2"
+            gcc: 11
+    env:
+      CUDA_PATH: /usr/local/cuda-${{ matrix.cuda }}
+      CMAKE_VERSION: 3.29.6
+    steps:
+      # Mandatory on hyperstack since a bootable volume is not re-usable yet.
+      - name: Install dependencies
         run: |
-          echo "name=$(echo '${{ inputs.profile }}' | sed 's|.*[[:space:]](\(.*\))|\1|')" >> "${GITHUB_OUTPUT}"
+          sudo apt update
+          sudo apt install -y checkinstall zlib1g-dev libssl-dev libclang-dev
+          wget https://github.com/Kitware/CMake/releases/download/v${{ env.CMAKE_VERSION }}/cmake-${{ env.CMAKE_VERSION }}.tar.gz
+          tar -zxvf cmake-${{ env.CMAKE_VERSION }}.tar.gz
+          cd cmake-${{ env.CMAKE_VERSION }}
+          ./bootstrap
+          make -j"$(nproc)"
+          sudo make install
 
-  run-benchmarks:
-    name: Run benchmarks
-    needs: parse-inputs
-    uses: ./.github/workflows/benchmark_gpu_integer_common.yml
-    with:
-      profile: ${{ needs.parse-inputs.outputs.profile }}
-      hardware_name: ${{ needs.parse-inputs.outputs.hardware_name }}
-      command: ${{ inputs.command }}
-      op_flavor: ${{ inputs.op_flavor }}
-      bench_type: ${{ inputs.bench_type }}
-      all_precisions: ${{ inputs.all_precisions }}
-    secrets: inherit
+      - name: Checkout tfhe-rs repo with tags
+        uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
+        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: Check device is detected
+        if: ${{ !cancelled() }}
+        run: nvidia-smi
+
+      - name: Run benchmarks with AVX512
+        run: |
+          make FAST_BENCH=TRUE BENCH_OP_FLAVOR=default bench_integer_gpu
+
+      - name: Parse benchmarks to csv
+        run: |
+          make PARSE_INTEGER_BENCH_CSV_FILE=${{ env.PARSE_INTEGER_BENCH_CSV_FILE }} \
+            parse_integer_benches
+
+      - name: Upload csv results artifact
+        uses: actions/upload-artifact@b4b15b8c7c6ac21ea08fcf65892d2ee8f75cf882
+        with:
+          name: ${{ github.sha }}_csv_integer
+          path: ${{ env.PARSE_INTEGER_BENCH_CSV_FILE }}
+
+      - name: Parse results
+        run: |
+          python3 ./ci/benchmark_parser.py target/criterion ${{ env.RESULTS_FILENAME }} \
+          --database tfhe_rs \
+          --hardware "n3-H100x1" \
+          --backend gpu \
+          --project-version "${{ 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@b4b15b8c7c6ac21ea08fcf65892d2ee8f75cf882
+        with:
+          name: ${{ github.sha }}_integer
+          path: ${{ env.RESULTS_FILENAME }}
+
+      - name: Checkout Slab repo
+        uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
+        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 }}"
+
+  slack-notify:
+    name: Slack Notification
+    needs: [ setup-instance, cuda-integer-benchmarks ]
+    runs-on: ubuntu-latest
+    if: ${{ always() && needs.cuda-integer-benchmarks.result != 'skipped' && failure() }}
+    continue-on-error: true
+    steps:
+      - name: Send message
+        uses: rtCamp/action-slack-notify@c33737706dea87cd7784c687dadc9adf1be59990
+        env:
+          SLACK_COLOR: ${{ needs.cuda-integer-benchmarks.result }}
+          SLACK_MESSAGE: "Integer GPU benchmarks finished with status: ${{ needs.cuda-integer-benchmarks.result }}. (${{ env.ACTION_RUN_URL }})"
+
+  teardown-instance:
+    name: Teardown instance (cuda-integer-benchmarks)
+    if: ${{ always() && needs.setup-instance.result != 'skipped' }}
+    needs: [ setup-instance, cuda-integer-benchmarks, slack-notify ]
+    runs-on: ubuntu-latest
+    steps:
+      - name: Stop instance
+        id: stop-instance
+        uses: zama-ai/slab-github-runner@801df0b8db5ea2b06128b7476c652f5ed5f193a8
+        with:
+          mode: stop
+          github-token: ${{ secrets.SLAB_ACTION_TOKEN }}
+          slab-url: ${{ secrets.SLAB_BASE_URL }}
+          job-secret: ${{ secrets.JOB_SECRET }}
+          label: ${{ needs.setup-instance.outputs.runner-name }}
+
+      - name: Slack Notification
+        if: ${{ failure() }}
+        continue-on-error: true
+        uses: rtCamp/action-slack-notify@c33737706dea87cd7784c687dadc9adf1be59990
+        env:
+          SLACK_COLOR: ${{ job.status }}
+          SLACK_MESSAGE: "Instance teardown (cuda-integer-benchmarks) finished with status: ${{ job.status }}. (${{ env.ACTION_RUN_URL }})"

.github/workflows/benchmark_gpu_integer_2H100_full.yml

@@ -1,40 +1,15 @@
-# Run ERC20 benchmarks on an instance with CUDA and return parsed results to Slab CI bot.
-name: Cuda ERC20 benchmarks - common
+# Run integer benchmarks on an instance with CUDA and return parsed results to Slab CI bot.
+name: Integer 2xH100 benchmarks
 
 on:
-  workflow_call:
-    inputs:
-      backend:
-        type: string
-        default: hyperstack
-      profile:
-        type: string
-        required: true
-      hardware_name:
-        type: string
-        required: true
-    secrets:
-      FHE_ACTIONS_TOKEN:
-        required: true
-      SLAB_ACTION_TOKEN:
-        required: true
-      SLAB_BASE_URL:
-        required: true
-      SLAB_URL:
-        required: true
-      JOB_SECRET:
-        required: true
-      SLACK_CHANNEL:
-        required: true
-      BOT_USERNAME:
-        required: true
-      SLACK_WEBHOOK:
-        required: true
+  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_${{ github.sha }}.json
-  PARSE_INTEGER_BENCH_CSV_FILE: tfhe_rs_integer_benches_${{ github.sha }}.csv
   ACTION_RUN_URL: ${{ github.server_url }}/${{ github.repository }}/actions/runs/${{ github.run_id }}
   RUST_BACKTRACE: "full"
   RUST_MIN_STACK: "8388608"
@@ -45,49 +20,63 @@ env:
 
 jobs:
   setup-instance:
-    name: Setup instance (cuda-erc20-benchmarks)
+    name: Setup instance (cuda-integer-full-2-gpu-benchmarks)
     runs-on: ubuntu-latest
-    if:  github.event_name == 'workflow_dispatch' ||
+    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@79939325c3c429837c10d6041e4fd8589d328bac
+        uses: zama-ai/slab-github-runner@801df0b8db5ea2b06128b7476c652f5ed5f193a8
         with:
           mode: start
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}
           slab-url: ${{ secrets.SLAB_BASE_URL }}
           job-secret: ${{ secrets.JOB_SECRET }}
-          backend: ${{ inputs.backend }}
-          profile: ${{ inputs.profile }}
+          backend: hyperstack
+          profile: 2-h100
 
-  cuda-erc20-benchmarks:
-    name: Cuda ERC20 benchmarks (${{ inputs.profile }})
+  cuda-integer-full-2-gpu-benchmarks:
+    name: Execute 2xH100 integer benchmarks
     needs: setup-instance
     runs-on: ${{ needs.setup-instance.outputs.runner-name }}
+    timeout-minutes: 1440 # 24 hours
+    continue-on-error: true
     strategy:
       fail-fast: false
-      # explicit include-based build matrix, of known valid options
+      max-parallel: 1
       matrix:
+        command: [integer_multi_bit]
+        op_flavor: [default]
+        # explicit include-based build matrix, of known valid options
         include:
           - os: ubuntu-22.04
             cuda: "12.2"
             gcc: 11
+    env:
+      CUDA_PATH: /usr/local/cuda-${{ 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 libclang-dev
+          wget https://github.com/Kitware/CMake/releases/download/v${{ env.CMAKE_VERSION }}/cmake-${{ env.CMAKE_VERSION }}.tar.gz
+          tar -zxvf cmake-${{ env.CMAKE_VERSION }}.tar.gz
+          cd cmake-${{ env.CMAKE_VERSION }}
+          ./bootstrap
+          make -j"$(nproc)"
+          sudo make install
+
       - name: Checkout tfhe-rs repo with tags
         uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
         with:
           fetch-depth: 0
           token: ${{ secrets.FHE_ACTIONS_TOKEN }}
 
-      - name: Setup Hyperstack dependencies
-        uses: ./.github/actions/hyperstack_setup
-        with:
-          cuda-version: ${{ matrix.cuda }}
-          gcc-version: ${{ matrix.gcc }}
-
       - name: Get benchmark details
         run: |
           {
@@ -102,32 +91,29 @@ jobs:
           echo "HOME=/home/ubuntu" >> "${GITHUB_ENV}"
 
       - name: Install rust
-        uses: dtolnay/rust-toolchain@a54c7afa936fefeb4456b2dd8068152669aa8203
+        uses: dtolnay/rust-toolchain@7b1c307e0dcbda6122208f10795a713336a9b35a
         with:
           toolchain: nightly
 
-      - name: Run benchmarks
+      - name: Export CUDA variables
+        if: ${{ !cancelled() }}
         run: |
-          make bench_hlapi_erc20_gpu
+          {
+            echo "CUDA_PATH=$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}"
 
-      - name: Parse results
+      # Specify the correct host compilers
+      - name: Export gcc and g++ variables
+        if: ${{ !cancelled() }}
         run: |
-          python3 ./ci/benchmark_parser.py target/criterion ${{ env.RESULTS_FILENAME }} \
-          --database tfhe_rs \
-          --hardware "${{ inputs.hardware_name }}" \
-          --backend gpu \
-          --project-version "${{ env.COMMIT_HASH }}" \
-          --branch ${{ github.ref_name }} \
-          --commit-date "${{ env.COMMIT_DATE }}" \
-          --bench-date "${{ env.BENCH_DATE }}" \
-          --walk-subdirs \
-          --name-suffix avx512
-
-      - name: Upload parsed results artifact
-        uses: actions/upload-artifact@6f51ac03b9356f520e9adb1b1b7802705f340c2b
-        with:
-          name: ${{ github.sha }}_erc20_${{ inputs.profile }}
-          path: ${{ env.RESULTS_FILENAME }}
+          {
+            echo "CC=/usr/bin/gcc-${{ matrix.gcc }}";
+            echo "CXX=/usr/bin/g++-${{ matrix.gcc }}";
+            echo "CUDAHOSTCXX=/usr/bin/g++-${{ matrix.gcc }}";
+          } >> "${GITHUB_ENV}"
 
       - name: Checkout Slab repo
         uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
@@ -136,6 +122,34 @@ jobs:
           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@b4b15b8c7c6ac21ea08fcf65892d2ee8f75cf882
+        with:
+          name: ${{ github.sha }}_${{ matrix.command }}_${{ matrix.op_flavor }}
+          path: ${{ env.RESULTS_FILENAME }}
+
       - name: Send data to Slab
         shell: bash
         run: |
@@ -144,26 +158,26 @@ jobs:
 
   slack-notify:
     name: Slack Notification
-    needs: [ setup-instance, cuda-erc20-benchmarks ]
+    needs: [ setup-instance, cuda-integer-full-2-gpu-benchmarks ]
     runs-on: ubuntu-latest
-    if: ${{ always() && needs.cuda-erc20-benchmarks.result != 'skipped' && failure() }}
+    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@c33737706dea87cd7784c687dadc9adf1be59990
         env:
-          SLACK_COLOR: ${{ needs.cuda-erc20-benchmarks.result }}
-          SLACK_MESSAGE: "Cuda ERC20 benchmarks (${{ inputs.profile }}) finished with status: ${{ needs.cuda-erc20-benchmarks.result }}. (${{ env.ACTION_RUN_URL }})"
+          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-erc20-${{ inputs.profile }}-benchmarks)
+    name: Teardown instance (cuda-integer-full-2-gpu-benchmarks)
     if: ${{ always() && needs.setup-instance.result != 'skipped' }}
-    needs: [ setup-instance, cuda-erc20-benchmarks, slack-notify ]
+    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@79939325c3c429837c10d6041e4fd8589d328bac
+        uses: zama-ai/slab-github-runner@801df0b8db5ea2b06128b7476c652f5ed5f193a8
         with:
           mode: stop
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}
@@ -177,4 +191,4 @@ jobs:
         uses: rtCamp/action-slack-notify@c33737706dea87cd7784c687dadc9adf1be59990
         env:
           SLACK_COLOR: ${{ job.status }}
-          SLACK_MESSAGE: "Instance teardown (cuda-erc20-${{ inputs.profile }}-benchmarks) finished with status: ${{ job.status }}. (${{ env.ACTION_RUN_URL }})"
+          SLACK_MESSAGE: "Instance teardown (cuda-integer-full-2-gpu-benchmarks) finished with status: ${{ job.status }}. (${{ env.ACTION_RUN_URL }})"

.github/workflows/benchmark_gpu_integer_common.yml

@@ -1,256 +0,0 @@
-# Run integer benchmarks on CUDA instance and return parsed results to Slab CI bot.
-name: Cuda benchmarks - common
-
-on:
-  workflow_call:
-    inputs:
-      backend:
-        type: string
-        default: hyperstack
-      profile:
-        type: string
-        required: true
-      hardware_name:
-        type: string
-        required: true
-      command: # Use a comma separated values to generate an array
-        type: string
-        required: true
-      op_flavor: # Use a comma separated values to generate an array
-        type: string
-        required: true
-      bench_type:
-        type: string
-        default: latency
-      all_precisions:
-        type: boolean
-        default: false
-    secrets:
-      FHE_ACTIONS_TOKEN:
-        required: true
-      SLAB_ACTION_TOKEN:
-        required: true
-      SLAB_BASE_URL:
-        required: true
-      SLAB_URL:
-        required: true
-      JOB_SECRET:
-        required: true
-      SLACK_CHANNEL:
-        required: true
-      BOT_USERNAME:
-        required: true
-      SLACK_WEBHOOK:
-        required: true
-
-env:
-  CARGO_TERM_COLOR: always
-  RESULTS_FILENAME: parsed_benchmark_results_${{ github.sha }}.json
-  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
-    outputs:
-      command: ${{ steps.set_command.outputs.command }}
-      op_flavor: ${{ steps.set_op_flavor.outputs.op_flavor }}
-      bench_type: ${{ steps.set_bench_type.outputs.bench_type }}
-    steps:
-      - name: Set single command
-        if: ${{ !contains(inputs.command, ',')}}
-        run: |
-          echo "COMMAND=[\"${{ inputs.command }}\"]" >> "${GITHUB_ENV}"
-
-      - name: Set multiple commands
-        if: ${{ contains(inputs.command, ',')}}
-        run: |
-          PARSED_COMMAND=$(echo "${{ inputs.command }}" | sed 's/[[:space:]]*,[[:space:]]*/\\", \\"/g')
-          echo "COMMAND=[\"${PARSED_COMMAND}\"]" >> "${GITHUB_ENV}"
-
-      - name: Set single operations flavor
-        if: ${{ !contains(inputs.op_flavor, ',')}}
-        run: |
-          echo "OP_FLAVOR=[\"${{ inputs.op_flavor }}\"]" >> "${GITHUB_ENV}"
-
-      - name: Set multiple operations flavors
-        if: ${{ contains(inputs.op_flavor, ',')}}
-        run: |
-          PARSED_OP_FLAVOR=$(echo "${{ inputs.op_flavor }}" | sed 's/[[:space:]]*,[[:space:]]*/", "/g')
-          echo "OP_FLAVOR=[\"${PARSED_OP_FLAVOR}\"]" >> "${GITHUB_ENV}"
-
-      - name: Set benchmark types
-        run: |
-          if [[ "${{ inputs.bench_type }}" == "both" ]]; then
-            echo "BENCH_TYPE=[\"latency\", \"throughput\"]" >> "${GITHUB_ENV}"
-          else
-            echo "BENCH_TYPE=[\"${{ inputs.bench_type }}\"]" >> "${GITHUB_ENV}"
-          fi
-
-      - name: Set command output
-        id: set_command
-        run: |
-          echo "command=${{ toJSON(env.COMMAND) }}" >> "${GITHUB_OUTPUT}"
-
-      - name: Set operation flavor output
-        id: set_op_flavor
-        run: |
-          echo "op_flavor=${{ toJSON(env.OP_FLAVOR) }}" >> "${GITHUB_OUTPUT}"
-
-      - name: Set benchmark types output
-        id: set_bench_type
-        run: |
-          echo "bench_type=${{ toJSON(env.BENCH_TYPE) }}" >> "${GITHUB_OUTPUT}"
-
-  setup-instance:
-    name: Setup instance (cuda-${{ inputs.profile }}-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@79939325c3c429837c10d6041e4fd8589d328bac
-        with:
-          mode: start
-          github-token: ${{ secrets.SLAB_ACTION_TOKEN }}
-          slab-url: ${{ secrets.SLAB_BASE_URL }}
-          job-secret: ${{ secrets.JOB_SECRET }}
-          backend: ${{ inputs.backend }}
-          profile: ${{ inputs.profile }}
-
-  cuda-benchmarks:
-    name: Cuda benchmarks (${{ inputs.profile }})
-    needs: [ prepare-matrix, 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: ${{ fromJSON(needs.prepare-matrix.outputs.command) }}
-        op_flavor: ${{ fromJSON(needs.prepare-matrix.outputs.op_flavor) }}
-        bench_type: ${{ fromJSON(needs.prepare-matrix.outputs.bench_type) }}
-        # explicit include-based build matrix, of known valid options
-        include:
-          - os: ubuntu-22.04
-            cuda: "12.2"
-            gcc: 11
-    steps:
-      - name: Checkout tfhe-rs repo with tags
-        uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
-        with:
-          fetch-depth: 0
-          token: ${{ secrets.FHE_ACTIONS_TOKEN }}
-
-      - name: Setup Hyperstack dependencies
-        uses: ./.github/actions/hyperstack_setup
-        with:
-          cuda-version: ${{ matrix.cuda }}
-          gcc-version: ${{ matrix.gcc }}
-
-      - name: Get benchmark details
-        run: |
-          {
-            echo "BENCH_DATE=$(date --iso-8601=seconds)";
-            echo "COMMIT_DATE=$(git --no-pager show -s --format=%cd --date=iso8601-strict ${{ github.sha }})";
-            echo "COMMIT_HASH=$(git describe --tags --dirty)";
-          } >> "${GITHUB_ENV}"
-
-      - name: Set up home
-        # "Install rust" step require root user to have a HOME directory which is not set.
-        run: |
-          echo "HOME=/home/ubuntu" >> "${GITHUB_ENV}"
-
-      - name: Install rust
-        uses: dtolnay/rust-toolchain@a54c7afa936fefeb4456b2dd8068152669aa8203
-        with:
-          toolchain: nightly
-
-      - name: Should run benchmarks with all precisions
-        if: inputs.all_precisions
-        run: |
-          echo "FAST_BENCH=FALSE" >> "${GITHUB_ENV}"
-
-      - name: Run benchmarks
-        run: |
-          make BENCH_OP_FLAVOR=${{ matrix.op_flavor }} BENCH_TYPE=${{ matrix.bench_type }} bench_${{ matrix.command }}_gpu
-
-      - name: Parse results
-        run: |
-          python3 ./ci/benchmark_parser.py target/criterion ${{ env.RESULTS_FILENAME }} \
-          --database tfhe_rs \
-          --hardware "${{ inputs.hardware_name }}" \
-          --backend gpu \
-          --project-version "${{ env.COMMIT_HASH }}" \
-          --branch ${{ github.ref_name }} \
-          --commit-date "${{ env.COMMIT_DATE }}" \
-          --bench-date "${{ env.BENCH_DATE }}" \
-          --walk-subdirs \
-          --name-suffix avx512 \
-          --bench-type ${{ matrix.bench_type }}
-
-      - name: Upload parsed results artifact
-        uses: actions/upload-artifact@6f51ac03b9356f520e9adb1b1b7802705f340c2b
-        with:
-          name: ${{ github.sha }}_${{ matrix.command }}_${{ matrix.op_flavor }}_${{ inputs.profile }}
-          path: ${{ env.RESULTS_FILENAME }}
-
-      - name: Checkout Slab repo
-        uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
-        with:
-          repository: zama-ai/slab
-          path: slab
-          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 }}"
-
-  slack-notify:
-    name: Slack Notification
-    needs: [ setup-instance, cuda-benchmarks ]
-    runs-on: ubuntu-latest
-    if: ${{ always() && needs.cuda-benchmarks.result != 'skipped' && failure() }}
-    continue-on-error: true
-    steps:
-      - name: Send message
-        uses: rtCamp/action-slack-notify@c33737706dea87cd7784c687dadc9adf1be59990
-        env:
-          SLACK_COLOR: ${{ needs.cuda-benchmarks.result }}
-          SLACK_MESSAGE: "Cuda benchmarks (${{ inputs.profile }}) finished with status: ${{ needs.cuda-benchmarks.result }}. (${{ env.ACTION_RUN_URL }})"
-
-  teardown-instance:
-    name: Teardown instance (cuda-${{ inputs.profile }}-benchmarks)
-    if: ${{ always() && needs.setup-instance.result != 'skipped' }}
-    needs: [ setup-instance, cuda-benchmarks, slack-notify ]
-    runs-on: ubuntu-latest
-    steps:
-      - name: Stop instance
-        id: stop-instance
-        uses: zama-ai/slab-github-runner@79939325c3c429837c10d6041e4fd8589d328bac
-        with:
-          mode: stop
-          github-token: ${{ secrets.SLAB_ACTION_TOKEN }}
-          slab-url: ${{ secrets.SLAB_BASE_URL }}
-          job-secret: ${{ secrets.JOB_SECRET }}
-          label: ${{ needs.setup-instance.outputs.runner-name }}
-
-      - name: Slack Notification
-        if: ${{ failure() }}
-        continue-on-error: true
-        uses: rtCamp/action-slack-notify@c33737706dea87cd7784c687dadc9adf1be59990
-        env:
-          SLACK_COLOR: ${{ job.status }}
-          SLACK_MESSAGE: "Instance teardown (cuda-${{ inputs.profile }}-benchmarks) finished with status: ${{ job.status }}. (${{ env.ACTION_RUN_URL }})"

.github/workflows/benchmark_gpu_integer_full.yml

@@ -0,0 +1,200 @@
+# Run all integer benchmarks on an instance with CUDA and return parsed results to Slab CI bot.
+name: Integer GPU full 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-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@801df0b8db5ea2b06128b7476c652f5ed5f193a8
+        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-integer-full-benchmarks:
+    name: Execute GPU integer benchmarks for all operations flavor
+    needs: setup-instance
+    runs-on: ${{ needs.setup-instance.outputs.runner-name }}
+    timeout-minutes: 1440 # 24 hours
+    continue-on-error: true
+    strategy:
+      fail-fast: false
+      max-parallel: 1
+      matrix:
+        command: [integer, integer_multi_bit]
+        op_flavor: [default]
+        # explicit include-based build matrix, of known valid options
+        include:
+          - os: ubuntu-22.04
+            cuda: "12.2"
+            gcc: 11
+    env:
+      CUDA_PATH: /usr/local/cuda-${{ 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 libclang-dev
+          wget https://github.com/Kitware/CMake/releases/download/v${{ env.CMAKE_VERSION }}/cmake-${{ env.CMAKE_VERSION }}.tar.gz
+          tar -zxvf cmake-${{ env.CMAKE_VERSION }}.tar.gz
+          cd cmake-${{ env.CMAKE_VERSION }}
+          ./bootstrap
+          make -j"$(nproc)"
+          sudo make install
+
+      - name: Checkout tfhe-rs repo with tags
+        uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
+        with:
+          fetch-depth: 0
+          token: ${{ secrets.FHE_ACTIONS_TOKEN }}
+
+      - name: Get benchmark details
+        run: |
+          {
+            echo "BENCH_DATE=$(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@11bd71901bbe5b1630ceea73d27597364c9af683
+        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 ${{ env.RESULTS_FILENAME }} \
+          --database tfhe_rs \
+          --hardware "n3-H100x1" \
+          --backend gpu \
+          --project-version "${{ 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@b4b15b8c7c6ac21ea08fcf65892d2ee8f75cf882
+        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-benchmarks ]
+    runs-on: ubuntu-latest
+    if: ${{ always() && needs.cuda-integer-full-benchmarks.result != 'skipped' && failure() }}
+    continue-on-error: true
+    steps:
+      - name: Send message
+        uses: rtCamp/action-slack-notify@c33737706dea87cd7784c687dadc9adf1be59990
+        env:
+          SLACK_COLOR: ${{ needs.cuda-integer-full-benchmarks.result }}
+          SLACK_MESSAGE: "Integer GPU full benchmarks finished with status: ${{ needs.cuda-integer-full-benchmarks.result }}. (${{ env.ACTION_RUN_URL }})"
+
+  teardown-instance:
+    name: Teardown instance (cuda-integer-full-benchmarks)
+    if: ${{ always() && needs.setup-instance.result != 'skipped' }}
+    needs: [ setup-instance, cuda-integer-full-benchmarks, slack-notify ]
+    runs-on: ubuntu-latest
+    steps:
+      - name: Stop instance
+        id: stop-instance
+        uses: zama-ai/slab-github-runner@801df0b8db5ea2b06128b7476c652f5ed5f193a8
+        with:
+          mode: stop
+          github-token: ${{ secrets.SLAB_ACTION_TOKEN }}
+          slab-url: ${{ secrets.SLAB_BASE_URL }}
+          job-secret: ${{ secrets.JOB_SECRET }}
+          label: ${{ needs.setup-instance.outputs.runner-name }}
+
+      - name: Slack Notification
+        if: ${{ failure() }}
+        continue-on-error: true
+        uses: rtCamp/action-slack-notify@c33737706dea87cd7784c687dadc9adf1be59990
+        env:
+          SLACK_COLOR: ${{ job.status }}
+          SLACK_MESSAGE: "Instance teardown (cuda-integer-full-benchmarks) finished with status: ${{ job.status }}. (${{ env.ACTION_RUN_URL }})"

.github/workflows/benchmark_gpu_integer_multi_bit.yml

@@ -0,0 +1,224 @@
+# Run integer benchmarks with multi-bit cryptographic parameters on an instance and return parsed results to Slab CI bot.
+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'
+
+env:
+  CARGO_TERM_COLOR: always
+  RESULTS_FILENAME: parsed_benchmark_results_${{ github.sha }}.json
+  PARSE_INTEGER_BENCH_CSV_FILE: tfhe_rs_integer_benches_${{ github.sha }}.csv
+  ACTION_RUN_URL: ${{ github.server_url }}/${{ github.repository }}/actions/runs/${{ github.run_id }}
+  RUST_BACKTRACE: "full"
+  RUST_MIN_STACK: "8388608"
+  SLACK_CHANNEL: ${{ secrets.SLACK_CHANNEL }}
+  SLACK_ICON: https://pbs.twimg.com/profile_images/1274014582265298945/OjBKP9kn_400x400.png
+  SLACK_USERNAME: ${{ secrets.BOT_USERNAME }}
+  SLACK_WEBHOOK: ${{ secrets.SLACK_WEBHOOK }}
+  FAST_BENCH: TRUE
+  BENCH_OP_FLAVOR: default
+
+jobs:
+  setup-instance:
+    name: Setup instance (cuda-integer-multi-bit-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@801df0b8db5ea2b06128b7476c652f5ed5f193a8
+        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-integer-multi-bit-benchmarks:
+    name: Execute GPU integer multi-bit benchmarks
+    needs: setup-instance
+    runs-on: ${{ needs.setup-instance.outputs.runner-name }}
+    timeout-minutes: 1440 # 24 hours
+    strategy:
+      fail-fast: false
+      # explicit include-based build matrix, of known valid options
+      matrix:
+        include:
+          - os: ubuntu-22.04
+            cuda: "12.2"
+            gcc: 11
+    env:
+      CUDA_PATH: /usr/local/cuda-${{ 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 libclang-dev
+          wget https://github.com/Kitware/CMake/releases/download/v${{ env.CMAKE_VERSION }}/cmake-${{ env.CMAKE_VERSION }}.tar.gz
+          tar -zxvf cmake-${{ env.CMAKE_VERSION }}.tar.gz
+          cd cmake-${{ env.CMAKE_VERSION }}
+          ./bootstrap
+          make -j"$(nproc)"
+          sudo make install
+
+      - name: Checkout tfhe-rs repo with tags
+        uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
+        with:
+          fetch-depth: 0
+          token: ${{ secrets.FHE_ACTIONS_TOKEN }}
+
+      - name: Get benchmark details
+        run: |
+          {
+            echo "BENCH_DATE=$(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 }}";
+          } >> "${GITHUB_ENV}"
+
+      - name: Should run benchmarks with all precisions
+        if: inputs.all_precisions
+        run: |
+          echo "FAST_BENCH=FALSE" >> "${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: ${{ !cancelled() }}
+        run: nvidia-smi
+
+      - name: Run multi-bit benchmarks with AVX512
+        run: |
+          make bench_unsigned_integer_multi_bit_gpu
+
+      - name: Parse benchmarks to csv
+        run: |
+          make PARSE_INTEGER_BENCH_CSV_FILE=${{ env.PARSE_INTEGER_BENCH_CSV_FILE }} \
+            parse_integer_benches
+
+      - name: Upload csv results artifact
+        uses: actions/upload-artifact@b4b15b8c7c6ac21ea08fcf65892d2ee8f75cf882
+        with:
+          name: ${{ github.sha }}_csv_integer
+          path: ${{ env.PARSE_INTEGER_BENCH_CSV_FILE }}
+
+      - name: Parse results
+        run: |
+          python3 ./ci/benchmark_parser.py target/criterion ${{ env.RESULTS_FILENAME }} \
+          --database tfhe_rs \
+          --hardware "n3-H100x1" \
+          --backend gpu \
+          --project-version "${{ 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@b4b15b8c7c6ac21ea08fcf65892d2ee8f75cf882
+        with:
+          name: ${{ github.sha }}_integer
+          path: ${{ env.RESULTS_FILENAME }}
+
+      - name: Checkout Slab repo
+        uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
+        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 }}"
+
+  slack-notify:
+    name: Slack Notification
+    needs: [ setup-instance, cuda-integer-multi-bit-benchmarks ]
+    runs-on: ubuntu-latest
+    if: ${{ always() && needs.cuda-integer-multi-bit-benchmarks.result != 'skipped' && failure() }}
+    continue-on-error: true
+    steps:
+      - name: Send message
+        uses: rtCamp/action-slack-notify@c33737706dea87cd7784c687dadc9adf1be59990
+        env:
+          SLACK_COLOR: ${{ needs.cuda-integer-multi-bit-benchmarks.result }}
+          SLACK_MESSAGE: "Integer GPU multi-bit benchmarks finished with status: ${{ needs.cuda-integer-multi-bit-benchmarks.result }}. (${{ env.ACTION_RUN_URL }})"
+
+  teardown-instance:
+    name: Teardown instance (cuda-integer-full-benchmarks)
+    if: ${{ always() && needs.setup-instance.result != 'skipped' }}
+    needs: [ setup-instance, cuda-integer-multi-bit-benchmarks, slack-notify ]
+    runs-on: ubuntu-latest
+    steps:
+      - name: Stop instance
+        id: stop-instance
+        uses: zama-ai/slab-github-runner@801df0b8db5ea2b06128b7476c652f5ed5f193a8
+        with:
+          mode: stop
+          github-token: ${{ secrets.SLAB_ACTION_TOKEN }}
+          slab-url: ${{ secrets.SLAB_BASE_URL }}
+          job-secret: ${{ secrets.JOB_SECRET }}
+          label: ${{ needs.setup-instance.outputs.runner-name }}
+
+      - name: Slack Notification
+        if: ${{ failure() }}
+        continue-on-error: true
+        uses: rtCamp/action-slack-notify@c33737706dea87cd7784c687dadc9adf1be59990
+        env:
+          SLACK_COLOR: ${{ job.status }}
+          SLACK_MESSAGE: "Instance teardown (cuda-integer-multi-bit-benchmarks) finished with status: ${{ job.status }}. (${{ env.ACTION_RUN_URL }})"

.github/workflows/benchmark_gpu_integer_multi_bit_multi_gpu.yml

@@ -0,0 +1,214 @@
+# Run 64-bit multi-bit integer benchmarks on an instance with CUDA and return parsed results to Slab CI bot.
+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'
+
+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
+  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 == '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@801df0b8db5ea2b06128b7476c652f5ed5f193a8
+        with:
+          mode: start
+          github-token: ${{ secrets.SLAB_ACTION_TOKEN }}
+          slab-url: ${{ secrets.SLAB_BASE_URL }}
+          job-secret: ${{ secrets.JOB_SECRET }}
+          backend: hyperstack
+          profile: multi-h100
+
+  cuda-integer-multi-bit-multi-gpu-benchmarks:
+    name: Execute multi GPU integer multi-bit benchmarks
+    needs: setup-instance
+    runs-on: ${{ needs.setup-instance.outputs.runner-name }}
+    timeout-minutes: 1440 # 24 hours
+    continue-on-error: true
+    strategy:
+      fail-fast: false
+      max-parallel: 1
+      matrix:
+        include:
+          - os: ubuntu-22.04
+            cuda: "12.2"
+            gcc: 11
+    env:
+      CUDA_PATH: /usr/local/cuda-${{ 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 libclang-dev
+          wget https://github.com/Kitware/CMake/releases/download/v${{ env.CMAKE_VERSION }}/cmake-${{ env.CMAKE_VERSION }}.tar.gz
+          tar -zxvf cmake-${{ env.CMAKE_VERSION }}.tar.gz
+          cd cmake-${{ env.CMAKE_VERSION }}
+          ./bootstrap
+          make -j"$(nproc)"
+          sudo make install
+
+      - name: Checkout tfhe-rs repo with tags
+        uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
+        with:
+          fetch-depth: 0
+          token: ${{ secrets.FHE_ACTIONS_TOKEN }}
+
+      - name: Get benchmark details
+        run: |
+          {
+            echo "BENCH_DATE=$(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@11bd71901bbe5b1630ceea73d27597364c9af683
+        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 bench_unsigned_integer_multi_bit_gpu
+
+      - name: Parse results
+        run: |
+          python3 ./ci/benchmark_parser.py target/criterion ${{ env.RESULTS_FILENAME }} \
+          --database tfhe_rs \
+          --hardware "n3-H100x8" \
+          --backend gpu \
+          --project-version "${{ 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@b4b15b8c7c6ac21ea08fcf65892d2ee8f75cf882
+        with:
+          name: ${{ github.sha }}_integer
+          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-multi-bit-multi-gpu-benchmarks ]
+    runs-on: ubuntu-latest
+    if: ${{ always() && needs.cuda-integer-multi-bit-multi-gpu-benchmarks.result != 'skipped' && failure() }}
+    continue-on-error: true
+    steps:
+      - name: Send message
+        uses: rtCamp/action-slack-notify@c33737706dea87cd7784c687dadc9adf1be59990
+        env:
+          SLACK_COLOR: ${{ needs.cuda-integer-multi-bit-multi-gpu-benchmarks.result }}
+          SLACK_MESSAGE: "Integer multi GPU multi-bit benchmarks finished with status: ${{ needs.cuda-integer-multi-bit-multi-gpu-benchmarks.result }}. (${{ env.ACTION_RUN_URL }})"
+
+  teardown-instance:
+    name: Teardown instance (cuda-integer-multi-bit-multi-gpu-benchmarks)
+    if: ${{ always() && needs.setup-instance.result != 'skipped' }}
+    needs: [ setup-instance, cuda-integer-multi-bit-multi-gpu-benchmarks ]
+    runs-on: ubuntu-latest
+    steps:
+      - name: Stop instance
+        id: stop-instance
+        uses: zama-ai/slab-github-runner@801df0b8db5ea2b06128b7476c652f5ed5f193a8
+        with:
+          mode: stop
+          github-token: ${{ secrets.SLAB_ACTION_TOKEN }}
+          slab-url: ${{ secrets.SLAB_BASE_URL }}
+          job-secret: ${{ secrets.JOB_SECRET }}
+          label: ${{ needs.setup-instance.outputs.runner-name }}
+
+      - name: Slack Notification
+        if: ${{ failure() }}
+        continue-on-error: true
+        uses: rtCamp/action-slack-notify@c33737706dea87cd7784c687dadc9adf1be59990
+        env:
+          SLACK_COLOR: ${{ job.status }}
+          SLACK_MESSAGE: "Instance teardown (cuda-integer-multi-bit-multi-gpu-benchmarks) finished with status: ${{ job.status }}. (${{ env.ACTION_RUN_URL }})"

.github/workflows/benchmark_gpu_integer_multi_gpu_full.yml

@@ -0,0 +1,194 @@
+# Run all integer benchmarks on an instance with CUDA and return parsed results to Slab CI bot.
+name: Integer multi GPU full 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-multi-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@801df0b8db5ea2b06128b7476c652f5ed5f193a8
+        with:
+          mode: start
+          github-token: ${{ secrets.SLAB_ACTION_TOKEN }}
+          slab-url: ${{ secrets.SLAB_BASE_URL }}
+          job-secret: ${{ secrets.JOB_SECRET }}
+          backend: hyperstack
+          profile: multi-h100
+
+  cuda-integer-full-multi-gpu-benchmarks:
+    name: Execute multi GPU 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 libclang-dev
+          wget https://github.com/Kitware/CMake/releases/download/v${{ env.CMAKE_VERSION }}/cmake-${{ env.CMAKE_VERSION }}.tar.gz
+          tar -zxvf cmake-${{ env.CMAKE_VERSION }}.tar.gz
+          cd cmake-${{ env.CMAKE_VERSION }}
+          ./bootstrap
+          make -j"$(nproc)"
+          sudo make install
+
+      - name: Checkout tfhe-rs repo with tags
+        uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
+        with:
+          fetch-depth: 0
+          token: ${{ secrets.FHE_ACTIONS_TOKEN }}
+
+      - name: Get benchmark details
+        run: |
+          {
+            echo "BENCH_DATE=$(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@11bd71901bbe5b1630ceea73d27597364c9af683
+        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 ${{ env.RESULTS_FILENAME }} \
+          --database tfhe_rs \
+          --hardware "n3-H100x8" \
+          --backend gpu \
+          --project-version "${{ 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@b4b15b8c7c6ac21ea08fcf65892d2ee8f75cf882
+        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-multi-gpu-benchmarks ]
+    runs-on: ubuntu-latest
+    if: ${{ always() && needs.cuda-integer-full-multi-gpu-benchmarks.result != 'skipped' && failure() }}
+    continue-on-error: true
+    steps:
+      - name: Send message
+        uses: rtCamp/action-slack-notify@c33737706dea87cd7784c687dadc9adf1be59990
+        env:
+          SLACK_COLOR: ${{ needs.cuda-integer-full-multi-gpu-benchmarks.result }}
+          SLACK_MESSAGE: "Integer GPU full benchmarks finished with status: ${{ needs.cuda-integer-full-multi-gpu-benchmarks.result }}. (${{ env.ACTION_RUN_URL }})"
+
+  teardown-instance:
+    name: Teardown instance (cuda-integer-full-multi-gpu-benchmarks)
+    if: ${{ always() && needs.setup-instance.result != 'skipped' }}
+    needs: [ setup-instance, cuda-integer-full-multi-gpu-benchmarks ]
+    runs-on: ubuntu-latest
+    steps:
+      - name: Stop instance
+        id: stop-instance
+        uses: zama-ai/slab-github-runner@801df0b8db5ea2b06128b7476c652f5ed5f193a8
+        with:
+          mode: stop
+          github-token: ${{ secrets.SLAB_ACTION_TOKEN }}
+          slab-url: ${{ secrets.SLAB_BASE_URL }}
+          job-secret: ${{ secrets.JOB_SECRET }}
+          label: ${{ needs.setup-instance.outputs.runner-name }}
+
+      - name: Slack Notification
+        if: ${{ failure() }}
+        continue-on-error: true
+        uses: rtCamp/action-slack-notify@c33737706dea87cd7784c687dadc9adf1be59990
+        env:
+          SLACK_COLOR: ${{ job.status }}
+          SLACK_MESSAGE: "Instance teardown (cuda-integer-full-multi-gpu-benchmarks) finished with status: ${{ job.status }}. (${{ env.ACTION_RUN_URL }})"

.github/workflows/benchmark_gpu_integer_weekly.yml

@@ -1,60 +0,0 @@
-# Run CUDA benchmarks on multiple Hyperstack VMs and return parsed results to Slab CI bot.
-name: Cuda weekly benchmarks
-
-on:
-  schedule:
-    # Weekly benchmarks will be triggered each Saturday at 1a.m.
-    - cron: '0 1 * * 6'
-
-jobs:
-  run-benchmarks-1-h100:
-    name: Run benchmarks (1xH100)
-    if: github.repository == 'zama-ai/tfhe-rs'
-    uses: ./.github/workflows/benchmark_gpu_integer_common.yml
-    with:
-      profile: single-h100
-      hardware_name: n3-H100x1
-      command: integer,integer_multi_bit
-      op_flavor: default
-      bench_type: latency
-      all_precisions: true
-    secrets: inherit
-
-  run-benchmarks-2-h100:
-    name: Run benchmarks (2xH100)
-    if: github.repository == 'zama-ai/tfhe-rs'
-    uses: ./.github/workflows/benchmark_gpu_integer_common.yml
-    with:
-      profile: 2-h100
-      hardware_name: n3-H100x2
-      command: integer_multi_bit
-      op_flavor: default
-      bench_type: latency
-      all_precisions: true
-    secrets: inherit
-
-  run-benchmarks-8-h100:
-    name: Run benchmarks (8xH100)
-    if: github.repository == 'zama-ai/tfhe-rs'
-    uses: ./.github/workflows/benchmark_gpu_integer_common.yml
-    with:
-      profile: multi-h100
-      hardware_name: n3-H100x8
-      command: integer_multi_bit
-      op_flavor: default
-      bench_type: latency
-      all_precisions: true
-    secrets: inherit
-
-  run-benchmarks-l40:
-    name: Run benchmarks (L40)
-    if: github.repository == 'zama-ai/tfhe-rs'
-    uses: ./.github/workflows/benchmark_gpu_integer_common.yml
-    with:
-      profile: l40
-      hardware_name: n3-L40x1
-      command: integer_multi_bit,integer_compression,pbs,ks
-      op_flavor: default
-      bench_type: latency
-      all_precisions: true
-    secrets: inherit

.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@801df0b8db5ea2b06128b7476c652f5ed5f193a8
+        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 libclang-dev
+          wget https://github.com/Kitware/CMake/releases/download/v${{ env.CMAKE_VERSION }}/cmake-${{ env.CMAKE_VERSION }}.tar.gz
+          tar -zxvf cmake-${{ env.CMAKE_VERSION }}.tar.gz
+          cd cmake-${{ env.CMAKE_VERSION }}
+          ./bootstrap
+          make -j"$(nproc)"
+          sudo make install
+
+      - name: Checkout tfhe-rs repo with tags
+        uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
+        with:
+          fetch-depth: 0
+          token: ${{ secrets.FHE_ACTIONS_TOKEN }}
+
+      - name: Get benchmark details
+        run: |
+          {
+            echo "BENCH_DATE=$(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@11bd71901bbe5b1630ceea73d27597364c9af683
+        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@b4b15b8c7c6ac21ea08fcf65892d2ee8f75cf882
+        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@c33737706dea87cd7784c687dadc9adf1be59990
+        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@801df0b8db5ea2b06128b7476c652f5ed5f193a8
+        with:
+          mode: stop
+          github-token: ${{ secrets.SLAB_ACTION_TOKEN }}
+          slab-url: ${{ secrets.SLAB_BASE_URL }}
+          job-secret: ${{ secrets.JOB_SECRET }}
+          label: ${{ needs.setup-instance.outputs.runner-name }}
+
+      - name: Slack Notification
+        if: ${{ failure() }}
+        continue-on-error: true
+        uses: rtCamp/action-slack-notify@c33737706dea87cd7784c687dadc9adf1be59990
+        env:
+          SLACK_COLOR: ${{ job.status }}
+          SLACK_MESSAGE: "Instance teardown (cuda-l40-benchmarks) finished with status: ${{ job.status }}. (${{ env.ACTION_RUN_URL }})"

.github/workflows/benchmark_integer.yml

@@ -8,14 +8,6 @@ on:
         description: "Run all precisions"
         type: boolean
         default: false
-      bench_type:
-        description: "Benchmarks type"
-        type: choice
-        default: latency
-        options:
-          - latency
-          - throughput
-          - both
 
   schedule:
     # Weekly benchmarks will be triggered each Saturday at 1a.m.
@@ -44,10 +36,10 @@ jobs:
       (github.event_name == 'schedule' && github.repository == 'zama-ai/tfhe-rs')
     outputs:
       op_flavor: ${{ steps.set_op_flavor.outputs.op_flavor }}
-      bench_type: ${{ steps.set_bench_type.outputs.bench_type }}
     steps:
       - name: Weekly benchmarks
-        if: github.event.schedule == '0 1 * * 6'
+        if: github.event_name == 'workflow_dispatch' ||
+          github.event.schedule == '0 1 * * 6'
         run: |
           echo "OP_FLAVOR=[\"default\"]" >> "${GITHUB_ENV}"
 
@@ -56,31 +48,11 @@ jobs:
         run: |
           echo "OP_FLAVOR=[\"default\", \"smart\", \"unchecked\", \"misc\"]" >> "${GITHUB_ENV}"
 
-      - name: Set benchmark types
-        if: github.event_name == 'workflow_dispatch'
-        run: |
-          echo "OP_FLAVOR=[\"default\"]" >> "${GITHUB_ENV}"
-          if [[ "${{ inputs.bench_type }}" == "both" ]]; then
-            echo "BENCH_TYPE=[\"latency\", \"throughput\"]" >> "${GITHUB_ENV}"
-          else
-            echo "BENCH_TYPE=[\"${{ inputs.bench_type }}\"]" >> "${GITHUB_ENV}"
-          fi
-
-      - name: Default benchmark type
-        if: github.event_name != 'workflow_dispatch'
-        run: |
-          echo "BENCH_TYPE=[\"latency\"]" >> "${GITHUB_ENV}"
-
-      - name: Set operation flavor output
-        id: set_op_flavor
-        run: |
+      -  name: Set operation flavor output
+         id: set_op_flavor
+         run: |
           echo "op_flavor=${{ toJSON(env.OP_FLAVOR) }}" >> "${GITHUB_OUTPUT}"
 
-      - name: Set benchmark types output
-        id: set_bench_type
-        run: |
-          echo "bench_type=${{ toJSON(env.BENCH_TYPE) }}" >> "${GITHUB_OUTPUT}"
-
   setup-instance:
     name: Setup instance (integer-benchmarks)
     needs: prepare-matrix
@@ -90,7 +62,7 @@ jobs:
     steps:
       - name: Start instance
         id: start-instance
-        uses: zama-ai/slab-github-runner@79939325c3c429837c10d6041e4fd8589d328bac
+        uses: zama-ai/slab-github-runner@801df0b8db5ea2b06128b7476c652f5ed5f193a8
         with:
           mode: start
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}
@@ -113,7 +85,6 @@ jobs:
       matrix:
         command: [ integer, integer_multi_bit]
         op_flavor: ${{ fromJson(needs.prepare-matrix.outputs.op_flavor) }}
-        bench_type: ${{ fromJSON(needs.prepare-matrix.outputs.bench_type) }}
     steps:
       - name: Checkout tfhe-rs repo with tags
         uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
@@ -129,8 +100,13 @@ jobs:
             echo "COMMIT_HASH=$(git describe --tags --dirty)";
           } >> "${GITHUB_ENV}"
 
+      - name: Set up home
+        # "Install rust" step require root user to have a HOME directory which is not set.
+        run: |
+          echo "HOME=/home/ubuntu" >> "${GITHUB_ENV}"
+
       - name: Install rust
-        uses: dtolnay/rust-toolchain@a54c7afa936fefeb4456b2dd8068152669aa8203
+        uses: dtolnay/rust-toolchain@7b1c307e0dcbda6122208f10795a713336a9b35a
         with:
           toolchain: nightly
 
@@ -148,13 +124,13 @@ jobs:
 
       - name: Run benchmarks with AVX512
         run: |
-          make BENCH_OP_FLAVOR=${{ matrix.op_flavor }} BENCH_TYPE=${{ matrix.bench_type }} bench_${{ matrix.command }}
+          make BENCH_OP_FLAVOR=${{ matrix.op_flavor }} bench_${{ matrix.command }}
 
-      # Run these benchmarks only once per benchmark type
+      # Run these benchmarks only once
       - name: Run compression benchmarks with AVX512
         if: matrix.op_flavor == 'default' && matrix.command == 'integer'
         run: |
-          make BENCH_TYPE=${{ matrix.bench_type }} bench_integer_compression
+          make bench_integer_compression
 
       - name: Parse results
         run: |
@@ -167,12 +143,12 @@ jobs:
           --bench-date "${{ env.BENCH_DATE }}" \
           --walk-subdirs \
           --name-suffix avx512 \
-          --bench-type ${{ matrix.bench_type }}
+          --throughput
 
       - name: Upload parsed results artifact
-        uses: actions/upload-artifact@6f51ac03b9356f520e9adb1b1b7802705f340c2b
+        uses: actions/upload-artifact@b4b15b8c7c6ac21ea08fcf65892d2ee8f75cf882
         with:
-          name: ${{ github.sha }}_${{ matrix.command }}_${{ matrix.op_flavor }}_${{ matrix.bench_type }}
+          name: ${{ github.sha }}_${{ matrix.command }}_${{ matrix.op_flavor }}
           path: ${{ env.RESULTS_FILENAME }}
 
       - name: Send data to Slab
@@ -197,7 +173,7 @@ jobs:
     steps:
       - name: Stop instance
         id: stop-instance
-        uses: zama-ai/slab-github-runner@79939325c3c429837c10d6041e4fd8589d328bac
+        uses: zama-ai/slab-github-runner@801df0b8db5ea2b06128b7476c652f5ed5f193a8
         with:
           mode: stop
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}

.github/workflows/benchmark_shortint.yml

@@ -56,7 +56,7 @@ jobs:
     steps:
       - name: Start instance
         id: start-instance
-        uses: zama-ai/slab-github-runner@79939325c3c429837c10d6041e4fd8589d328bac
+        uses: zama-ai/slab-github-runner@801df0b8db5ea2b06128b7476c652f5ed5f193a8
         with:
           mode: start
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}
@@ -92,8 +92,13 @@ jobs:
             echo "COMMIT_HASH=$(git describe --tags --dirty)";
           } >> "${GITHUB_ENV}"
 
+      - name: Set up home
+        # "Install rust" step require root user to have a HOME directory which is not set.
+        run: |
+          echo "HOME=/home/ubuntu" >> "${GITHUB_ENV}"
+
       - name: Install rust
-        uses: dtolnay/rust-toolchain@a54c7afa936fefeb4456b2dd8068152669aa8203
+        uses: dtolnay/rust-toolchain@7b1c307e0dcbda6122208f10795a713336a9b35a
         with:
           toolchain: nightly
 
@@ -120,7 +125,8 @@ jobs:
           --commit-date "${COMMIT_DATE}" \
           --bench-date "${{ env.BENCH_DATE }}" \
           --walk-subdirs \
-          --name-suffix avx512
+          --name-suffix avx512 \
+          --throughput
 
       # This small benchmark needs to be executed only once.
       - name: Measure key sizes
@@ -136,7 +142,7 @@ jobs:
           --append-results
 
       - name: Upload parsed results artifact
-        uses: actions/upload-artifact@6f51ac03b9356f520e9adb1b1b7802705f340c2b
+        uses: actions/upload-artifact@b4b15b8c7c6ac21ea08fcf65892d2ee8f75cf882
         with:
           name: ${{ github.sha }}_shortint_${{ matrix.op_flavor }}
           path: ${{ env.RESULTS_FILENAME }}
@@ -163,7 +169,7 @@ jobs:
     steps:
       - name: Stop instance
         id: stop-instance
-        uses: zama-ai/slab-github-runner@79939325c3c429837c10d6041e4fd8589d328bac
+        uses: zama-ai/slab-github-runner@801df0b8db5ea2b06128b7476c652f5ed5f193a8
         with:
           mode: stop
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}

.github/workflows/benchmark_signed_integer.yml

@@ -8,14 +8,6 @@ on:
         description: "Run all precisions"
         type: boolean
         default: false
-      bench_type:
-        description: "Benchmarks type"
-        type: choice
-        default: latency
-        options:
-          - latency
-          - throughput
-          - both
 
   schedule:
     # Weekly benchmarks will be triggered each Saturday at 1a.m.
@@ -44,10 +36,10 @@ jobs:
       (github.event_name == 'schedule' && github.repository == 'zama-ai/tfhe-rs')
     outputs:
       op_flavor: ${{ steps.set_op_flavor.outputs.op_flavor }}
-      bench_type: ${{ steps.set_bench_type.outputs.bench_type }}
     steps:
       - name: Weekly benchmarks
-        if: github.event.schedule == '0 1 * * 6'
+        if: github.event_name == 'workflow_dispatch' ||
+          github.event.schedule == '0 1 * * 6'
         run: |
           echo "OP_FLAVOR=[\"default\"]" >> "${GITHUB_ENV}"
 
@@ -56,31 +48,11 @@ jobs:
         run: |
           echo "OP_FLAVOR=[\"default\", \"unchecked\"]" >> "${GITHUB_ENV}"
 
-      - name: Set benchmark types
-        if: github.event_name == 'workflow_dispatch'
-        run: |
-          echo "OP_FLAVOR=[\"default\"]" >> "${GITHUB_ENV}"
-          if [[ "${{ inputs.bench_type }}" == "both" ]]; then
-            echo "BENCH_TYPE=[\"latency\", \"throughput\"]" >> "${GITHUB_ENV}"
-          else
-            echo "BENCH_TYPE=[\"${{ inputs.bench_type }}\"]" >> "${GITHUB_ENV}"
-          fi
-
-      - name: Default benchmark type
-        if: github.event_name != 'workflow_dispatch'
-        run: |
-          echo "BENCH_TYPE=[\"latency\"]" >> "${GITHUB_ENV}"
-
       - name: Set operation flavor output
         id: set_op_flavor
         run: |
           echo "op_flavor=${{ toJSON(env.OP_FLAVOR) }}" >> "${GITHUB_OUTPUT}"
 
-      - name: Set benchmark types output
-        id: set_bench_type
-        run: |
-          echo "bench_type=${{ toJSON(env.BENCH_TYPE) }}" >> "${GITHUB_OUTPUT}"
-
   setup-instance:
     name: Setup instance (signed-integer-benchmarks)
     needs: prepare-matrix
@@ -90,7 +62,7 @@ jobs:
     steps:
       - name: Start instance
         id: start-instance
-        uses: zama-ai/slab-github-runner@79939325c3c429837c10d6041e4fd8589d328bac
+        uses: zama-ai/slab-github-runner@801df0b8db5ea2b06128b7476c652f5ed5f193a8
         with:
           mode: start
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}
@@ -112,8 +84,7 @@ jobs:
       max-parallel: 1
       matrix:
         command: [ integer, integer_multi_bit ]
-        op_flavor: ${{ fromJSON(needs.prepare-matrix.outputs.op_flavor) }}
-        bench_type: ${{ fromJSON(needs.prepare-matrix.outputs.bench_type) }}
+        op_flavor: [ default, unchecked ]
     steps:
       - name: Checkout tfhe-rs repo with tags
         uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
@@ -129,8 +100,13 @@ jobs:
             echo "COMMIT_HASH=$(git describe --tags --dirty)";
           } >> "${GITHUB_ENV}"
 
+      - name: Set up home
+        # "Install rust" step require root user to have a HOME directory which is not set.
+        run: |
+          echo "HOME=/home/ubuntu" >> "${GITHUB_ENV}"
+
       - name: Install rust
-        uses: dtolnay/rust-toolchain@a54c7afa936fefeb4456b2dd8068152669aa8203
+        uses: dtolnay/rust-toolchain@7b1c307e0dcbda6122208f10795a713336a9b35a
         with:
           toolchain: nightly
 
@@ -148,7 +124,7 @@ jobs:
 
       - name: Run benchmarks with AVX512
         run: |
-          make BENCH_OP_FLAVOR=${{ matrix.op_flavor }} BENCH_TYPE=${{ matrix.bench_type }} bench_signed_${{ matrix.command }}
+          make BENCH_OP_FLAVOR=${{ matrix.op_flavor }} bench_signed_${{ matrix.command }}
 
       - name: Parse results
         run: |
@@ -161,12 +137,12 @@ jobs:
           --bench-date "${{ env.BENCH_DATE }}" \
           --walk-subdirs \
           --name-suffix avx512 \
-          --bench-type ${{ matrix.bench_type }}
+          --throughput
 
       - name: Upload parsed results artifact
-        uses: actions/upload-artifact@6f51ac03b9356f520e9adb1b1b7802705f340c2b
+        uses: actions/upload-artifact@b4b15b8c7c6ac21ea08fcf65892d2ee8f75cf882
         with:
-          name: ${{ github.sha }}_${{ matrix.command }}_${{ matrix.op_flavor }}_${{ matrix.bench_type }}
+          name: ${{ github.sha }}_${{ matrix.command }}_${{ matrix.op_flavor }}
           path: ${{ env.RESULTS_FILENAME }}
 
       - name: Send data to Slab
@@ -191,7 +167,7 @@ jobs:
     steps:
       - name: Stop instance
         id: stop-instance
-        uses: zama-ai/slab-github-runner@79939325c3c429837c10d6041e4fd8589d328bac
+        uses: zama-ai/slab-github-runner@801df0b8db5ea2b06128b7476c652f5ed5f193a8
         with:
           mode: stop
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}

.github/workflows/benchmark_tfhe_fft.yml

@@ -1,143 +0,0 @@
-# Run FFT benchmarks on an AWS instance and return parsed results to Slab CI bot.
-name: FFT benchmarks
-
-env:
-  CARGO_TERM_COLOR: always
-  ACTION_RUN_URL: ${{ github.server_url }}/${{ github.repository }}/actions/runs/${{ github.run_id }}
-  RESULTS_FILENAME: parsed_benchmark_results_${{ github.sha }}.json
-  RUST_BACKTRACE: "full"
-  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:
-  workflow_dispatch:
-  push:
-    branches:
-      - "main"
-    paths:
-      - tfhe-fft/**
-      - .github/workflows/benchmark_tfhe_fft.yml
-  schedule:
-    # Job will be triggered each Thursday at 11p.m.
-    - cron: '0 23 * * 4'
-
-jobs:
-  setup-ec2:
-    name: Setup EC2 instance (fft-benchmarks)
-    runs-on: ubuntu-latest
-    outputs:
-      runner-name: ${{ steps.start-instance.outputs.label }}
-    steps:
-      - name: Start instance
-        id: start-instance
-        uses: zama-ai/slab-github-runner@79939325c3c429837c10d6041e4fd8589d328bac
-        with:
-          mode: start
-          github-token: ${{ secrets.SLAB_ACTION_TOKEN }}
-          slab-url: ${{ secrets.SLAB_BASE_URL }}
-          job-secret: ${{ secrets.JOB_SECRET }}
-          backend: aws
-          profile: bench
-
-  fft-benchmarks:
-    name: Execute FFT benchmarks in EC2
-    needs: setup-ec2
-    concurrency:
-      group: ${{ github.workflow }}_${{ github.ref }}
-      cancel-in-progress: true
-    runs-on: ${{ needs.setup-ec2.outputs.runner-name }}
-    steps:
-      - name: Checkout tfhe-rs repo with tags
-        uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
-        with:
-          fetch-depth: 0
-
-      - name: Get benchmark details
-        run: |
-          {
-            echo "BENCH_DATE=$(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: Install rust
-        uses: actions-rs/toolchain@16499b5e05bf2e26879000db0c1d13f7e13fa3af
-        with:
-          toolchain: nightly
-          override: true
-
-      - name: Run benchmarks with AVX512
-        run: |
-          make bench_fft
-
-      - name: Parse AVX512 results
-        run: |
-          python3 ./ci/fft_benchmark_parser.py target/criterion ${{ env.RESULTS_FILENAME }} \
-          --database concrete_fft \
-          --hardware "hpc7a.96xlarge" \
-          --project-version "${{ env.COMMIT_HASH }}" \
-          --branch ${{ github.ref_name }} \
-          --commit-date "${{ env.COMMIT_DATE }}" \
-          --bench-date "${{ env.BENCH_DATE }}" \
-          --name-suffix avx512
-
-      - name: Upload parsed results artifact
-        uses: actions/upload-artifact@6f51ac03b9356f520e9adb1b1b7802705f340c2b
-        with:
-          name: ${{ github.sha }}_fft
-          path: ${{ env.RESULTS_FILENAME }}
-
-      - name: Checkout Slab repo
-        uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
-        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@c33737706dea87cd7784c687dadc9adf1be59990
-        env:
-          SLACK_COLOR: ${{ job.status }}
-          SLACK_MESSAGE: "tfhe-fft benchmarks failed. (${{ env.ACTION_RUN_URL }})"
-
-  teardown-ec2:
-    name: Teardown EC2 instance (fft-benchmarks)
-    if: ${{ always() && needs.setup-ec2.result != 'skipped' }}
-    needs: [ setup-ec2, fft-benchmarks ]
-    runs-on: ubuntu-latest
-    steps:
-      - name: Stop instance
-        id: stop-instance
-        uses: zama-ai/slab-github-runner@79939325c3c429837c10d6041e4fd8589d328bac
-        with:
-          mode: stop
-          github-token: ${{ secrets.SLAB_ACTION_TOKEN }}
-          slab-url: ${{ secrets.SLAB_BASE_URL }}
-          job-secret: ${{ secrets.JOB_SECRET }}
-          label: ${{ needs.setup-ec2.outputs.runner-name }}
-
-      - name: Slack Notification
-        if: ${{ failure() }}
-        continue-on-error: true
-        uses: rtCamp/action-slack-notify@c33737706dea87cd7784c687dadc9adf1be59990
-        env:
-          SLACK_COLOR: ${{ job.status }}
-          SLACK_MESSAGE: "EC2 teardown (fft-benchmarks) failed. (${{ env.ACTION_RUN_URL }})"

.github/workflows/benchmark_tfhe_ntt.yml

@@ -1,143 +0,0 @@
-# Run NTT benchmarks on an AWS instance and return parsed results to Slab CI bot.
-name: NTT benchmarks
-
-env:
-  CARGO_TERM_COLOR: always
-  ACTION_RUN_URL: ${{ github.server_url }}/${{ github.repository }}/actions/runs/${{ github.run_id }}
-  RESULTS_FILENAME: parsed_benchmark_results_${{ github.sha }}.json
-  RUST_BACKTRACE: "full"
-  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:
-  workflow_dispatch:
-  push:
-    branches:
-      - "main"
-    paths:
-      - tfhe-ntt/**
-      - .github/workflows/benchmark_tfhe_ntt.yml
-  schedule:
-    # Job will be triggered each Friday at 11p.m.
-    - cron: "0 23 * * 5"
-
-jobs:
-  setup-ec2:
-    name: Setup EC2 instance (ntt-benchmarks)
-    runs-on: ubuntu-latest
-    outputs:
-      runner-name: ${{ steps.start-instance.outputs.label }}
-    steps:
-      - name: Start instance
-        id: start-instance
-        uses: zama-ai/slab-github-runner@79939325c3c429837c10d6041e4fd8589d328bac
-        with:
-          mode: start
-          github-token: ${{ secrets.SLAB_ACTION_TOKEN }}
-          slab-url: ${{ secrets.SLAB_BASE_URL }}
-          job-secret: ${{ secrets.JOB_SECRET }}
-          backend: aws
-          profile: bench
-
-  ntt-benchmarks:
-    name: Execute NTT benchmarks in EC2
-    needs: setup-ec2
-    concurrency:
-      group: ${{ github.workflow }}_${{ github.ref }}
-      cancel-in-progress: true
-    runs-on: ${{ needs.setup-ec2.outputs.runner-name }}
-    steps:
-      - name: Checkout tfhe-rs repo with tags
-        uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
-        with:
-          fetch-depth: 0
-
-      - name: Get benchmark details
-        run: |
-          {
-            echo "BENCH_DATE=$(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: Install rust
-        uses: actions-rs/toolchain@16499b5e05bf2e26879000db0c1d13f7e13fa3af
-        with:
-          toolchain: nightly
-          override: true
-
-      - name: Run benchmarks
-        run: |
-          make bench_ntt
-
-      - name: Parse results
-        run: |
-          python3 ./ci/ntt_benchmark_parser.py target/criterion ${{ env.RESULTS_FILENAME }} \
-          --database concrete_ntt \
-          --hardware "hpc7a.96xlarge" \
-          --project-version "${{ env.COMMIT_HASH }}" \
-          --branch ${{ github.ref_name }} \
-          --commit-date "${{ env.COMMIT_DATE }}" \
-          --bench-date "${{ env.BENCH_DATE }}" \
-          --name-suffix avx512
-
-      - name: Upload parsed results artifact
-        uses: actions/upload-artifact@6f51ac03b9356f520e9adb1b1b7802705f340c2b
-        with:
-          name: ${{ github.sha }}_ntt
-          path: ${{ env.RESULTS_FILENAME }}
-
-      - name: Checkout Slab repo
-        uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
-        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@c33737706dea87cd7784c687dadc9adf1be59990
-        env:
-          SLACK_COLOR: ${{ job.status }}
-          SLACK_MESSAGE: "tfhe-ntt benchmarks failed. (${{ env.ACTION_RUN_URL }})"
-
-  teardown-ec2:
-    name: Teardown EC2 instance (ntt-benchmarks)
-    if: ${{ always() && needs.setup-ec2.result != 'skipped' }}
-    needs: [setup-ec2, ntt-benchmarks]
-    runs-on: ubuntu-latest
-    steps:
-      - name: Stop instance
-        id: stop-instance
-        uses: zama-ai/slab-github-runner@79939325c3c429837c10d6041e4fd8589d328bac
-        with:
-          mode: stop
-          github-token: ${{ secrets.SLAB_ACTION_TOKEN }}
-          slab-url: ${{ secrets.SLAB_BASE_URL }}
-          job-secret: ${{ secrets.JOB_SECRET }}
-          label: ${{ needs.setup-ec2.outputs.runner-name }}
-
-      - name: Slack Notification
-        if: ${{ failure() }}
-        continue-on-error: true
-        uses: rtCamp/action-slack-notify@c33737706dea87cd7784c687dadc9adf1be59990
-        env:
-          SLACK_COLOR: ${{ job.status }}
-          SLACK_MESSAGE: "EC2 teardown (ntt-benchmarks) failed. (${{ env.ACTION_RUN_URL }})"

.github/workflows/benchmark_tfhe_zk_pok.yml

@@ -1,172 +0,0 @@
-# Run benchmarks of the tfhe-zk-pok crate on an instance and return parsed results to Slab CI bot.
-name: tfhe-zk-pok benchmarks
-
-on:
-  workflow_dispatch:
-  push:
-    branches:
-      - main
-  schedule:
-    # Weekly benchmarks will be triggered each Saturday at 3a.m.
-    - cron: '0 3 * * 6'
-env:
-  CARGO_TERM_COLOR: always
-  RESULTS_FILENAME: parsed_benchmark_results_${{ github.sha }}.json
-  PARSE_INTEGER_BENCH_CSV_FILE: tfhe_rs_integer_benches_${{ github.sha }}.csv
-  ACTION_RUN_URL: ${{ github.server_url }}/${{ github.repository }}/actions/runs/${{ github.run_id }}
-  RUST_BACKTRACE: "full"
-  RUST_MIN_STACK: "8388608"
-  SLACK_CHANNEL: ${{ secrets.SLACK_CHANNEL }}
-  SLACK_ICON: https://pbs.twimg.com/profile_images/1274014582265298945/OjBKP9kn_400x400.png
-  SLACK_USERNAME: ${{ secrets.BOT_USERNAME }}
-  SLACK_WEBHOOK: ${{ secrets.SLACK_WEBHOOK }}
-
-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')
-    outputs:
-      zk_pok_changed: ${{ steps.changed-files.outputs.zk_pok_any_changed }}
-    steps:
-      - name: Checkout tfhe-rs
-        uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
-        with:
-          fetch-depth: 0
-
-      - name: Check for file changes
-        id: changed-files
-        uses: tj-actions/changed-files@d6e91a2266cdb9d62096cebf1e8546899c6aa18f
-        with:
-          since_last_remote_commit: true
-          files_yaml: |
-            zk_pok:
-              - tfhe-zk-pok/**
-              - .github/workflows/benchmark_tfhe_zk_pok.yml
-
-  setup-instance:
-    name: Setup instance (tfhe-zk-pok-benchmarks)
-    runs-on: ubuntu-latest
-    needs: should-run
-    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.zk_pok_changed == 'true')
-    outputs:
-      runner-name: ${{ steps.start-instance.outputs.label }}
-    steps:
-      - name: Start instance
-        id: start-instance
-        uses: zama-ai/slab-github-runner@79939325c3c429837c10d6041e4fd8589d328bac
-        with:
-          mode: start
-          github-token: ${{ secrets.SLAB_ACTION_TOKEN }}
-          slab-url: ${{ secrets.SLAB_BASE_URL }}
-          job-secret: ${{ secrets.JOB_SECRET }}
-          backend: aws
-          profile: bench
-
-  tfhe-zk-pok-benchmarks:
-    name: Execute tfhe-zk-pok benchmarks
-    if: needs.setup-instance.result != 'skipped'
-    needs: setup-instance
-    concurrency:
-      group: ${{ github.workflow }}_${{github.event_name}}_${{ github.ref }}${{ github.ref == 'refs/heads/main' && github.sha || '' }}
-      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@11bd71901bbe5b1630ceea73d27597364c9af683
-        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: Install rust
-        uses: dtolnay/rust-toolchain@a54c7afa936fefeb4456b2dd8068152669aa8203
-        with:
-          toolchain: nightly
-
-      - name: Checkout Slab repo
-        uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
-        with:
-          repository: zama-ai/slab
-          path: slab
-          token: ${{ secrets.FHE_ACTIONS_TOKEN }}
-
-      - name: Run benchmarks
-        run: |
-          make bench_tfhe_zk_pok
-
-      - name: Parse results
-        run: |
-          python3 ./ci/benchmark_parser.py target/criterion ${{ env.RESULTS_FILENAME }} \
-          --database tfhe_rs \
-          --crate tfhe-zk-pok \
-          --hardware "hpc7a.96xlarge" \
-          --backend cpu \
-          --project-version "${{ env.COMMIT_HASH }}" \
-          --branch ${{ github.ref_name }} \
-          --commit-date "${{ env.COMMIT_DATE }}" \
-          --bench-date "${{ env.BENCH_DATE }}" \
-          --walk-subdirs \
-          --name-suffix avx512
-
-      - name: Upload parsed results artifact
-        uses: actions/upload-artifact@6f51ac03b9356f520e9adb1b1b7802705f340c2b
-        with:
-          name: ${{ github.sha }}_tfhe_zk_pok
-          path: ${{ env.RESULTS_FILENAME }}
-
-      - name: Checkout Slab repo
-        uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
-        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@c33737706dea87cd7784c687dadc9adf1be59990
-        env:
-          SLACK_COLOR: ${{ job.status }}
-          SLACK_MESSAGE: "tfhe-zk-pok benchmarks finished with status: ${{ job.status }}. (${{ env.ACTION_RUN_URL }})"
-
-  teardown-instance:
-    name: Teardown instance (tfhe-zk-pok-benchmarks)
-    if: ${{ always() && needs.setup-instance.result != 'skipped' }}
-    needs: [ setup-instance, tfhe-zk-pok-benchmarks ]
-    runs-on: ubuntu-latest
-    steps:
-      - name: Stop instance
-        id: stop-instance
-        uses: zama-ai/slab-github-runner@79939325c3c429837c10d6041e4fd8589d328bac
-        with:
-          mode: stop
-          github-token: ${{ secrets.SLAB_ACTION_TOKEN }}
-          slab-url: ${{ secrets.SLAB_BASE_URL }}
-          job-secret: ${{ secrets.JOB_SECRET }}
-          label: ${{ needs.setup-instance.outputs.runner-name }}
-
-      - name: Slack Notification
-        if: ${{ failure() }}
-        continue-on-error: true
-        uses: rtCamp/action-slack-notify@c33737706dea87cd7784c687dadc9adf1be59990
-        env:
-          SLACK_COLOR: ${{ job.status }}
-          SLACK_MESSAGE: "Instance teardown (tfhe-zk-pok-benchmarks) finished with status: ${{ job.status }}. (${{ env.ACTION_RUN_URL }})"

.github/workflows/benchmark_wasm_client.yml

@@ -36,17 +36,16 @@ jobs:
         uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
         with:
           fetch-depth: 0
-          token: ${{ secrets.FHE_ACTIONS_TOKEN }}
 
       - name: Check for file changes
         id: changed-files
-        uses: tj-actions/changed-files@d6e91a2266cdb9d62096cebf1e8546899c6aa18f
+        uses: tj-actions/changed-files@c3a1bb2c992d77180ae65be6ae6c166cf40f857c
         with:
           since_last_remote_commit: true
           files_yaml: |
             wasm_bench:
               - tfhe/Cargo.toml
-              - tfhe-csprng/**
+              - concrete-csprng/**
               - tfhe-zk-pok/**
               - tfhe/src/**
               - '!tfhe/src/c_api/**'
@@ -65,7 +64,7 @@ jobs:
     steps:
       - name: Start instance
         id: start-instance
-        uses: zama-ai/slab-github-runner@79939325c3c429837c10d6041e4fd8589d328bac
+        uses: zama-ai/slab-github-runner@801df0b8db5ea2b06128b7476c652f5ed5f193a8
         with:
           mode: start
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}
@@ -99,7 +98,7 @@ jobs:
           } >> "${GITHUB_ENV}"
 
       - name: Install rust
-        uses: dtolnay/rust-toolchain@a54c7afa936fefeb4456b2dd8068152669aa8203
+        uses: dtolnay/rust-toolchain@7b1c307e0dcbda6122208f10795a713336a9b35a
         with:
           toolchain: nightly
 
@@ -109,7 +108,7 @@ jobs:
 
       - name: Node cache restoration
         id: node-cache
-        uses: actions/cache/restore@1bd1e32a3bdc45362d1e726936510720a7c30a57 #v4.2.0
+        uses: actions/cache/restore@6849a6489940f00c2f30c0fb92c6274307ccb58a #v4.1.2
         with:
           path: |
             ~/.nvm
@@ -122,7 +121,7 @@ jobs:
           make install_node
 
       - name: Node cache save
-        uses: actions/cache/save@1bd1e32a3bdc45362d1e726936510720a7c30a57 #v4.2.0
+        uses: actions/cache/save@6849a6489940f00c2f30c0fb92c6274307ccb58a #v4.1.2
         if: steps.node-cache.outputs.cache-hit != 'true'
         with:
           path: |
@@ -166,7 +165,7 @@ jobs:
           --append-results
 
       - name: Upload parsed results artifact
-        uses: actions/upload-artifact@6f51ac03b9356f520e9adb1b1b7802705f340c2b
+        uses: actions/upload-artifact@b4b15b8c7c6ac21ea08fcf65892d2ee8f75cf882
         with:
           name: ${{ github.sha }}_wasm_${{ matrix.browser }}
           path: ${{ env.RESULTS_FILENAME }}
@@ -200,7 +199,7 @@ jobs:
     steps:
       - name: Stop instance
         id: stop-instance
-        uses: zama-ai/slab-github-runner@79939325c3c429837c10d6041e4fd8589d328bac
+        uses: zama-ai/slab-github-runner@801df0b8db5ea2b06128b7476c652f5ed5f193a8
         with:
           mode: stop
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}

.github/workflows/benchmark_zk_pke.yml

@@ -3,16 +3,6 @@ name: PKE ZK benchmarks
 
 on:
   workflow_dispatch:
-    inputs:
-      bench_type:
-        description: "Benchmarks type"
-        type: choice
-        default: latency
-        options:
-          - latency
-          - throughput
-          - both
-
   push:
     branches:
       - main
@@ -43,18 +33,16 @@ jobs:
         uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
         with:
           fetch-depth: 0
-          token: ${{ secrets.FHE_ACTIONS_TOKEN }}
 
       - name: Check for file changes
         id: changed-files
-        uses: tj-actions/changed-files@d6e91a2266cdb9d62096cebf1e8546899c6aa18f
+        uses: tj-actions/changed-files@c3a1bb2c992d77180ae65be6ae6c166cf40f857c
         with:
           since_last_remote_commit: true
           files_yaml: |
             zk_pok:
               - tfhe/Cargo.toml
-              - tfhe-csprng/**
-              - tfhe-fft/**
+              - concrete-csprng/**
               - tfhe-zk-pok/**
               - tfhe/src/core_crypto/**
               - tfhe/src/shortint/**
@@ -63,37 +51,10 @@ jobs:
               - tfhe/benches/integer/zk_pke.rs
               - .github/workflows/zk_pke_benchmark.yml
 
-  prepare-matrix:
-    name: Prepare operations matrix
-    runs-on: ubuntu-latest
-    if: github.event_name != 'schedule' ||
-      (github.event_name == 'schedule' && github.repository == 'zama-ai/tfhe-rs')
-    outputs:
-      bench_type: ${{ steps.set_bench_type.outputs.bench_type }}
-    steps:
-      - name: Set benchmark types
-        if: github.event_name == 'workflow_dispatch'
-        run: |
-          if [[ "${{ inputs.bench_type }}" == "both" ]]; then
-            echo "BENCH_TYPE=[\"latency\", \"throughput\"]" >> "${GITHUB_ENV}"
-          else
-            echo "BENCH_TYPE=[\"${{ inputs.bench_type }}\"]" >> "${GITHUB_ENV}"
-          fi
-
-      - name: Default benchmark type
-        if: github.event_name != 'workflow_dispatch'
-        run: |
-          echo "BENCH_TYPE=[\"latency\"]" >> "${GITHUB_ENV}"
-
-      - name: Set benchmark types output
-        id: set_bench_type
-        run: |
-          echo "bench_type=${{ toJSON(env.BENCH_TYPE) }}" >> "${GITHUB_OUTPUT}"
-
   setup-instance:
     name: Setup instance (pke-zk-benchmarks)
     runs-on: ubuntu-latest
-    needs: [ should-run, prepare-matrix ]
+    needs: should-run
     if: github.event_name == 'workflow_dispatch' ||
       (github.event_name == 'schedule' && github.repository == 'zama-ai/tfhe-rs') ||
       (github.event_name == 'push' &&
@@ -104,7 +65,7 @@ jobs:
     steps:
       - name: Start instance
         id: start-instance
-        uses: zama-ai/slab-github-runner@79939325c3c429837c10d6041e4fd8589d328bac
+        uses: zama-ai/slab-github-runner@801df0b8db5ea2b06128b7476c652f5ed5f193a8
         with:
           mode: start
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}
@@ -116,15 +77,11 @@ jobs:
   pke-zk-benchmarks:
     name: Execute PKE ZK benchmarks
     if: needs.setup-instance.result != 'skipped'
-    needs: [ prepare-matrix, setup-instance ]
+    needs: setup-instance
     concurrency:
       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 }}
-    strategy:
-      max-parallel: 1
-      matrix:
-        bench_type: ${{ fromJSON(needs.prepare-matrix.outputs.bench_type) }}
     steps:
       - name: Checkout tfhe-rs repo with tags
         uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
@@ -140,8 +97,13 @@ jobs:
             echo "COMMIT_HASH=$(git describe --tags --dirty)";
           } >> "${GITHUB_ENV}"
 
+      - name: Set up home
+        # "Install rust" step require root user to have a HOME directory which is not set.
+        run: |
+          echo "HOME=/home/ubuntu" >> "${GITHUB_ENV}"
+
       - name: Install rust
-        uses: dtolnay/rust-toolchain@a54c7afa936fefeb4456b2dd8068152669aa8203
+        uses: dtolnay/rust-toolchain@7b1c307e0dcbda6122208f10795a713336a9b35a
         with:
           toolchain: nightly
 
@@ -154,7 +116,7 @@ jobs:
 
       - name: Run benchmarks with AVX512
         run: |
-          make BENCH_TYPE=${{ matrix.bench_type }} bench_integer_zk
+          make bench_integer_zk
 
       - name: Parse results
         run: |
@@ -168,7 +130,7 @@ jobs:
           --bench-date "${{ env.BENCH_DATE }}" \
           --walk-subdirs \
           --name-suffix avx512 \
-          --bench-type ${{ matrix.bench_type }}
+          --throughput
 
       - name: Parse CRS sizes results
         run: |
@@ -177,7 +139,7 @@ jobs:
           --append-results
 
       - name: Upload parsed results artifact
-        uses: actions/upload-artifact@6f51ac03b9356f520e9adb1b1b7802705f340c2b
+        uses: actions/upload-artifact@b4b15b8c7c6ac21ea08fcf65892d2ee8f75cf882
         with:
           name: ${{ github.sha }}_integer_zk
           path: ${{ env.RESULTS_FILENAME }}
@@ -211,7 +173,7 @@ jobs:
     steps:
       - name: Stop instance
         id: stop-instance
-        uses: zama-ai/slab-github-runner@79939325c3c429837c10d6041e4fd8589d328bac
+        uses: zama-ai/slab-github-runner@801df0b8db5ea2b06128b7476c652f5ed5f193a8
         with:
           mode: stop
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}

.github/workflows/cargo_build.yml

@@ -1,4 +1,4 @@
-name: Cargo Build TFHE-rs
+name: Cargo Build
 
 on:
   pull_request:
@@ -28,7 +28,7 @@ jobs:
       - uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
 
       - name: Install latest stable
-        uses: dtolnay/rust-toolchain@a54c7afa936fefeb4456b2dd8068152669aa8203
+        uses: dtolnay/rust-toolchain@7b1c307e0dcbda6122208f10795a713336a9b35a
         with:
           toolchain: stable
 
@@ -47,10 +47,10 @@ jobs:
         run: |
           make pcc
 
-      - name: Build tfhe-csprng
+      - name: Build concrete-csprng
         if: ${{ contains(matrix.os, 'ubuntu') }}
         run: |
-          make build_tfhe_csprng
+          make build_concrete_csprng
 
       - name: Build Release core
         if: ${{ contains(matrix.os, 'ubuntu') }}

.github/workflows/cargo_build_tfhe_fft.yml

@@ -1,44 +0,0 @@
-# Build tfhe-fft
-name: Cargo Build tfhe-fft
-
-on:
-  pull_request:
-
-env:
-  CARGO_TERM_COLOR: always
-
-concurrency:
-  group: ${{ github.workflow }}-${{ github.head_ref }}
-  cancel-in-progress: true
-
-jobs:
-  cargo-builds-fft:
-    runs-on: ${{ matrix.runner_type }}
-
-    strategy:
-      matrix:
-        runner_type: [ubuntu-latest, macos-latest, windows-latest]
-      fail-fast: false
-
-    steps:
-      - uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
-
-      - name: Install Rust
-        uses: actions-rs/toolchain@16499b5e05bf2e26879000db0c1d13f7e13fa3af
-        with:
-          toolchain: stable
-          override: true
-
-      - name: Run pcc checks
-        if: matrix.runner_type == 'ubuntu-latest'
-        run: |
-          sudo apt install -y libfftw3-dev
-          make pcc_fft
-
-      - name: Build release
-        run: |
-          make build_fft
-
-      - name: Build release no-std
-        run: |
-          make build_fft_no_std

.github/workflows/cargo_build_tfhe_ntt.yml

@@ -1,40 +0,0 @@
-# Build tfhe-ntt
-name: Cargo Build tfhe-ntt
-
-on:
-  pull_request:
-
-env:
-  CARGO_TERM_COLOR: always
-
-concurrency:
-  group: ${{ github.workflow }}-${{ github.head_ref }}
-  cancel-in-progress: true
-
-jobs:
-  cargo-builds:
-    runs-on: ${{ matrix.os }}
-    strategy:
-      matrix:
-        os: [ubuntu-latest, macos-latest, windows-latest]
-      fail-fast: false
-    steps:
-      - uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
-
-      - name: Install Rust
-        uses: actions-rs/toolchain@16499b5e05bf2e26879000db0c1d13f7e13fa3af
-        with:
-          toolchain: stable
-          override: true
-
-      - name: Run pcc checks
-        run: |
-          make pcc_ntt
-
-      - name: Build release
-        run: |
-          make build_ntt
-
-      - name: Build release no-std
-        run: |
-          make build_ntt_no_std

.github/workflows/cargo_test_fft.yml

@@ -1,71 +0,0 @@
-# Test tfhe-fft
-name: Cargo Test tfhe-fft
-
-on:
-  pull_request:
-
-env:
-  CARGO_TERM_COLOR: always
-
-concurrency:
-  group: ${{ github.workflow }}-${{ github.head_ref }}
-  cancel-in-progress: true
-
-jobs:
-  cargo-tests:
-    runs-on: ${{ matrix.runner_type }}
-    strategy:
-      matrix:
-        runner_type: [ubuntu-latest, macos-latest, windows-latest]
-      fail-fast: false
-    steps:
-      - uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
-
-      - name: Install Rust
-        uses: actions-rs/toolchain@16499b5e05bf2e26879000db0c1d13f7e13fa3af
-        with:
-          toolchain: stable
-          override: true
-
-      - name: Test debug
-        run: |
-          make test_fft
-
-      - name: Test serialization
-        run: make test_fft_serde
-
-      - name: Test no-std
-        run: |
-          make test_fft_no_std
-
-  cargo-tests-nightly:
-    runs-on: ${{ matrix.runner_type }}
-    strategy:
-      matrix:
-        runner_type: [ubuntu-latest, macos-latest, windows-latest]
-    steps:
-      - uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
-
-      - name: Install Rust
-        uses: actions-rs/toolchain@16499b5e05bf2e26879000db0c1d13f7e13fa3af
-        with:
-          toolchain: nightly
-          override: true
-
-      - name: Test nightly
-        run: |
-          make test_fft_nightly
-
-      - name: Test no-std nightly
-        run: |
-          make test_fft_no_std_nightly
-
-  cargo-tests-node-js:
-    runs-on: "ubuntu-latest"
-    steps:
-      - uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
-
-      - name: Test node js
-        run: |
-          make install_node
-          make test_fft_node_js_ci

.github/workflows/cargo_test_ntt.yml

@@ -1,54 +0,0 @@
-# Test tfhe-ntt
-name: Cargo Test tfhe-ntt
-
-on:
-  pull_request:
-
-env:
-  CARGO_TERM_COLOR: always
-
-concurrency:
-  group: ${{ github.workflow }}-${{ github.head_ref }}
-  cancel-in-progress: true
-
-jobs:
-  cargo-tests:
-    runs-on: ${{ matrix.os }}
-    strategy:
-      matrix:
-        os: [ubuntu-latest, macos-latest, windows-latest]
-      fail-fast: false
-    steps:
-      - uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
-
-      - name: Install Rust
-        uses: actions-rs/toolchain@16499b5e05bf2e26879000db0c1d13f7e13fa3af
-        with:
-          toolchain: stable
-          override: true
-
-      - name: Test debug
-        run: make test_ntt
-
-      - name: Test no-std
-        run: make test_ntt_no_std
-
-  cargo-tests-nightly:
-    runs-on: ${{ matrix.os }}
-    strategy:
-      matrix:
-        os: [ubuntu-latest, macos-latest, windows-latest]
-    steps:
-      - uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
-
-      - name: Install Rust
-        uses: actions-rs/toolchain@16499b5e05bf2e26879000db0c1d13f7e13fa3af
-        with:
-          toolchain: nightly
-          override: true
-
-      - name: Test nightly
-        run: make test_ntt_nightly
-
-      - name: Test no-std nightly
-        run: make test_ntt_no_std_nightly

.github/workflows/check_triggering_actor.yml

@@ -1,29 +0,0 @@
-# Check if triggering actor is a collaborator and has write access
-name: Check Triggering Actor
-
-on:
-  workflow_call:
-    secrets:
-      TOKEN:
-        required: true
-
-jobs:
-  check-actor-permission:
-    runs-on: ubuntu-latest
-    steps:
-      - name: Get User Permission
-        id: check-access
-        uses: actions-cool/check-user-permission@956b2e73cdfe3bcb819bb7225e490cb3b18fd76e # v2.2.1
-        with:
-          require: write
-          username: ${{ github.triggering_actor }}
-        env:
-          GITHUB_TOKEN: ${{ secrets.TOKEN }}
-
-      - name: Check User Permission
-        if: steps.check-access.outputs.require-result == 'false'
-        run: |
-          echo "${{ github.triggering_actor }} does not have permissions on this repo."
-          echo "Current permission level is ${{ steps.check-access.outputs.user-permission }}"
-          echo "Job originally triggered by ${{ github.actor }}"
-          exit 1

.github/workflows/ci_lint.yml

@@ -27,7 +27,7 @@ jobs:
           make lint_workflow
 
       - name: Ensure SHA pinned actions
-        uses: zgosalvez/github-actions-ensure-sha-pinned-actions@6ae615f6475d2ede5ad88bea6baa7a1d5e93ffaa # v3.0.19
+        uses: zgosalvez/github-actions-ensure-sha-pinned-actions@38608ef4fb69adae7f1eac6eeb88e67b7d083bfd # v3.0.16
         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@79939325c3c429837c10d6041e4fd8589d328bac
+        uses: zama-ai/slab-github-runner@801df0b8db5ea2b06128b7476c652f5ed5f193a8
         with:
           mode: start
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}
@@ -47,19 +47,19 @@ jobs:
         uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
 
       - name: Install latest stable
-        uses: dtolnay/rust-toolchain@a54c7afa936fefeb4456b2dd8068152669aa8203
+        uses: dtolnay/rust-toolchain@7b1c307e0dcbda6122208f10795a713336a9b35a
         with:
           toolchain: stable
 
       - name: Check for file changes
         id: changed-files
-        uses: tj-actions/changed-files@d6e91a2266cdb9d62096cebf1e8546899c6aa18f
+        uses: tj-actions/changed-files@c3a1bb2c992d77180ae65be6ae6c166cf40f857c
         with:
           files_yaml: |
             tfhe:
               - tfhe/src/**
-            tfhe_csprng:
-              - tfhe-csprng/src/**
+            concrete_csprng:
+              - concrete-csprng/src/**
 
       - name: Generate Keys
         if: steps.changed-files.outputs.tfhe_any_changed == 'true'
@@ -83,7 +83,7 @@ jobs:
           make test_shortint_cov
 
       - name: Upload tfhe coverage to Codecov
-        uses: codecov/codecov-action@1e68e06f1dbfde0e4cefc87efeba9e4643565303
+        uses: codecov/codecov-action@b9fd7d16f6d7d1b5d2bec1a2887e65ceed900238
         if: steps.changed-files.outputs.tfhe_any_changed == 'true'
         with:
           token: ${{ secrets.CODECOV_TOKEN }}
@@ -97,7 +97,7 @@ jobs:
           make test_integer_cov
 
       - name: Upload tfhe coverage to Codecov
-        uses: codecov/codecov-action@1e68e06f1dbfde0e4cefc87efeba9e4643565303
+        uses: codecov/codecov-action@b9fd7d16f6d7d1b5d2bec1a2887e65ceed900238
         if: steps.changed-files.outputs.tfhe_any_changed == 'true'
         with:
           token: ${{ secrets.CODECOV_TOKEN }}
@@ -121,7 +121,7 @@ jobs:
     steps:
       - name: Stop instance
         id: stop-instance
-        uses: zama-ai/slab-github-runner@79939325c3c429837c10d6041e4fd8589d328bac
+        uses: zama-ai/slab-github-runner@801df0b8db5ea2b06128b7476c652f5ed5f193a8
         with:
           mode: stop
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}

.github/workflows/csprng_randomness_tests.yml

@@ -27,7 +27,7 @@ jobs:
     steps:
       - name: Start instance
         id: start-instance
-        uses: zama-ai/slab-github-runner@79939325c3c429837c10d6041e4fd8589d328bac
+        uses: zama-ai/slab-github-runner@801df0b8db5ea2b06128b7476c652f5ed5f193a8
         with:
           mode: start
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}
@@ -51,7 +51,7 @@ jobs:
           token: ${{ secrets.FHE_ACTIONS_TOKEN }}
 
       - name: Install latest stable
-        uses: dtolnay/rust-toolchain@a54c7afa936fefeb4456b2dd8068152669aa8203
+        uses: dtolnay/rust-toolchain@7b1c307e0dcbda6122208f10795a713336a9b35a
         with:
           toolchain: stable
 
@@ -65,7 +65,7 @@ jobs:
         uses: rtCamp/action-slack-notify@c33737706dea87cd7784c687dadc9adf1be59990
         env:
           SLACK_COLOR: ${{ job.status }}
-          SLACK_MESSAGE: "tfhe-csprng randomness check finished with status: ${{ job.status }}. (${{ env.ACTION_RUN_URL }})"
+          SLACK_MESSAGE: "concrete-csprng randomness check finished with status: ${{ job.status }}. (${{ env.ACTION_RUN_URL }})"
 
   teardown-instance:
     name: Teardown instance (csprng-randomness-tests)
@@ -75,7 +75,7 @@ jobs:
     steps:
       - name: Stop instance
         id: stop-instance
-        uses: zama-ai/slab-github-runner@79939325c3c429837c10d6041e4fd8589d328bac
+        uses: zama-ai/slab-github-runner@801df0b8db5ea2b06128b7476c652f5ed5f193a8
         with:
           mode: stop
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}

.github/workflows/gpu_4090_tests.yml

@@ -40,7 +40,7 @@ jobs:
           token: ${{ secrets.FHE_ACTIONS_TOKEN }}
 
       - name: Install latest stable
-        uses: dtolnay/rust-toolchain@a54c7afa936fefeb4456b2dd8068152669aa8203
+        uses: dtolnay/rust-toolchain@7b1c307e0dcbda6122208f10795a713336a9b35a
         with:
           toolchain: stable
 

.github/workflows/gpu_fast_h100_tests.yml

@@ -31,11 +31,10 @@ jobs:
         uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
         with:
           fetch-depth: 0
-          token: ${{ secrets.FHE_ACTIONS_TOKEN }}
 
       - name: Check for file changes
         id: changed-files
-        uses: tj-actions/changed-files@d6e91a2266cdb9d62096cebf1e8546899c6aa18f
+        uses: tj-actions/changed-files@c3a1bb2c992d77180ae65be6ae6c166cf40f857c
         with:
           since_last_remote_commit: true
           files_yaml: |
@@ -45,9 +44,6 @@ jobs:
               - backends/tfhe-cuda-backend/**
               - tfhe/src/core_crypto/gpu/**
               - tfhe/src/integer/gpu/**
-              - tfhe/src/integer/server_key/radix_parallel/tests_unsigned/**
-              - tfhe/src/integer/server_key/radix_parallel/tests_signed/**
-              - tfhe/src/integer/server_key/radix_parallel/tests_cases_unsigned.rs
               - tfhe/src/shortint/parameters/**
               - tfhe/src/high_level_api/**
               - tfhe/src/c_api/**
@@ -68,7 +64,7 @@ jobs:
     steps:
       - name: Start instance
         id: start-instance
-        uses: zama-ai/slab-github-runner@79939325c3c429837c10d6041e4fd8589d328bac
+        uses: zama-ai/slab-github-runner@801df0b8db5ea2b06128b7476c652f5ed5f193a8
         with:
           mode: start
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}
@@ -94,28 +90,60 @@ jobs:
           - os: ubuntu-22.04
             cuda: "12.2"
             gcc: 11 
+    env:
+      CUDA_PATH: /usr/local/cuda-${{ matrix.cuda }}
+      CMAKE_VERSION: 3.29.6
     steps:
+      # Mandatory on hyperstack since a bootable volume is not re-usable yet.
+      - name: Install dependencies
+        run: |
+          sudo apt update
+          sudo apt install -y checkinstall zlib1g-dev libssl-dev libclang-dev
+          wget https://github.com/Kitware/CMake/releases/download/v${{ env.CMAKE_VERSION }}/cmake-${{ env.CMAKE_VERSION }}.tar.gz
+          tar -zxvf cmake-${{ env.CMAKE_VERSION }}.tar.gz
+          cd cmake-${{ env.CMAKE_VERSION }}
+          ./bootstrap
+          make -j"$(nproc)"
+          sudo make install
+
       - name: Checkout tfhe-rs
         uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
         with:
           persist-credentials: 'false'
           token: ${{ secrets.FHE_ACTIONS_TOKEN }}
 
-      - name: Setup Hyperstack dependencies
-        uses: ./.github/actions/hyperstack_setup
-        with:
-          cuda-version: ${{ matrix.cuda }}
-          gcc-version: ${{ matrix.gcc }}
-
       - name: Set up home
         run: |
           echo "HOME=/home/ubuntu" >> "${GITHUB_ENV}"
 
       - name: Install latest stable
-        uses: dtolnay/rust-toolchain@a54c7afa936fefeb4456b2dd8068152669aa8203
+        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
@@ -155,7 +183,7 @@ jobs:
     steps:
       - name: Stop instance
         id: stop-instance
-        uses: zama-ai/slab-github-runner@79939325c3c429837c10d6041e4fd8589d328bac
+        uses: zama-ai/slab-github-runner@801df0b8db5ea2b06128b7476c652f5ed5f193a8
         with:
           mode: stop
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}

.github/workflows/gpu_fast_tests.yml

@@ -30,11 +30,10 @@ jobs:
         uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
         with:
           fetch-depth: 0
-          token: ${{ secrets.FHE_ACTIONS_TOKEN }}
 
       - name: Check for file changes
         id: changed-files
-        uses: tj-actions/changed-files@d6e91a2266cdb9d62096cebf1e8546899c6aa18f
+        uses: tj-actions/changed-files@c3a1bb2c992d77180ae65be6ae6c166cf40f857c
         with:
           since_last_remote_commit: true
           files_yaml: |
@@ -43,9 +42,6 @@ jobs:
               - tfhe/build.rs
               - backends/tfhe-cuda-backend/**
               - tfhe/src/core_crypto/gpu/**
-              - tfhe/src/integer/server_key/radix_parallel/tests_unsigned/**
-              - tfhe/src/integer/server_key/radix_parallel/tests_signed/**
-              - tfhe/src/integer/server_key/radix_parallel/tests_cases_unsigned.rs
               - tfhe/src/integer/gpu/**
               - tfhe/src/shortint/parameters/**
               - tfhe/src/high_level_api/**
@@ -66,7 +62,7 @@ jobs:
     steps:
       - name: Start instance
         id: start-instance
-        uses: zama-ai/slab-github-runner@79939325c3c429837c10d6041e4fd8589d328bac
+        uses: zama-ai/slab-github-runner@801df0b8db5ea2b06128b7476c652f5ed5f193a8
         with:
           mode: start
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}
@@ -92,28 +88,60 @@ jobs:
           - os: ubuntu-22.04
             cuda: "12.2"
             gcc: 11 
+    env:
+      CUDA_PATH: /usr/local/cuda-${{ matrix.cuda }}
+      CMAKE_VERSION: 3.29.6
     steps:
+      # Mandatory on hyperstack since a bootable volume is not re-usable yet.
+      - name: Install dependencies
+        run: |
+          sudo apt update
+          sudo apt install -y checkinstall zlib1g-dev libssl-dev libclang-dev
+          wget https://github.com/Kitware/CMake/releases/download/v${{ env.CMAKE_VERSION }}/cmake-${{ env.CMAKE_VERSION }}.tar.gz
+          tar -zxvf cmake-${{ env.CMAKE_VERSION }}.tar.gz
+          cd cmake-${{ env.CMAKE_VERSION }}
+          ./bootstrap
+          make -j"$(nproc)"
+          sudo make install
+
       - name: Checkout tfhe-rs
         uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
         with:
           persist-credentials: 'false'
           token: ${{ secrets.FHE_ACTIONS_TOKEN }}
 
-      - name: Setup Hyperstack dependencies
-        uses: ./.github/actions/hyperstack_setup
-        with:
-          cuda-version: ${{ matrix.cuda }}
-          gcc-version: ${{ matrix.gcc }}
-
       - name: Set up home
         run: |
           echo "HOME=/home/ubuntu" >> "${GITHUB_ENV}"
 
       - name: Install latest stable
-        uses: dtolnay/rust-toolchain@a54c7afa936fefeb4456b2dd8068152669aa8203
+        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: |
           make test_core_crypto_gpu
@@ -153,7 +181,7 @@ jobs:
     steps:
       - name: Stop instance
         id: stop-instance
-        uses: zama-ai/slab-github-runner@79939325c3c429837c10d6041e4fd8589d328bac
+        uses: zama-ai/slab-github-runner@801df0b8db5ea2b06128b7476c652f5ed5f193a8
         with:
           mode: stop
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}

.github/workflows/gpu_full_h100_tests.yml

@@ -25,7 +25,7 @@ jobs:
     steps:
       - name: Start instance
         id: start-instance
-        uses: zama-ai/slab-github-runner@79939325c3c429837c10d6041e4fd8589d328bac
+        uses: zama-ai/slab-github-runner@801df0b8db5ea2b06128b7476c652f5ed5f193a8
         with:
           mode: start
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}
@@ -49,6 +49,9 @@ jobs:
           - os: ubuntu-22.04
             cuda: "12.2"
             gcc: 11 
+    env:
+      CUDA_PATH: /usr/local/cuda-${{ matrix.cuda }}
+      CMAKE_VERSION: 3.29.6
     steps:
       # Mandatory on hyperstack since a bootable volume is not re-usable yet.
       - name: Install dependencies
@@ -68,21 +71,38 @@ jobs:
           persist-credentials: 'false'
           token: ${{ secrets.FHE_ACTIONS_TOKEN }}
 
-      - name: Setup Hyperstack dependencies
-        uses: ./.github/actions/hyperstack_setup
-        with:
-          cuda-version: ${{ matrix.cuda }}
-          gcc-version: ${{ matrix.gcc }}
-
       - name: Set up home
         run: |
           echo "HOME=/home/ubuntu" >> "${GITHUB_ENV}"
 
       - name: Install latest stable
-        uses: dtolnay/rust-toolchain@a54c7afa936fefeb4456b2dd8068152669aa8203
+        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, integer and internal CUDA backend tests
         run: |
           make test_gpu
@@ -119,7 +139,7 @@ jobs:
     steps:
       - name: Stop instance
         id: stop-instance
-        uses: zama-ai/slab-github-runner@79939325c3c429837c10d6041e4fd8589d328bac
+        uses: zama-ai/slab-github-runner@801df0b8db5ea2b06128b7476c652f5ed5f193a8
         with:
           mode: stop
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}

.github/workflows/gpu_full_multi_gpu_tests.yml

@@ -31,11 +31,10 @@ jobs:
         uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
         with:
           fetch-depth: 0
-          token: ${{ secrets.FHE_ACTIONS_TOKEN }}
 
       - name: Check for file changes
         id: changed-files
-        uses: tj-actions/changed-files@d6e91a2266cdb9d62096cebf1e8546899c6aa18f
+        uses: tj-actions/changed-files@c3a1bb2c992d77180ae65be6ae6c166cf40f857c
         with:
           since_last_remote_commit: true
           files_yaml: |
@@ -44,9 +43,6 @@ jobs:
               - tfhe/build.rs
               - backends/tfhe-cuda-backend/**
               - tfhe/src/core_crypto/gpu/**
-              - tfhe/src/integer/server_key/radix_parallel/tests_unsigned/**
-              - tfhe/src/integer/server_key/radix_parallel/tests_signed/**
-              - tfhe/src/integer/server_key/radix_parallel/tests_cases_unsigned.rs
               - tfhe/src/integer/gpu/**
               - tfhe/src/shortint/parameters/**
               - tfhe/src/high_level_api/**
@@ -68,7 +64,7 @@ jobs:
     steps:
       - name: Start instance
         id: start-instance
-        uses: zama-ai/slab-github-runner@79939325c3c429837c10d6041e4fd8589d328bac
+        uses: zama-ai/slab-github-runner@801df0b8db5ea2b06128b7476c652f5ed5f193a8
         with:
           mode: start
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}
@@ -94,28 +90,60 @@ jobs:
           - os: ubuntu-22.04
             cuda: "12.2"
             gcc: 11 
+    env:
+      CUDA_PATH: /usr/local/cuda-${{ matrix.cuda }}
+      CMAKE_VERSION: 3.29.6
     steps:
+      # Mandatory on hyperstack since a bootable volume is not re-usable yet.
+      - name: Install dependencies
+        run: |
+          sudo apt update
+          sudo apt install -y checkinstall zlib1g-dev libssl-dev libclang-dev
+          wget https://github.com/Kitware/CMake/releases/download/v${{ env.CMAKE_VERSION }}/cmake-${{ env.CMAKE_VERSION }}.tar.gz
+          tar -zxvf cmake-${{ env.CMAKE_VERSION }}.tar.gz
+          cd cmake-${{ env.CMAKE_VERSION }}
+          ./bootstrap
+          make -j"$(nproc)"
+          sudo make install
+
       - name: Checkout tfhe-rs
         uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
         with:
           persist-credentials: 'false'
           token: ${{ secrets.FHE_ACTIONS_TOKEN }}
 
-      - name: Setup Hyperstack dependencies
-        uses: ./.github/actions/hyperstack_setup
-        with:
-          cuda-version: ${{ matrix.cuda }}
-          gcc-version: ${{ matrix.gcc }}
-
       - name: Set up home
         run: |
           echo "HOME=/home/ubuntu" >> "${GITHUB_ENV}"
 
       - name: Install latest stable
-        uses: dtolnay/rust-toolchain@a54c7afa936fefeb4456b2dd8068152669aa8203
+        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
@@ -158,7 +186,7 @@ jobs:
     steps:
       - name: Stop instance
         id: stop-instance
-        uses: zama-ai/slab-github-runner@79939325c3c429837c10d6041e4fd8589d328bac
+        uses: zama-ai/slab-github-runner@801df0b8db5ea2b06128b7476c652f5ed5f193a8
         with:
           mode: stop
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}

.github/workflows/gpu_integer_long_run_tests.yml

@@ -1,115 +0,0 @@
-name: Long Run Tests on GPU
-
-env:
-  CARGO_TERM_COLOR: always
-  ACTION_RUN_URL: ${{ github.server_url }}/${{ github.repository }}/actions/runs/${{ github.run_id }}
-  RUSTFLAGS: "-C target-cpu=native"
-  RUST_BACKTRACE: "full"
-  RUST_MIN_STACK: "8388608"
-  SLACK_CHANNEL: ${{ secrets.SLACK_CHANNEL }}
-  SLACK_ICON: https://pbs.twimg.com/profile_images/1274014582265298945/OjBKP9kn_400x400.png
-  SLACK_USERNAME: ${{ secrets.BOT_USERNAME }}
-  SLACK_WEBHOOK: ${{ secrets.SLACK_WEBHOOK }}
-
-on:
-  # Allows you to run this workflow manually from the Actions tab as an alternative.
-  workflow_dispatch:
-  schedule:
-    # Weekly tests will be triggered each Friday at 9p.m.
-    - cron: "0 21 * * 5"
-
-jobs:
-  setup-instance:
-    name: Setup instance (gpu-tests)
-    if: github.event_name != 'schedule' ||
-      (github.event_name == 'schedule' && github.repository == 'zama-ai/tfhe-rs')
-    runs-on: ubuntu-latest
-    outputs:
-      runner-name: ${{ steps.start-instance.outputs.label }}
-    steps:
-      - name: Start instance
-        id: start-instance
-        uses: zama-ai/slab-github-runner@79939325c3c429837c10d6041e4fd8589d328bac
-        with:
-          mode: start
-          github-token: ${{ secrets.SLAB_ACTION_TOKEN }}
-          slab-url: ${{ secrets.SLAB_BASE_URL }}
-          job-secret: ${{ secrets.JOB_SECRET }}
-          backend: hyperstack
-          profile: multi-gpu-test
-
-  cuda-tests:
-    name: Long run GPU tests
-    needs: [ setup-instance ]
-    concurrency:
-      group: ${{ github.workflow }}_${{github.event_name}}_${{ github.ref }}
-      cancel-in-progress: true
-    runs-on: ${{ needs.setup-instance.outputs.runner-name }}
-    strategy:
-      fail-fast: false
-      # explicit include-based build matrix, of known valid options
-      matrix:
-        include:
-          - os: ubuntu-22.04
-            cuda: "12.2"
-            gcc: 11 
-    timeout-minutes: 4320 # 72 hours
-    steps:
-      - name: Checkout tfhe-rs
-        uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
-
-      - name: Setup Hyperstack dependencies
-        uses: ./.github/actions/hyperstack_setup
-        with:
-          cuda-version: ${{ matrix.cuda }}
-          gcc-version: ${{ matrix.gcc }}
-
-      - name: Set up home
-        run: |
-          echo "HOME=/home/ubuntu" >> "${GITHUB_ENV}"
-
-      - name: Install latest stable
-        uses: dtolnay/rust-toolchain@a54c7afa936fefeb4456b2dd8068152669aa8203
-        with:
-          toolchain: stable
-
-      - name: Run tests
-        run: |
-          make test_integer_long_run_gpu
-
-  slack-notify:
-    name: Slack Notification
-    needs: [ setup-instance, cuda-tests ]
-    runs-on: ubuntu-latest
-    if: ${{ always() && needs.cuda-tests.result != 'skipped' && failure() }}
-    continue-on-error: true
-    steps:
-      - name: Send message
-        uses: rtCamp/action-slack-notify@c33737706dea87cd7784c687dadc9adf1be59990
-        env:
-          SLACK_COLOR: ${{ needs.cuda-tests.result }}
-          SLACK_MESSAGE: "Integer GPU long run tests finished with status: ${{ needs.cuda-tests.result }}. (${{ env.ACTION_RUN_URL }})"
-
-  teardown-instance:
-    name: Teardown instance (gpu-tests)
-    if: ${{ always() && needs.setup-instance.result != 'skipped' }}
-    needs: [ setup-instance, cuda-tests ]
-    runs-on: ubuntu-latest
-    steps:
-      - name: Stop instance
-        id: stop-instance
-        uses: zama-ai/slab-github-runner@79939325c3c429837c10d6041e4fd8589d328bac
-        with:
-          mode: stop
-          github-token: ${{ secrets.SLAB_ACTION_TOKEN }}
-          slab-url: ${{ secrets.SLAB_BASE_URL }}
-          job-secret: ${{ secrets.JOB_SECRET }}
-          label: ${{ needs.setup-instance.outputs.runner-name }}
-
-      - name: Slack Notification
-        if: ${{ failure() }}
-        continue-on-error: true
-        uses: rtCamp/action-slack-notify@c33737706dea87cd7784c687dadc9adf1be59990
-        env:
-          SLACK_COLOR: ${{ job.status }}
-          SLACK_MESSAGE: "Instance teardown (gpu-long-run-tests) finished with status: ${{ job.status }}. (${{ env.ACTION_RUN_URL }})"

.github/workflows/gpu_pcc.yml

@@ -24,7 +24,7 @@ jobs:
     steps:
       - name: Start instance
         id: start-instance
-        uses: zama-ai/slab-github-runner@79939325c3c429837c10d6041e4fd8589d328bac
+        uses: zama-ai/slab-github-runner@801df0b8db5ea2b06128b7476c652f5ed5f193a8
         with:
           mode: start
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}
@@ -63,7 +63,7 @@ jobs:
           echo "HOME=/home/ubuntu" >> "${GITHUB_ENV}"
 
       - name: Install latest stable
-        uses: dtolnay/rust-toolchain@a54c7afa936fefeb4456b2dd8068152669aa8203
+        uses: dtolnay/rust-toolchain@7b1c307e0dcbda6122208f10795a713336a9b35a
         with:
           toolchain: stable
 
@@ -110,7 +110,7 @@ jobs:
     steps:
       - name: Stop instance
         id: stop-instance
-        uses: zama-ai/slab-github-runner@79939325c3c429837c10d6041e4fd8589d328bac
+        uses: zama-ai/slab-github-runner@801df0b8db5ea2b06128b7476c652f5ed5f193a8
         with:
           mode: stop
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}

.github/workflows/gpu_signed_integer_classic_tests.yml

@@ -35,7 +35,7 @@ jobs:
 
       - name: Check for file changes
         id: changed-files
-        uses: tj-actions/changed-files@d6e91a2266cdb9d62096cebf1e8546899c6aa18f
+        uses: tj-actions/changed-files@c3a1bb2c992d77180ae65be6ae6c166cf40f857c
         with:
           since_last_remote_commit: true
           files_yaml: |
@@ -44,9 +44,6 @@ jobs:
               - tfhe/build.rs
               - backends/tfhe-cuda-backend/**
               - tfhe/src/core_crypto/gpu/**
-              - tfhe/src/integer/server_key/radix_parallel/tests_unsigned/**
-              - tfhe/src/integer/server_key/radix_parallel/tests_signed/**
-              - tfhe/src/integer/server_key/radix_parallel/tests_cases_unsigned.rs
               - tfhe/src/integer/gpu/**
               - tfhe/src/shortint/parameters/**
               - tfhe/src/high_level_api/**
@@ -68,7 +65,7 @@ jobs:
     steps:
       - name: Start instance
         id: start-instance
-        uses: zama-ai/slab-github-runner@79939325c3c429837c10d6041e4fd8589d328bac
+        uses: zama-ai/slab-github-runner@801df0b8db5ea2b06128b7476c652f5ed5f193a8
         with:
           mode: start
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}
@@ -94,25 +91,58 @@ jobs:
           - os: ubuntu-22.04
             cuda: "12.2"
             gcc: 11 
+    env:
+      CUDA_PATH: /usr/local/cuda-${{ matrix.cuda }}
+      CMAKE_VERSION: 3.29.6
     steps:
+      # Mandatory on hyperstack since a bootable volume is not re-usable yet.
+      - name: Install dependencies
+        run: |
+          sudo apt update
+          sudo apt install -y checkinstall zlib1g-dev libssl-dev libclang-dev
+          wget https://github.com/Kitware/CMake/releases/download/v${{ env.CMAKE_VERSION }}/cmake-${{ env.CMAKE_VERSION }}.tar.gz
+          tar -zxvf cmake-${{ env.CMAKE_VERSION }}.tar.gz
+          cd cmake-${{ env.CMAKE_VERSION }}
+          ./bootstrap
+          make -j"$(nproc)"
+          sudo make install
+
+
       - name: Checkout tfhe-rs
         uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
 
-      - name: Setup Hyperstack dependencies
-        uses: ./.github/actions/hyperstack_setup
-        with:
-          cuda-version: ${{ matrix.cuda }}
-          gcc-version: ${{ matrix.gcc }}
-
       - name: Set up home
         run: |
           echo "HOME=/home/ubuntu" >> "${GITHUB_ENV}"
 
       - name: Install latest stable
-        uses: dtolnay/rust-toolchain@a54c7afa936fefeb4456b2dd8068152669aa8203
+        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
@@ -138,7 +168,7 @@ jobs:
     steps:
       - name: Stop instance
         id: stop-instance
-        uses: zama-ai/slab-github-runner@79939325c3c429837c10d6041e4fd8589d328bac
+        uses: zama-ai/slab-github-runner@801df0b8db5ea2b06128b7476c652f5ed5f193a8
         with:
           mode: stop
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}

.github/workflows/gpu_signed_integer_h100_tests.yml

@@ -35,7 +35,7 @@ jobs:
 
       - name: Check for file changes
         id: changed-files
-        uses: tj-actions/changed-files@d6e91a2266cdb9d62096cebf1e8546899c6aa18f
+        uses: tj-actions/changed-files@c3a1bb2c992d77180ae65be6ae6c166cf40f857c
         with:
           since_last_remote_commit: true
           files_yaml: |
@@ -44,9 +44,6 @@ jobs:
               - tfhe/build.rs
               - backends/tfhe-cuda-backend/**
               - tfhe/src/core_crypto/gpu/**
-              - tfhe/src/integer/server_key/radix_parallel/tests_unsigned/**
-              - tfhe/src/integer/server_key/radix_parallel/tests_signed/**
-              - tfhe/src/integer/server_key/radix_parallel/tests_cases_unsigned.rs
               - tfhe/src/integer/gpu/**
               - tfhe/src/shortint/parameters/**
               - tfhe/src/high_level_api/**
@@ -68,7 +65,7 @@ jobs:
     steps:
       - name: Start instance
         id: start-instance
-        uses: zama-ai/slab-github-runner@79939325c3c429837c10d6041e4fd8589d328bac
+        uses: zama-ai/slab-github-runner@801df0b8db5ea2b06128b7476c652f5ed5f193a8
         with:
           mode: start
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}
@@ -94,25 +91,58 @@ jobs:
           - os: ubuntu-22.04
             cuda: "12.2"
             gcc: 11 
+    env:
+      CUDA_PATH: /usr/local/cuda-${{ matrix.cuda }}
+      CMAKE_VERSION: 3.29.6
     steps:
+      # Mandatory on hyperstack since a bootable volume is not re-usable yet.
+      - name: Install dependencies
+        run: |
+          sudo apt update
+          sudo apt install -y checkinstall zlib1g-dev libssl-dev libclang-dev
+          wget https://github.com/Kitware/CMake/releases/download/v${{ env.CMAKE_VERSION }}/cmake-${{ env.CMAKE_VERSION }}.tar.gz
+          tar -zxvf cmake-${{ env.CMAKE_VERSION }}.tar.gz
+          cd cmake-${{ env.CMAKE_VERSION }}
+          ./bootstrap
+          make -j"$(nproc)"
+          sudo make install
+
+
       - name: Checkout tfhe-rs
         uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
 
-      - name: Setup Hyperstack dependencies
-        uses: ./.github/actions/hyperstack_setup
-        with:
-          cuda-version: ${{ matrix.cuda }}
-          gcc-version: ${{ matrix.gcc }}
-
       - name: Set up home
         run: |
           echo "HOME=/home/ubuntu" >> "${GITHUB_ENV}"
 
       - name: Install latest stable
-        uses: dtolnay/rust-toolchain@a54c7afa936fefeb4456b2dd8068152669aa8203
+        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 multi-bit tests
         run: |
           BIG_TESTS_INSTANCE=TRUE make test_signed_integer_multi_bit_gpu_ci
@@ -138,7 +168,7 @@ jobs:
     steps:
       - name: Stop instance
         id: stop-instance
-        uses: zama-ai/slab-github-runner@79939325c3c429837c10d6041e4fd8589d328bac
+        uses: zama-ai/slab-github-runner@801df0b8db5ea2b06128b7476c652f5ed5f193a8
         with:
           mode: stop
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}

.github/workflows/gpu_signed_integer_tests.yml

@@ -38,11 +38,10 @@ jobs:
         uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
         with:
           fetch-depth: 0
-          token: ${{ secrets.FHE_ACTIONS_TOKEN }}
 
       - name: Check for file changes
         id: changed-files
-        uses: tj-actions/changed-files@d6e91a2266cdb9d62096cebf1e8546899c6aa18f
+        uses: tj-actions/changed-files@c3a1bb2c992d77180ae65be6ae6c166cf40f857c
         with:
           since_last_remote_commit: true
           files_yaml: |
@@ -51,9 +50,6 @@ jobs:
               - tfhe/build.rs
               - backends/tfhe-cuda-backend/**
               - tfhe/src/core_crypto/gpu/**
-              - tfhe/src/integer/server_key/radix_parallel/tests_unsigned/**
-              - tfhe/src/integer/server_key/radix_parallel/tests_signed/**
-              - tfhe/src/integer/server_key/radix_parallel/tests_cases_unsigned.rs
               - tfhe/src/integer/gpu/**
               - tfhe/src/shortint/parameters/**
               - tfhe/src/high_level_api/**
@@ -75,7 +71,7 @@ jobs:
     steps:
       - name: Start instance
         id: start-instance
-        uses: zama-ai/slab-github-runner@79939325c3c429837c10d6041e4fd8589d328bac
+        uses: zama-ai/slab-github-runner@801df0b8db5ea2b06128b7476c652f5ed5f193a8
         with:
           mode: start
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}
@@ -101,28 +97,57 @@ jobs:
           - os: ubuntu-22.04
             cuda: "12.2"
             gcc: 11 
+    env:
+      CUDA_PATH: /usr/local/cuda-${{ matrix.cuda }}
+      CMAKE_VERSION: 3.29.6
     steps:
+      # Mandatory on hyperstack since a bootable volume is not re-usable yet.
+      - name: Install dependencies
+        run: |
+          sudo apt update
+          sudo apt install -y checkinstall zlib1g-dev libssl-dev libclang-dev
+          wget https://github.com/Kitware/CMake/releases/download/v${{ env.CMAKE_VERSION }}/cmake-${{ env.CMAKE_VERSION }}.tar.gz
+          tar -zxvf cmake-${{ env.CMAKE_VERSION }}.tar.gz
+          cd cmake-${{ env.CMAKE_VERSION }}
+          ./bootstrap
+          make -j"$(nproc)"
+          sudo make install
+
+
       - name: Checkout tfhe-rs
         uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
         with:
           persist-credentials: 'false'
           token: ${{ secrets.FHE_ACTIONS_TOKEN }}
 
-      - name: Setup Hyperstack dependencies
-        uses: ./.github/actions/hyperstack_setup
-        with:
-          cuda-version: ${{ matrix.cuda }}
-          gcc-version: ${{ matrix.gcc }}
-
       - name: Set up home
         run: |
           echo "HOME=/home/ubuntu" >> "${GITHUB_ENV}"
 
       - name: Install latest stable
-        uses: dtolnay/rust-toolchain@a54c7afa936fefeb4456b2dd8068152669aa8203
+        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";
+          } >> "${GITHUB_ENV}"
+
       - name: Should run nightly tests
         if: github.event_name == 'schedule'
         run: |
@@ -131,6 +156,10 @@ jobs:
             echo "NIGHTLY_TESTS=TRUE";
           } >> "${GITHUB_ENV}"
 
+      - name: Check device is detected
+        if: ${{ !cancelled() }}
+        run: nvidia-smi
+
       - name: Run signed integer multi-bit tests
         run: |
           make test_signed_integer_multi_bit_gpu_ci
@@ -156,7 +185,7 @@ jobs:
     steps:
       - name: Stop instance
         id: stop-instance
-        uses: zama-ai/slab-github-runner@79939325c3c429837c10d6041e4fd8589d328bac
+        uses: zama-ai/slab-github-runner@801df0b8db5ea2b06128b7476c652f5ed5f193a8
         with:
           mode: stop
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}

.github/workflows/gpu_unsigned_integer_classic_tests.yml

@@ -35,7 +35,7 @@ jobs:
 
       - name: Check for file changes
         id: changed-files
-        uses: tj-actions/changed-files@d6e91a2266cdb9d62096cebf1e8546899c6aa18f
+        uses: tj-actions/changed-files@c3a1bb2c992d77180ae65be6ae6c166cf40f857c
         with:
           since_last_remote_commit: true
           files_yaml: |
@@ -44,9 +44,6 @@ jobs:
               - tfhe/build.rs
               - backends/tfhe-cuda-backend/**
               - tfhe/src/core_crypto/gpu/**
-              - tfhe/src/integer/server_key/radix_parallel/tests_unsigned/**
-              - tfhe/src/integer/server_key/radix_parallel/tests_signed/**
-              - tfhe/src/integer/server_key/radix_parallel/tests_cases_unsigned.rs
               - tfhe/src/integer/gpu/**
               - tfhe/src/shortint/parameters/**
               - tfhe/src/high_level_api/**
@@ -68,7 +65,7 @@ jobs:
     steps:
       - name: Start instance
         id: start-instance
-        uses: zama-ai/slab-github-runner@79939325c3c429837c10d6041e4fd8589d328bac
+        uses: zama-ai/slab-github-runner@801df0b8db5ea2b06128b7476c652f5ed5f193a8
         with:
           mode: start
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}
@@ -94,25 +91,58 @@ jobs:
           - os: ubuntu-22.04
             cuda: "12.2"
             gcc: 11 
+    env:
+      CUDA_PATH: /usr/local/cuda-${{ matrix.cuda }}
+      CMAKE_VERSION: 3.29.6
     steps:
+      # Mandatory on hyperstack since a bootable volume is not re-usable yet.
+      - name: Install dependencies
+        run: |
+          sudo apt update
+          sudo apt install -y checkinstall zlib1g-dev libssl-dev libclang-dev
+          wget https://github.com/Kitware/CMake/releases/download/v${{ env.CMAKE_VERSION }}/cmake-${{ env.CMAKE_VERSION }}.tar.gz
+          tar -zxvf cmake-${{ env.CMAKE_VERSION }}.tar.gz
+          cd cmake-${{ env.CMAKE_VERSION }}
+          ./bootstrap
+          make -j"$(nproc)"
+          sudo make install
+
+
       - name: Checkout tfhe-rs
         uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
 
-      - name: Setup Hyperstack dependencies
-        uses: ./.github/actions/hyperstack_setup
-        with:
-          cuda-version: ${{ matrix.cuda }}
-          gcc-version: ${{ matrix.gcc }}
-
       - name: Set up home
         run: |
           echo "HOME=/home/ubuntu" >> "${GITHUB_ENV}"
 
       - name: Install latest stable
-        uses: dtolnay/rust-toolchain@a54c7afa936fefeb4456b2dd8068152669aa8203
+        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
@@ -138,7 +168,7 @@ jobs:
     steps:
       - name: Stop instance
         id: stop-instance
-        uses: zama-ai/slab-github-runner@79939325c3c429837c10d6041e4fd8589d328bac
+        uses: zama-ai/slab-github-runner@801df0b8db5ea2b06128b7476c652f5ed5f193a8
         with:
           mode: stop
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}

.github/workflows/gpu_unsigned_integer_h100_tests.yml

@@ -35,7 +35,7 @@ jobs:
 
       - name: Check for file changes
         id: changed-files
-        uses: tj-actions/changed-files@d6e91a2266cdb9d62096cebf1e8546899c6aa18f
+        uses: tj-actions/changed-files@c3a1bb2c992d77180ae65be6ae6c166cf40f857c
         with:
           since_last_remote_commit: true
           files_yaml: |
@@ -44,9 +44,6 @@ jobs:
               - tfhe/build.rs
               - backends/tfhe-cuda-backend/**
               - tfhe/src/core_crypto/gpu/**
-              - tfhe/src/integer/server_key/radix_parallel/tests_unsigned/**
-              - tfhe/src/integer/server_key/radix_parallel/tests_signed/**
-              - tfhe/src/integer/server_key/radix_parallel/tests_cases_unsigned.rs
               - tfhe/src/integer/gpu/**
               - tfhe/src/shortint/parameters/**
               - tfhe/src/high_level_api/**
@@ -68,7 +65,7 @@ jobs:
     steps:
       - name: Start instance
         id: start-instance
-        uses: zama-ai/slab-github-runner@79939325c3c429837c10d6041e4fd8589d328bac
+        uses: zama-ai/slab-github-runner@801df0b8db5ea2b06128b7476c652f5ed5f193a8
         with:
           mode: start
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}
@@ -94,25 +91,58 @@ jobs:
           - os: ubuntu-22.04
             cuda: "12.2"
             gcc: 11 
+    env:
+      CUDA_PATH: /usr/local/cuda-${{ matrix.cuda }}
+      CMAKE_VERSION: 3.29.6
     steps:
+      # Mandatory on hyperstack since a bootable volume is not re-usable yet.
+      - name: Install dependencies
+        run: |
+          sudo apt update
+          sudo apt install -y checkinstall zlib1g-dev libssl-dev libclang-dev
+          wget https://github.com/Kitware/CMake/releases/download/v${{ env.CMAKE_VERSION }}/cmake-${{ env.CMAKE_VERSION }}.tar.gz
+          tar -zxvf cmake-${{ env.CMAKE_VERSION }}.tar.gz
+          cd cmake-${{ env.CMAKE_VERSION }}
+          ./bootstrap
+          make -j"$(nproc)"
+          sudo make install
+
+
       - name: Checkout tfhe-rs
         uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
 
-      - name: Setup Hyperstack dependencies
-        uses: ./.github/actions/hyperstack_setup
-        with:
-          cuda-version: ${{ matrix.cuda }}
-          gcc-version: ${{ matrix.gcc }}
-
       - name: Set up home
         run: |
           echo "HOME=/home/ubuntu" >> "${GITHUB_ENV}"
 
       - name: Install latest stable
-        uses: dtolnay/rust-toolchain@a54c7afa936fefeb4456b2dd8068152669aa8203
+        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 multi-bit tests
         run: |
           BIG_TESTS_INSTANCE=TRUE make test_unsigned_integer_multi_bit_gpu_ci
@@ -138,7 +168,7 @@ jobs:
     steps:
       - name: Stop instance
         id: stop-instance
-        uses: zama-ai/slab-github-runner@79939325c3c429837c10d6041e4fd8589d328bac
+        uses: zama-ai/slab-github-runner@801df0b8db5ea2b06128b7476c652f5ed5f193a8
         with:
           mode: stop
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}

.github/workflows/gpu_unsigned_integer_tests.yml

@@ -41,7 +41,7 @@ jobs:
 
       - name: Check for file changes
         id: changed-files
-        uses: tj-actions/changed-files@d6e91a2266cdb9d62096cebf1e8546899c6aa18f
+        uses: tj-actions/changed-files@c3a1bb2c992d77180ae65be6ae6c166cf40f857c
         with:
           since_last_remote_commit: true
           files_yaml: |
@@ -50,9 +50,6 @@ jobs:
               - tfhe/build.rs
               - backends/tfhe-cuda-backend/**
               - tfhe/src/core_crypto/gpu/**
-              - tfhe/src/integer/server_key/radix_parallel/tests_unsigned/**
-              - tfhe/src/integer/server_key/radix_parallel/tests_signed/**
-              - tfhe/src/integer/server_key/radix_parallel/tests_cases_unsigned.rs
               - tfhe/src/integer/gpu/**
               - tfhe/src/shortint/parameters/**
               - tfhe/src/high_level_api/**
@@ -74,7 +71,7 @@ jobs:
     steps:
       - name: Start instance
         id: start-instance
-        uses: zama-ai/slab-github-runner@79939325c3c429837c10d6041e4fd8589d328bac
+        uses: zama-ai/slab-github-runner@801df0b8db5ea2b06128b7476c652f5ed5f193a8
         with:
           mode: start
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}
@@ -100,25 +97,54 @@ jobs:
           - os: ubuntu-22.04
             cuda: "12.2"
             gcc: 11
+    env:
+      CUDA_PATH: /usr/local/cuda-${{ matrix.cuda }}
+      CMAKE_VERSION: 3.29.6
     steps:
+      # Mandatory on hyperstack since a bootable volume is not re-usable yet.
+      - name: Install dependencies
+        run: |
+          sudo apt update
+          sudo apt install -y checkinstall zlib1g-dev libssl-dev libclang-dev
+          wget https://github.com/Kitware/CMake/releases/download/v${{ env.CMAKE_VERSION }}/cmake-${{ env.CMAKE_VERSION }}.tar.gz
+          tar -zxvf cmake-${{ env.CMAKE_VERSION }}.tar.gz
+          cd cmake-${{ env.CMAKE_VERSION }}
+          ./bootstrap
+          make -j"$(nproc)"
+          sudo make install
+
+
       - name: Checkout tfhe-rs
         uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
 
-      - name: Setup Hyperstack dependencies
-        uses: ./.github/actions/hyperstack_setup
-        with:
-          cuda-version: ${{ matrix.cuda }}
-          gcc-version: ${{ matrix.gcc }}
-
       - name: Set up home
         run: |
           echo "HOME=/home/ubuntu" >> "${GITHUB_ENV}"
 
       - name: Install latest stable
-        uses: dtolnay/rust-toolchain@a54c7afa936fefeb4456b2dd8068152669aa8203
+        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";
+          } >> "${GITHUB_ENV}"
+
       - name: Should run nightly tests
         if: github.event_name == 'schedule'
         run: |
@@ -127,6 +153,10 @@ jobs:
             echo "NIGHTLY_TESTS=TRUE";
           } >> "${GITHUB_ENV}"
 
+      - name: Check device is detected
+        if: ${{ !cancelled() }}
+        run: nvidia-smi
+
       - name: Run unsigned integer multi-bit tests
         run: |
           make test_unsigned_integer_multi_bit_gpu_ci
@@ -152,7 +182,7 @@ jobs:
     steps:
       - name: Stop instance
         id: stop-instance
-        uses: zama-ai/slab-github-runner@79939325c3c429837c10d6041e4fd8589d328bac
+        uses: zama-ai/slab-github-runner@801df0b8db5ea2b06128b7476c652f5ed5f193a8
         with:
           mode: stop
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}

.github/workflows/integer_long_run_tests.yml

@@ -1,95 +0,0 @@
-name: AWS Long Run Tests on CPU
-
-env:
-  CARGO_TERM_COLOR: always
-  ACTION_RUN_URL: ${{ github.server_url }}/${{ github.repository }}/actions/runs/${{ github.run_id }}
-  RUSTFLAGS: "-C target-cpu=native"
-  RUST_BACKTRACE: "full"
-  RUST_MIN_STACK: "8388608"
-  SLACK_CHANNEL: ${{ secrets.SLACK_CHANNEL }}
-  SLACK_ICON: https://pbs.twimg.com/profile_images/1274014582265298945/OjBKP9kn_400x400.png
-  SLACK_USERNAME: ${{ secrets.BOT_USERNAME }}
-  SLACK_WEBHOOK: ${{ secrets.SLACK_WEBHOOK }}
-
-on:
-  # Allows you to run this workflow manually from the Actions tab as an alternative.
-  workflow_dispatch:
-  schedule:
-    # Weekly tests will be triggered each Friday at 9p.m.
-    - cron: "0 21 * * 5"
-
-jobs:
-  setup-instance:
-    name: Setup instance (cpu-tests)
-    if: github.event_name != 'schedule' ||
-      (github.event_name == 'schedule' && github.repository == 'zama-ai/tfhe-rs')
-    runs-on: ubuntu-latest
-    outputs:
-      runner-name: ${{ steps.start-instance.outputs.label }}
-    steps:
-      - name: Start instance
-        id: start-instance
-        uses: zama-ai/slab-github-runner@79939325c3c429837c10d6041e4fd8589d328bac
-        with:
-          mode: start
-          github-token: ${{ secrets.SLAB_ACTION_TOKEN }}
-          slab-url: ${{ secrets.SLAB_BASE_URL }}
-          job-secret: ${{ secrets.JOB_SECRET }}
-          backend: aws
-          profile: cpu-big
-
-  cpu-tests:
-    name: Long run CPU tests
-    needs: [ setup-instance ]
-    concurrency:
-      group: ${{ github.workflow }}_${{github.event_name}}_${{ github.ref }}
-      cancel-in-progress: true
-    runs-on: ${{ needs.setup-instance.outputs.runner-name }}
-    timeout-minutes: 4320 # 72 hours
-    steps:
-      - name: Checkout tfhe-rs
-        uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
-        with:
-          persist-credentials: 'false'
-          token: ${{ secrets.FHE_ACTIONS_TOKEN }}
-
-      - name: Install latest stable
-        uses: dtolnay/rust-toolchain@a54c7afa936fefeb4456b2dd8068152669aa8203
-        with:
-          toolchain: stable
-
-      - name: Run tests
-        run: |
-          make test_integer_long_run
-
-      - name: Slack Notification
-        if: ${{ failure() }}
-        continue-on-error: true
-        uses: rtCamp/action-slack-notify@c33737706dea87cd7784c687dadc9adf1be59990
-        env:
-          SLACK_COLOR: ${{ job.status }}
-          SLACK_MESSAGE: "CPU long run tests finished with status: ${{ job.status }}. (${{ env.ACTION_RUN_URL }})"
-
-  teardown-instance:
-    name: Teardown instance (cpu-tests)
-    if: ${{ always() && needs.setup-instance.result != 'skipped' }}
-    needs: [ setup-instance, cpu-tests ]
-    runs-on: ubuntu-latest
-    steps:
-      - name: Stop instance
-        id: stop-instance
-        uses: zama-ai/slab-github-runner@79939325c3c429837c10d6041e4fd8589d328bac
-        with:
-          mode: stop
-          github-token: ${{ secrets.SLAB_ACTION_TOKEN }}
-          slab-url: ${{ secrets.SLAB_BASE_URL }}
-          job-secret: ${{ secrets.JOB_SECRET }}
-          label: ${{ needs.setup-instance.outputs.runner-name }}
-
-      - name: Slack Notification
-        if: ${{ failure() }}
-        continue-on-error: true
-        uses: rtCamp/action-slack-notify@c33737706dea87cd7784c687dadc9adf1be59990
-        env:
-          SLACK_COLOR: ${{ job.status }}
-          SLACK_MESSAGE: "Instance teardown (cpu-long-run-tests) finished with status: ${{ job.status }}. (${{ env.ACTION_RUN_URL }})"

.github/workflows/m1_tests.yml

@@ -3,7 +3,7 @@ name: Tests on M1 CPU
 on:
   workflow_dispatch:
   pull_request:
-    types: [labeled]
+    types: [ labeled ]
   # Have a nightly build for M1 tests
   schedule:
     # * is a special character in YAML so you have to quote this string
@@ -27,7 +27,7 @@ concurrency:
   cancel-in-progress: true
 
 jobs:
-  cargo-builds-m1:
+  cargo-builds:
     if: ${{ (github.event_name == 'schedule' &&  github.repository == 'zama-ai/tfhe-rs') || github.event_name == 'workflow_dispatch' || contains(github.event.label.name, 'm1_test') }}
     runs-on: ["self-hosted", "m1mac"]
     # 12 hours, default is 6 hours, hopefully this is more than enough
@@ -36,57 +36,20 @@ jobs:
     steps:
       - uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
         with:
-          persist-credentials: "false"
+          persist-credentials: 'false'
 
       - name: Install latest stable
-        uses: dtolnay/rust-toolchain@a54c7afa936fefeb4456b2dd8068152669aa8203
+        uses: dtolnay/rust-toolchain@7b1c307e0dcbda6122208f10795a713336a9b35a
         with:
           toolchain: stable
 
-      - name: Run pcc FFT checks
-        run: |
-          make pcc_fft
-
-      - name: Build FFT release
-        run: |
-          make build_fft
-
-      - name: Build FFT release no-std
-        run: |
-          make build_fft_no_std
-
-      - name: Run FFT tests
-        run: |
-          make test_fft
-          make test_fft_serde
-          make test_fft_nightly
-          make test_fft_no_std
-          make test_fft_no_std_nightly
-          # we don't run the js stuff here as it's causing issues with the M1 config
-
-      - name: Run pcc NTT checks
-        run: |
-          make pcc_ntt
-
-      - name: Build NTT release
-        run: |
-          make build_ntt
-
-      - name: Build NTT release no-std
-        run: |
-          make build_ntt_no_std
-
-      - name: Run NTT tests
-        run: |
-          make test_ntt_all
-
       - name: Run pcc checks
         run: |
           make pcc
 
-      - name: Build tfhe-csprng
+      - name: Build concrete-csprng
         run: |
-          make build_tfhe_csprng
+          make build_concrete_csprng
 
       - name: Build Release core
         run: |
@@ -112,9 +75,9 @@ jobs:
         run: |
           make build_c_api
 
-      - name: Run tfhe-csprng tests
+      - name: Run concrete-csprng tests
         run: |
-          make test_tfhe_csprng
+          make test_concrete_csprng
 
       - name: Run tfhe-zk-pok tests
         run: |
@@ -174,7 +137,7 @@ jobs:
     name: Remove m1_test label
     runs-on: ubuntu-latest
     needs:
-      - cargo-builds-m1
+      - cargo-builds
     if: ${{ always() }}
     steps:
       - uses: actions-ecosystem/action-remove-labels@2ce5d41b4b6aa8503e285553f75ed56e0a40bae0
@@ -184,13 +147,13 @@ jobs:
           github_token: ${{ secrets.GITHUB_TOKEN }}
 
       - name: Slack Notification
-        if: ${{ needs.cargo-builds-m1.result != 'skipped' }}
+        if: ${{ needs.cargo-builds.result != 'skipped' }}
         continue-on-error: true
         uses: rtCamp/action-slack-notify@c33737706dea87cd7784c687dadc9adf1be59990
         env:
-          SLACK_COLOR: ${{ needs.cargo-builds-m1.result }}
+          SLACK_COLOR: ${{ needs.cargo-builds.result }}
           SLACK_CHANNEL: ${{ secrets.SLACK_CHANNEL }}
           SLACK_ICON: https://pbs.twimg.com/profile_images/1274014582265298945/OjBKP9kn_400x400.png
-          SLACK_MESSAGE: "M1 tests finished with status: ${{ needs.cargo-builds-m1.result }}. (${{ env.ACTION_RUN_URL }})"
+          SLACK_MESSAGE: "M1 tests finished with status: ${{ needs.cargo-builds.result }}. (${{ env.ACTION_RUN_URL }})"
           SLACK_USERNAME: ${{ secrets.BOT_USERNAME }}
           SLACK_WEBHOOK: ${{ secrets.SLACK_WEBHOOK }}

.github/workflows/make_release.yml

@@ -46,11 +46,10 @@ jobs:
         uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
         with:
           fetch-depth: 0
-          token: ${{ secrets.FHE_ACTIONS_TOKEN }}
       - name: Prepare package
         run: |
           cargo package -p tfhe
-      - uses: actions/upload-artifact@6f51ac03b9356f520e9adb1b1b7802705f340c2b # v4.5.0
+      - uses: actions/upload-artifact@b4b15b8c7c6ac21ea08fcf65892d2ee8f75cf882 # v4.4.3
         with:
           name: crate
           path: target/package/*.crate
@@ -85,7 +84,6 @@ jobs:
         uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
         with:
           fetch-depth: 0
-          token: ${{ secrets.FHE_ACTIONS_TOKEN }}
       - name: Create NPM version tag
         if: ${{ inputs.npm_latest_tag }}
         run: |

.github/workflows/make_release_concrete_csprng.yml

@@ -1,4 +1,4 @@
-name: Publish tfhe-csprng release
+name: Publish concrete-csprng release
 
 on:
   workflow_dispatch:
@@ -19,7 +19,7 @@ jobs:
       READ_ORG_TOKEN: ${{ secrets.READ_ORG_TOKEN }}
 
   publish_release:
-    name: Publish tfhe-csprng Release
+    name: Publish concrete-csprng Release
     needs: verify_tag
     runs-on: ubuntu-latest
     steps:
@@ -27,14 +27,13 @@ jobs:
         uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
         with:
           fetch-depth: 0
-          token: ${{ secrets.FHE_ACTIONS_TOKEN }}
 
       - name: Publish crate.io package
         env:
           CRATES_TOKEN: ${{ secrets.CARGO_REGISTRY_TOKEN }}
           DRY_RUN: ${{ inputs.dry_run && '--dry-run' || '' }}
         run: |
-          cargo publish -p tfhe-csprng --token ${{ env.CRATES_TOKEN }} ${{ env.DRY_RUN }}
+          cargo publish -p concrete-csprng --token ${{ env.CRATES_TOKEN }} ${{ env.DRY_RUN }}
 
       - name: Slack Notification
         if: ${{ failure() }}
@@ -44,6 +43,6 @@ jobs:
           SLACK_COLOR: ${{ job.status }}
           SLACK_CHANNEL: ${{ secrets.SLACK_CHANNEL }}
           SLACK_ICON: https://pbs.twimg.com/profile_images/1274014582265298945/OjBKP9kn_400x400.png
-          SLACK_MESSAGE: "tfhe-csprng release finished with status: ${{ job.status }}. (${{ 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

@@ -36,7 +36,7 @@ jobs:
     steps:
       - name: Start instance
         id: start-instance
-        uses: zama-ai/slab-github-runner@79939325c3c429837c10d6041e4fd8589d328bac
+        uses: zama-ai/slab-github-runner@801df0b8db5ea2b06128b7476c652f5ed5f193a8
         with:
           mode: start
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}
@@ -64,14 +64,13 @@ jobs:
         uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
         with:
           fetch-depth: 0
-          token: ${{ secrets.FHE_ACTIONS_TOKEN }}
 
       - name: Set up home
         run: |
           echo "HOME=/home/ubuntu" >> "${GITHUB_ENV}"
 
       - name: Install latest stable
-        uses: dtolnay/rust-toolchain@a54c7afa936fefeb4456b2dd8068152669aa8203
+        uses: dtolnay/rust-toolchain@7b1c307e0dcbda6122208f10795a713336a9b35a
         with:
           toolchain: stable
 
@@ -120,7 +119,7 @@ jobs:
     steps:
       - name: Stop instance
         id: stop-instance
-        uses: zama-ai/slab-github-runner@79939325c3c429837c10d6041e4fd8589d328bac
+        uses: zama-ai/slab-github-runner@801df0b8db5ea2b06128b7476c652f5ed5f193a8
         with:
           mode: stop
           github-token: ${{ secrets.SLAB_ACTION_TOKEN }}

.github/workflows/make_release_tfhe_fft.yml

@@ -1,49 +0,0 @@
-# Publish new release of tfhe-fft
-name: Publish tfhe-fft release
-
-on:
-  workflow_dispatch:
-    inputs:
-      dry_run:
-        description: "Dry-run"
-        type: boolean
-        default: true
-
-env:
-  ACTION_RUN_URL: ${{ github.server_url }}/${{ github.repository }}/actions/runs/${{ github.run_id }}
-
-jobs:
-  verify_tag:
-    uses: ./.github/workflows/verify_tagged_commit.yml
-    secrets:
-      RELEASE_TEAM: ${{ secrets.RELEASE_TEAM }}
-      READ_ORG_TOKEN: ${{ secrets.READ_ORG_TOKEN }}
-
-  publish_release:
-    name: Publish tfhe-fft Release
-    runs-on: ubuntu-latest
-    needs: verify_tag
-    steps:
-      - name: Checkout
-        uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
-        with:
-          fetch-depth: 0
-
-      - name: Publish crate.io package
-        env:
-          CRATES_TOKEN: ${{ secrets.CARGO_REGISTRY_TOKEN }}
-          DRY_RUN: ${{ inputs.dry_run && '--dry-run' || '' }}
-        run: |
-          cargo publish -p tfhe-fft --token ${{ env.CRATES_TOKEN }} ${{ env.DRY_RUN }}
-
-      - name: Slack Notification
-        if: ${{ failure() }}
-        continue-on-error: true
-        uses: rtCamp/action-slack-notify@c33737706dea87cd7784c687dadc9adf1be59990
-        env:
-          SLACK_COLOR: ${{ job.status }}
-          SLACK_CHANNEL: ${{ secrets.SLACK_CHANNEL }}
-          SLACK_ICON: https://pbs.twimg.com/profile_images/1274014582265298945/OjBKP9kn_400x400.png
-          SLACK_MESSAGE: "tfhe-fft release failed: (${{ env.ACTION_RUN_URL }})"
-          SLACK_USERNAME: ${{ secrets.BOT_USERNAME }}
-          SLACK_WEBHOOK: ${{ secrets.SLACK_WEBHOOK }}

.github/workflows/make_release_tfhe_ntt.yml

@@ -1,49 +0,0 @@
-# Publish new release of tfhe-ntt
-name: Publish tfhe-ntt release
-
-on:
-  workflow_dispatch:
-    inputs:
-      dry_run:
-        description: "Dry-run"
-        type: boolean
-        default: true
-
-env:
-  ACTION_RUN_URL: ${{ github.server_url }}/${{ github.repository }}/actions/runs/${{ github.run_id }}
-
-jobs:
-  verify_tag:
-    uses: ./.github/workflows/verify_tagged_commit.yml
-    secrets:
-      RELEASE_TEAM: ${{ secrets.RELEASE_TEAM }}
-      READ_ORG_TOKEN: ${{ secrets.READ_ORG_TOKEN }}
-
-  publish_release:
-    name: Publish tfhe-ntt Release
-    runs-on: ubuntu-latest
-    needs: verify_tag
-    steps:
-      - name: Checkout
-        uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
-        with:
-          fetch-depth: 0
-
-      - name: Publish crate.io package
-        env:
-          CRATES_TOKEN: ${{ secrets.CARGO_REGISTRY_TOKEN }}
-          DRY_RUN: ${{ inputs.dry_run && '--dry-run' || '' }}
-        run: |
-          cargo publish -p tfhe-ntt --token ${{ env.CRATES_TOKEN }} ${{ env.DRY_RUN }}
-
-      - name: Slack Notification
-        if: ${{ failure() }}
-        continue-on-error: true
-        uses: rtCamp/action-slack-notify@c33737706dea87cd7784c687dadc9adf1be59990
-        env:
-          SLACK_COLOR: ${{ job.status }}
-          SLACK_CHANNEL: ${{ secrets.SLACK_CHANNEL }}
-          SLACK_ICON: https://pbs.twimg.com/profile_images/1274014582265298945/OjBKP9kn_400x400.png
-          SLACK_MESSAGE: "tfhe-ntt release failed: (${{ env.ACTION_RUN_URL }})"
-          SLACK_USERNAME: ${{ secrets.BOT_USERNAME }}
-          SLACK_WEBHOOK: ${{ secrets.SLACK_WEBHOOK }}

.github/workflows/make_release_tfhe_versionable.yml

@@ -27,7 +27,6 @@ jobs:
         uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
         with:
           fetch-depth: 0
-          token: ${{ secrets.FHE_ACTIONS_TOKEN }}
 
       - name: Publish proc-macro crate
         env:

.github/workflows/make_release_zk_pok.yml

@@ -28,7 +28,6 @@ jobs:
         uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
         with:
           fetch-depth: 0
-          token: ${{ secrets.FHE_ACTIONS_TOKEN }}
 
       - name: Publish crate.io package
         env:

.github/workflows/sync_on_push.yml

@@ -16,7 +16,6 @@ jobs:
         uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
         with:
           fetch-depth: 0
-          token: ${{ secrets.FHE_ACTIONS_TOKEN }}
       - name: git-sync
         uses: wei/git-sync@55c6b63b4f21607da0e9877ca9b4d11a29fc6d83
         with:

.gitignore

@@ -13,7 +13,6 @@ target/
 
 # Some of our bench outputs
 /tfhe/benchmarks_parameters
-/tfhe-zk-pok/benchmarks_parameters
 **/*.csv
 
 # dieharder run log
@@ -29,8 +28,6 @@ backends/tfhe-cuda-backend/cuda/cmake-build-debug/
 tfhe/web_wasm_parallel_tests/server.PID
 venv/
 web-test-runner/
-node_modules/
-package-lock.json
 
 # Dir used for backward compatibility test data
 tfhe/tfhe-backward-compat-data/

Cargo.toml

@@ -2,12 +2,10 @@
 resolver = "2"
 members = [
     "tfhe",
-    "tfhe-fft",
-    "tfhe-ntt",
     "tfhe-zk-pok",
     "tasks",
     "apps/trivium",
-    "tfhe-csprng",
+    "concrete-csprng",
     "backends/tfhe-cuda-backend",
     "utils/tfhe-versionable",
     "utils/tfhe-versionable-derive",
@@ -18,17 +16,6 @@ exclude = [
     "utils/cargo-tfhe-lints-inner",
     "utils/cargo-tfhe-lints"
 ]
-[workspace.dependencies]
-aligned-vec = { version = "0.6", default-features = false }
-bytemuck = "1.14.3"
-dyn-stack = { version = "0.11", default-features = false }
-itertools = "0.13"
-num-complex = "0.4"
-pulp = { version = "0.20.0", default-features = false }
-rand = "0.8"
-rayon = "1"
-serde = { version = "1.0", default-features = false }
-wasm-bindgen = ">=0.2.86,<0.2.94"
 
 [profile.bench]
 lto = "fat"

README.md

@@ -70,12 +70,26 @@ production-ready library for all the advanced features of TFHE.
 ### Cargo.toml configuration
 To use the latest version of `TFHE-rs` in your project, you first need to add it as a dependency in your `Cargo.toml`:
 
++ For x86_64-based machines running Unix-like OSes:
+
 ```toml
-tfhe = { version = "*", features = ["boolean", "shortint", "integer"] }
+tfhe = { version = "*", features = ["boolean", "shortint", "integer", "x86_64-unix"] }
+```
+
++ For Apple Silicon or aarch64-based machines running Unix-like OSes:
+
+```toml
+tfhe = { version = "*", features = ["boolean", "shortint", "integer", "aarch64-unix"] }
+```
+
++ For x86_64-based machines with the [`rdseed instruction`](https://en.wikipedia.org/wiki/RDRAND) running Windows:
+
+```toml
+tfhe = { version = "*", features = ["boolean", "shortint", "integer", "x86_64"] }
 ```
 
 > [!Note]
-> Note: You need to use a Rust version >= 1.81 to compile TFHE-rs.
+> Note: You need to use a Rust version >= 1.73 to compile TFHE-rs.
 
 > [!Note]
 > Note: aarch64-based machines are not yet supported for Windows as it's currently missing an entropy source to be able to seed the [CSPRNGs](https://en.wikipedia.org/wiki/Cryptographically_secure_pseudorandom_number_generator) used in TFHE-rs.

apps/trivium/Cargo.toml

@@ -6,8 +6,15 @@ edition = "2021"
 # See more keys and their definitions at https://doc.rust-lang.org/cargo/reference/manifest.html
 
 [dependencies]
-rayon = { workspace = true }
-tfhe = { path = "../../tfhe", features = [ "boolean", "shortint", "integer" ] }
+rayon = { version = "1.7.0"}
+
+[target.'cfg(target_arch = "x86_64")'.dependencies.tfhe]
+path = "../../tfhe"
+features = [ "boolean", "shortint", "integer", "x86_64" ]
+
+[target.'cfg(target_arch = "aarch64")'.dependencies.tfhe]
+path = "../../tfhe"
+features = [ "boolean", "shortint", "integer", "aarch64-unix" ]
 
 [dev-dependencies]
 criterion = { version = "0.5.1", features = [ "html_reports" ]}

backends/tfhe-cuda-backend/Cargo.toml

@@ -1,6 +1,6 @@
 [package]
 name = "tfhe-cuda-backend"
-version = "0.6.0"
+version = "0.5.0"
 edition = "2021"
 authors = ["Zama team"]
 license = "BSD-3-Clause-Clear"

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

@@ -27,23 +27,12 @@ inline void cuda_error(cudaError_t code, const char *file, int line) {
     std::abort();                                                              \
   }
 
-cudaEvent_t cuda_create_event(uint32_t gpu_index);
-
-void cuda_event_record(cudaEvent_t event, cudaStream_t stream,
-                       uint32_t gpu_index);
-void cuda_stream_wait_event(cudaStream_t stream, cudaEvent_t event,
-                            uint32_t gpu_index);
-
-void cuda_event_destroy(cudaEvent_t event, uint32_t gpu_index);
-
 cudaStream_t cuda_create_stream(uint32_t gpu_index);
 
 void cuda_destroy_stream(cudaStream_t stream, uint32_t gpu_index);
 
 void cuda_synchronize_stream(cudaStream_t stream, uint32_t gpu_index);
 
-uint32_t cuda_is_available();
-
 void *cuda_malloc(uint64_t size, uint32_t gpu_index);
 
 void *cuda_malloc_async(uint64_t size, cudaStream_t stream, uint32_t gpu_index);

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

@@ -38,7 +38,6 @@ template <typename Torus> struct int_compression {
 
       scratch_packing_keyswitch_lwe_list_to_glwe_64(
           streams[0], gpu_indexes[0], &fp_ks_buffer,
-          compression_params.small_lwe_dimension,
           compression_params.glwe_dimension, compression_params.polynomial_size,
           num_radix_blocks, true);
     }
@@ -103,12 +102,13 @@ template <typename Torus> struct int_decompression {
       };
 
       generate_device_accumulator<Torus>(
-          streams[0], gpu_indexes[0], carry_extract_lut->get_lut(0, 0),
+          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, 0);
+      carry_extract_lut->broadcast_lut(streams, gpu_indexes, gpu_indexes[0]);
     }
   }
   void release(cudaStream_t const *streams, uint32_t const *gpu_indexes,

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

@@ -35,8 +35,6 @@ enum CMP_ORDERING { IS_INFERIOR = 0, IS_EQUAL = 1, IS_SUPERIOR = 2 };
 
 enum SIGNED_OPERATION { ADDITION = 1, SUBTRACTION = -1 };
 
-enum outputFlag { FLAG_NONE = 0, FLAG_OVERFLOW = 1, FLAG_CARRY = 2 };
-
 extern "C" {
 void scratch_cuda_apply_univariate_lut_kb_64(
     void *const *streams, uint32_t const *gpu_indexes, uint32_t gpu_count,
@@ -46,14 +44,7 @@ void scratch_cuda_apply_univariate_lut_kb_64(
     uint32_t grouping_factor, uint32_t input_lwe_ciphertext_count,
     uint32_t message_modulus, uint32_t carry_modulus, PBS_TYPE pbs_type,
     bool allocate_gpu_memory);
-void scratch_cuda_apply_many_univariate_lut_kb_64(
-    void *const *streams, uint32_t const *gpu_indexes, uint32_t gpu_count,
-    int8_t **mem_ptr, void const *input_lut, uint32_t lwe_dimension,
-    uint32_t glwe_dimension, uint32_t polynomial_size, uint32_t ks_level,
-    uint32_t ks_base_log, uint32_t pbs_level, uint32_t pbs_base_log,
-    uint32_t grouping_factor, uint32_t num_radix_blocks,
-    uint32_t message_modulus, uint32_t carry_modulus, PBS_TYPE pbs_type,
-    uint32_t num_many_lut, bool allocate_gpu_memory);
+
 void cuda_apply_univariate_lut_kb_64(void *const *streams,
                                      uint32_t const *gpu_indexes,
                                      uint32_t gpu_count, void *output_radix_lwe,
@@ -112,19 +103,17 @@ void cleanup_cuda_full_propagation(void *const *streams,
 
 void scratch_cuda_integer_mult_radix_ciphertext_kb_64(
     void *const *streams, uint32_t const *gpu_indexes, uint32_t gpu_count,
-    int8_t **mem_ptr, bool const is_boolean_left, bool const is_boolean_right,
-    uint32_t message_modulus, uint32_t carry_modulus, uint32_t glwe_dimension,
-    uint32_t lwe_dimension, uint32_t polynomial_size, uint32_t pbs_base_log,
-    uint32_t pbs_level, uint32_t ks_base_log, uint32_t ks_level,
-    uint32_t grouping_factor, uint32_t num_blocks, PBS_TYPE pbs_type,
-    bool allocate_gpu_memory);
+    int8_t **mem_ptr, uint32_t message_modulus, uint32_t carry_modulus,
+    uint32_t glwe_dimension, uint32_t lwe_dimension, uint32_t polynomial_size,
+    uint32_t pbs_base_log, uint32_t pbs_level, uint32_t ks_base_log,
+    uint32_t ks_level, uint32_t grouping_factor, uint32_t num_blocks,
+    PBS_TYPE pbs_type, bool allocate_gpu_memory);
 
 void cuda_integer_mult_radix_ciphertext_kb_64(
     void *const *streams, uint32_t const *gpu_indexes, uint32_t gpu_count,
-    void *radix_lwe_out, void const *radix_lwe_left, bool const is_bool_left,
-    void const *radix_lwe_right, bool const is_bool_right, void *const *bsks,
-    void *const *ksks, int8_t *mem_ptr, uint32_t polynomial_size,
-    uint32_t num_blocks);
+    void *radix_lwe_out, void const *radix_lwe_left,
+    void const *radix_lwe_right, void *const *bsks, void *const *ksks,
+    int8_t *mem_ptr, uint32_t polynomial_size, uint32_t num_blocks);
 
 void cleanup_cuda_integer_mult(void *const *streams,
                                uint32_t const *gpu_indexes, uint32_t gpu_count,
@@ -291,61 +280,23 @@ void scratch_cuda_propagate_single_carry_kb_64_inplace(
     uint32_t big_lwe_dimension, uint32_t small_lwe_dimension, uint32_t ks_level,
     uint32_t ks_base_log, uint32_t pbs_level, uint32_t pbs_base_log,
     uint32_t grouping_factor, uint32_t num_blocks, uint32_t message_modulus,
-    uint32_t carry_modulus, PBS_TYPE pbs_type, uint32_t requested_flag,
-    uint32_t uses_carry, bool allocate_gpu_memory);
-
-void scratch_cuda_add_and_propagate_single_carry_kb_64_inplace(
-    void *const *streams, uint32_t const *gpu_indexes, uint32_t gpu_count,
-    int8_t **mem_ptr, uint32_t glwe_dimension, uint32_t polynomial_size,
-    uint32_t big_lwe_dimension, uint32_t small_lwe_dimension, uint32_t ks_level,
-    uint32_t ks_base_log, uint32_t pbs_level, uint32_t pbs_base_log,
-    uint32_t grouping_factor, uint32_t num_blocks, uint32_t message_modulus,
-    uint32_t carry_modulus, PBS_TYPE pbs_type, uint32_t requested_flag,
-    uint32_t uses_carry, bool allocate_gpu_memory);
+    uint32_t carry_modulus, PBS_TYPE pbs_type, bool allocate_gpu_memory);
 
 void cuda_propagate_single_carry_kb_64_inplace(
     void *const *streams, uint32_t const *gpu_indexes, uint32_t gpu_count,
-    void *lwe_array, void *carry_out, const void *carry_in, int8_t *mem_ptr,
-    void *const *bsks, void *const *ksks, uint32_t num_blocks,
-    uint32_t requested_flag, uint32_t uses_carry);
+    void *lwe_array, void *carry_out, int8_t *mem_ptr, void *const *bsks,
+    void *const *ksks, uint32_t num_blocks);
 
-void cuda_add_and_propagate_single_carry_kb_64_inplace(
+void cuda_propagate_single_carry_get_input_carries_kb_64_inplace(
     void *const *streams, uint32_t const *gpu_indexes, uint32_t gpu_count,
-    void *lhs_array, const void *rhs_array, void *carry_out,
-    const void *carry_in, int8_t *mem_ptr, void *const *bsks, void *const *ksks,
-    uint32_t num_blocks, uint32_t requested_flag, uint32_t uses_carry);
+    void *lwe_array, void *carry_out, void *input_carries, int8_t *mem_ptr,
+    void *const *bsks, void *const *ksks, uint32_t num_blocks);
 
 void cleanup_cuda_propagate_single_carry(void *const *streams,
                                          uint32_t const *gpu_indexes,
                                          uint32_t gpu_count,
                                          int8_t **mem_ptr_void);
 
-void cleanup_cuda_add_and_propagate_single_carry(void *const *streams,
-                                                 uint32_t const *gpu_indexes,
-                                                 uint32_t gpu_count,
-                                                 int8_t **mem_ptr_void);
-
-void scratch_cuda_integer_overflowing_sub_kb_64_inplace(
-    void *const *streams, uint32_t const *gpu_indexes, uint32_t gpu_count,
-    int8_t **mem_ptr, uint32_t glwe_dimension, uint32_t polynomial_size,
-    uint32_t big_lwe_dimension, uint32_t small_lwe_dimension, uint32_t ks_level,
-    uint32_t ks_base_log, uint32_t pbs_level, uint32_t pbs_base_log,
-    uint32_t grouping_factor, uint32_t num_blocks, uint32_t message_modulus,
-    uint32_t carry_modulus, PBS_TYPE pbs_type, uint32_t compute_overflow,
-    bool allocate_gpu_memory);
-
-void cuda_integer_overflowing_sub_kb_64_inplace(
-    void *const *streams, uint32_t const *gpu_indexes, uint32_t gpu_count,
-    void *lhs_array, const void *rhs_array, void *overflow_block,
-    const void *input_borrow, int8_t *mem_ptr, void *const *bsks,
-    void *const *ksks, uint32_t num_blocks, uint32_t compute_overflow,
-    uint32_t uses_input_borrow);
-
-void cleanup_cuda_integer_overflowing_sub(void *const *streams,
-                                          uint32_t const *gpu_indexes,
-                                          uint32_t gpu_count,
-                                          int8_t **mem_ptr_void);
-
 void scratch_cuda_integer_radix_partial_sum_ciphertexts_vec_kb_64(
     void *const *streams, uint32_t const *gpu_indexes, uint32_t gpu_count,
     int8_t **mem_ptr, uint32_t glwe_dimension, uint32_t polynomial_size,
@@ -365,6 +316,25 @@ void cleanup_cuda_integer_radix_partial_sum_ciphertexts_vec(
     void *const *streams, uint32_t const *gpu_indexes, uint32_t gpu_count,
     int8_t **mem_ptr_void);
 
+void scratch_cuda_integer_radix_overflowing_sub_kb_64(
+    void *const *streams, uint32_t const *gpu_indexes, uint32_t gpu_count,
+    int8_t **mem_ptr, uint32_t glwe_dimension, uint32_t polynomial_size,
+    uint32_t big_lwe_dimension, uint32_t small_lwe_dimension, uint32_t ks_level,
+    uint32_t ks_base_log, uint32_t pbs_level, uint32_t pbs_base_log,
+    uint32_t grouping_factor, uint32_t num_blocks, uint32_t message_modulus,
+    uint32_t carry_modulus, PBS_TYPE pbs_type, bool allocate_gpu_memory);
+
+void cuda_integer_radix_overflowing_sub_kb_64(
+    void *const *streams, uint32_t const *gpu_indexes, uint32_t gpu_count,
+    void *radix_lwe_out, void *radix_lwe_overflowed, void const *radix_lwe_left,
+    void const *radix_lwe_right, int8_t *mem_ptr, void *const *bsks,
+    void *const *ksks, uint32_t num_blocks_in_radix);
+
+void cleanup_cuda_integer_radix_overflowing_sub(void *const *streams,
+                                                uint32_t const *gpu_indexes,
+                                                uint32_t gpu_count,
+                                                int8_t **mem_ptr_void);
+
 void scratch_cuda_integer_scalar_mul_kb_64(
     void *const *streams, uint32_t const *gpu_indexes, uint32_t gpu_count,
     int8_t **mem_ptr, uint32_t glwe_dimension, uint32_t polynomial_size,
@@ -387,23 +357,42 @@ void cleanup_cuda_integer_radix_scalar_mul(void *const *streams,
 
 void scratch_cuda_integer_div_rem_radix_ciphertext_kb_64(
     void *const *streams, uint32_t const *gpu_indexes, uint32_t gpu_count,
-    bool is_signed, int8_t **mem_ptr, uint32_t glwe_dimension,
-    uint32_t polynomial_size, uint32_t big_lwe_dimension,
-    uint32_t small_lwe_dimension, uint32_t ks_level, uint32_t ks_base_log,
-    uint32_t pbs_level, uint32_t pbs_base_log, uint32_t grouping_factor,
-    uint32_t num_blocks, uint32_t message_modulus, uint32_t carry_modulus,
-    PBS_TYPE pbs_type, bool allocate_gpu_memory);
+    int8_t **mem_ptr, uint32_t glwe_dimension, uint32_t polynomial_size,
+    uint32_t big_lwe_dimension, uint32_t small_lwe_dimension, uint32_t ks_level,
+    uint32_t ks_base_log, uint32_t pbs_level, uint32_t pbs_base_log,
+    uint32_t grouping_factor, uint32_t num_blocks, uint32_t message_modulus,
+    uint32_t carry_modulus, PBS_TYPE pbs_type, bool allocate_gpu_memory);
 
 void cuda_integer_div_rem_radix_ciphertext_kb_64(
     void *const *streams, uint32_t const *gpu_indexes, uint32_t gpu_count,
     void *quotient, void *remainder, void const *numerator, void const *divisor,
-    bool is_signed, int8_t *mem_ptr, void *const *bsks, void *const *ksks,
+    int8_t *mem_ptr, void *const *bsks, void *const *ksks,
     uint32_t num_blocks_in_radix);
 
 void cleanup_cuda_integer_div_rem(void *const *streams,
                                   uint32_t const *gpu_indexes,
                                   uint32_t gpu_count, int8_t **mem_ptr_void);
 
+void scratch_cuda_signed_overflowing_add_or_sub_radix_ciphertext_kb_64(
+    void *const *streams, uint32_t const *gpu_indexes, uint32_t gpu_count,
+    int8_t **mem_ptr, uint32_t glwe_dimension, uint32_t polynomial_size,
+    uint32_t big_lwe_dimension, uint32_t small_lwe_dimension, uint32_t ks_level,
+    uint32_t ks_base_log, uint32_t pbs_level, uint32_t pbs_base_log,
+    uint32_t grouping_factor, uint32_t num_blocks, int8_t signed_operation,
+    uint32_t message_modulus, uint32_t carry_modulus, PBS_TYPE pbs_type,
+    bool allocate_gpu_memory);
+
+void cuda_signed_overflowing_add_or_sub_radix_ciphertext_kb_64(
+    void *const *streams, uint32_t const *gpu_indexes, uint32_t gpu_count,
+    void *lhs, void const *rhs, void *overflowed, int8_t signed_operation,
+    int8_t *mem_ptr, void *const *bsks, void *const *ksks,
+    uint32_t num_blocks_in_radix);
+
+void cleanup_signed_overflowing_add_or_sub(void *const *streams,
+                                           uint32_t const *gpu_indexes,
+                                           uint32_t gpu_count,
+                                           int8_t **mem_ptr_void);
+
 void scratch_cuda_integer_compute_prefix_sum_hillis_steele_64(
     void *const *streams, uint32_t const *gpu_indexes, uint32_t gpu_count,
     int8_t **mem_ptr, void const *input_lut, uint32_t lwe_dimension,
@@ -428,60 +417,5 @@ void cuda_integer_reverse_blocks_64_inplace(void *const *streams,
                                             uint32_t num_blocks,
                                             uint32_t lwe_size);
 
-void scratch_cuda_integer_abs_inplace_radix_ciphertext_kb_64(
-    void *const *streams, uint32_t const *gpu_indexes, uint32_t gpu_count,
-    int8_t **mem_ptr, bool is_signed, uint32_t glwe_dimension,
-    uint32_t polynomial_size, uint32_t big_lwe_dimension,
-    uint32_t small_lwe_dimension, uint32_t ks_level, uint32_t ks_base_log,
-    uint32_t pbs_level, uint32_t pbs_base_log, uint32_t grouping_factor,
-    uint32_t num_blocks, uint32_t message_modulus, uint32_t carry_modulus,
-    PBS_TYPE pbs_type, bool allocate_gpu_memory);
-
-void cuda_integer_abs_inplace_radix_ciphertext_kb_64(
-    void *const *streams, uint32_t const *gpu_indexes, uint32_t gpu_count,
-    void *ct, int8_t *mem_ptr, bool is_signed, void *const *bsks,
-    void *const *ksks, uint32_t num_blocks);
-
-void cleanup_cuda_integer_abs_inplace(void *const *streams,
-                                      uint32_t const *gpu_indexes,
-                                      uint32_t gpu_count,
-                                      int8_t **mem_ptr_void);
-
-void scratch_cuda_integer_are_all_comparisons_block_true_kb_64(
-    void *const *streams, uint32_t const *gpu_indexes, uint32_t gpu_count,
-    int8_t **mem_ptr, uint32_t glwe_dimension, uint32_t polynomial_size,
-    uint32_t big_lwe_dimension, uint32_t small_lwe_dimension, uint32_t ks_level,
-    uint32_t ks_base_log, uint32_t pbs_level, uint32_t pbs_base_log,
-    uint32_t grouping_factor, uint32_t num_radix_blocks,
-    uint32_t message_modulus, uint32_t carry_modulus, PBS_TYPE pbs_type,
-    bool allocate_gpu_memory);
-
-void cuda_integer_are_all_comparisons_block_true_kb_64(
-    void *const *streams, uint32_t const *gpu_indexes, uint32_t gpu_count,
-    void *lwe_array_out, void const *lwe_array_in, int8_t *mem_ptr,
-    void *const *bsks, void *const *ksks, uint32_t num_radix_blocks);
-
-void cleanup_cuda_integer_are_all_comparisons_block_true(
-    void *const *streams, uint32_t const *gpu_indexes, uint32_t gpu_count,
-    int8_t **mem_ptr_void);
-
-void scratch_cuda_integer_is_at_least_one_comparisons_block_true_kb_64(
-    void *const *streams, uint32_t const *gpu_indexes, uint32_t gpu_count,
-    int8_t **mem_ptr, uint32_t glwe_dimension, uint32_t polynomial_size,
-    uint32_t big_lwe_dimension, uint32_t small_lwe_dimension, uint32_t ks_level,
-    uint32_t ks_base_log, uint32_t pbs_level, uint32_t pbs_base_log,
-    uint32_t grouping_factor, uint32_t num_radix_blocks,
-    uint32_t message_modulus, uint32_t carry_modulus, PBS_TYPE pbs_type,
-    bool allocate_gpu_memory);
-
-void cuda_integer_is_at_least_one_comparisons_block_true_kb_64(
-    void *const *streams, uint32_t const *gpu_indexes, uint32_t gpu_count,
-    void *lwe_array_out, void const *lwe_array_in, int8_t *mem_ptr,
-    void *const *bsks, void *const *ksks, uint32_t num_radix_blocks);
-
-void cleanup_cuda_integer_is_at_least_one_comparisons_block_true(
-    void *const *streams, uint32_t const *gpu_indexes, uint32_t gpu_count,
-    int8_t **mem_ptr_void);
-
 } // extern C
 #endif // CUDA_INTEGER_H

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

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

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

@@ -27,7 +27,6 @@ void cuda_add_lwe_ciphertext_vector_64(void *stream, uint32_t gpu_index,
                                        void const *lwe_array_in_2,
                                        uint32_t input_lwe_dimension,
                                        uint32_t input_lwe_ciphertext_count);
-
 void cuda_add_lwe_ciphertext_vector_plaintext_vector_32(
     void *stream, uint32_t gpu_index, void *lwe_array_out,
     void const *lwe_array_in, void const *plaintext_array_in,

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

@@ -28,7 +28,7 @@ void cuda_tbc_multi_bit_programmable_bootstrap_lwe_ciphertext_vector(
     pbs_buffer<Torus, MULTI_BIT> *pbs_buffer, uint32_t lwe_dimension,
     uint32_t glwe_dimension, uint32_t polynomial_size, uint32_t grouping_factor,
     uint32_t base_log, uint32_t level_count, uint32_t num_samples,
-    uint32_t num_many_lut, uint32_t lut_stride);
+    uint32_t lut_count, uint32_t lut_stride);
 #endif
 
 template <typename Torus>
@@ -46,7 +46,7 @@ void cuda_cg_multi_bit_programmable_bootstrap_lwe_ciphertext_vector(
     pbs_buffer<Torus, MULTI_BIT> *pbs_buffer, uint32_t lwe_dimension,
     uint32_t glwe_dimension, uint32_t polynomial_size, uint32_t grouping_factor,
     uint32_t base_log, uint32_t level_count, uint32_t num_samples,
-    uint32_t num_many_lut, uint32_t lut_stride);
+    uint32_t lut_count, uint32_t lut_stride);
 
 template <typename Torus>
 void scratch_cuda_multi_bit_programmable_bootstrap(
@@ -63,7 +63,7 @@ void cuda_multi_bit_programmable_bootstrap_lwe_ciphertext_vector(
     pbs_buffer<Torus, MULTI_BIT> *pbs_buffer, uint32_t lwe_dimension,
     uint32_t glwe_dimension, uint32_t polynomial_size, uint32_t grouping_factor,
     uint32_t base_log, uint32_t level_count, uint32_t num_samples,
-    uint32_t num_many_lut, uint32_t lut_stride);
+    uint32_t lut_count, uint32_t lut_stride);
 
 template <typename Torus>
 uint64_t get_buffer_size_full_sm_multibit_programmable_bootstrap_keybundle(
@@ -106,7 +106,7 @@ template <typename Torus> struct pbs_buffer<Torus, PBS_TYPE::MULTI_BIT> {
   uint32_t lwe_chunk_size;
   double2 *keybundle_fft;
   Torus *global_accumulator;
-  double2 *global_join_buffer;
+  double2 *global_accumulator_fft;
 
   PBS_VARIANT pbs_variant;
 
@@ -225,12 +225,10 @@ template <typename Torus> struct pbs_buffer<Torus, PBS_TYPE::MULTI_BIT> {
           num_blocks_keybundle * (polynomial_size / 2) * sizeof(double2),
           stream, gpu_index);
       global_accumulator = (Torus *)cuda_malloc_async(
-          input_lwe_ciphertext_count * (glwe_dimension + 1) * polynomial_size *
-              sizeof(Torus),
-          stream, gpu_index);
-      global_join_buffer = (double2 *)cuda_malloc_async(
-          level_count * (glwe_dimension + 1) * input_lwe_ciphertext_count *
-              (polynomial_size / 2) * sizeof(double2),
+          num_blocks_acc_step_one * polynomial_size * sizeof(Torus), stream,
+          gpu_index);
+      global_accumulator_fft = (double2 *)cuda_malloc_async(
+          num_blocks_acc_step_one * (polynomial_size / 2) * sizeof(double2),
           stream, gpu_index);
     }
   }
@@ -262,7 +260,7 @@ template <typename Torus> struct pbs_buffer<Torus, PBS_TYPE::MULTI_BIT> {
 
     cuda_drop_async(keybundle_fft, stream, gpu_index);
     cuda_drop_async(global_accumulator, stream, gpu_index);
-    cuda_drop_async(global_join_buffer, stream, gpu_index);
+    cuda_drop_async(global_accumulator_fft, stream, gpu_index);
   }
 };
 

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

@@ -69,7 +69,7 @@ template <typename Torus> struct pbs_buffer<Torus, PBS_TYPE::CLASSICAL> {
   int8_t *d_mem;
 
   Torus *global_accumulator;
-  double2 *global_join_buffer;
+  double2 *global_accumulator_fft;
 
   PBS_VARIANT pbs_variant;
 
@@ -114,7 +114,7 @@ template <typename Torus> struct pbs_buffer<Torus, PBS_TYPE::CLASSICAL> {
         // Otherwise, both kernels run all in shared memory
         d_mem = (int8_t *)cuda_malloc_async(device_mem, stream, gpu_index);
 
-        global_join_buffer = (double2 *)cuda_malloc_async(
+        global_accumulator_fft = (double2 *)cuda_malloc_async(
             (glwe_dimension + 1) * level_count * input_lwe_ciphertext_count *
                 (polynomial_size / 2) * sizeof(double2),
             stream, gpu_index);
@@ -147,7 +147,7 @@ template <typename Torus> struct pbs_buffer<Torus, PBS_TYPE::CLASSICAL> {
         // Otherwise, both kernels run all in shared memory
         d_mem = (int8_t *)cuda_malloc_async(device_mem, stream, gpu_index);
 
-        global_join_buffer = (double2 *)cuda_malloc_async(
+        global_accumulator_fft = (double2 *)cuda_malloc_async(
             (glwe_dimension + 1) * level_count * input_lwe_ciphertext_count *
                 polynomial_size / 2 * sizeof(double2),
             stream, gpu_index);
@@ -194,7 +194,7 @@ template <typename Torus> struct pbs_buffer<Torus, PBS_TYPE::CLASSICAL> {
         // Otherwise, both kernels run all in shared memory
         d_mem = (int8_t *)cuda_malloc_async(device_mem, stream, gpu_index);
 
-        global_join_buffer = (double2 *)cuda_malloc_async(
+        global_accumulator_fft = (double2 *)cuda_malloc_async(
             (glwe_dimension + 1) * level_count * input_lwe_ciphertext_count *
                 polynomial_size / 2 * sizeof(double2),
             stream, gpu_index);
@@ -208,7 +208,7 @@ template <typename Torus> struct pbs_buffer<Torus, PBS_TYPE::CLASSICAL> {
 
   void release(cudaStream_t stream, uint32_t gpu_index) {
     cuda_drop_async(d_mem, stream, gpu_index);
-    cuda_drop_async(global_join_buffer, stream, gpu_index);
+    cuda_drop_async(global_accumulator_fft, stream, gpu_index);
 
     if (pbs_variant == DEFAULT)
       cuda_drop_async(global_accumulator, stream, gpu_index);
@@ -255,7 +255,7 @@ void cuda_programmable_bootstrap_cg_lwe_ciphertext_vector(
     Torus const *lwe_input_indexes, double2 const *bootstrapping_key,
     pbs_buffer<Torus, CLASSICAL> *buffer, uint32_t lwe_dimension,
     uint32_t glwe_dimension, uint32_t polynomial_size, uint32_t base_log,
-    uint32_t level_count, uint32_t num_samples, uint32_t num_many_lut,
+    uint32_t level_count, uint32_t num_samples, uint32_t lut_count,
     uint32_t lut_stride);
 
 template <typename Torus>
@@ -266,7 +266,7 @@ void cuda_programmable_bootstrap_lwe_ciphertext_vector(
     Torus const *lwe_input_indexes, double2 const *bootstrapping_key,
     pbs_buffer<Torus, CLASSICAL> *buffer, uint32_t lwe_dimension,
     uint32_t glwe_dimension, uint32_t polynomial_size, uint32_t base_log,
-    uint32_t level_count, uint32_t num_samples, uint32_t num_many_lut,
+    uint32_t level_count, uint32_t num_samples, uint32_t lut_count,
     uint32_t lut_stride);
 
 #if (CUDA_ARCH >= 900)
@@ -278,7 +278,7 @@ void cuda_programmable_bootstrap_tbc_lwe_ciphertext_vector(
     Torus const *lwe_input_indexes, double2 const *bootstrapping_key,
     pbs_buffer<Torus, CLASSICAL> *buffer, uint32_t lwe_dimension,
     uint32_t glwe_dimension, uint32_t polynomial_size, uint32_t base_log,
-    uint32_t level_count, uint32_t num_samples, uint32_t num_many_lut,
+    uint32_t level_count, uint32_t num_samples, uint32_t lut_count,
     uint32_t lut_stride);
 
 template <typename Torus>

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

@@ -69,7 +69,7 @@ void cuda_programmable_bootstrap_lwe_ciphertext_vector_32(
     void const *lwe_input_indexes, void const *bootstrapping_key,
     int8_t *buffer, uint32_t lwe_dimension, uint32_t glwe_dimension,
     uint32_t polynomial_size, uint32_t base_log, uint32_t level_count,
-    uint32_t num_samples, uint32_t num_many_lut, uint32_t lut_stride);
+    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,
@@ -78,7 +78,7 @@ void cuda_programmable_bootstrap_lwe_ciphertext_vector_64(
     void const *lwe_input_indexes, void const *bootstrapping_key,
     int8_t *buffer, uint32_t lwe_dimension, uint32_t glwe_dimension,
     uint32_t polynomial_size, uint32_t base_log, uint32_t level_count,
-    uint32_t num_samples, uint32_t num_many_lut, uint32_t lut_stride);
+    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);

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

@@ -27,7 +27,7 @@ void cuda_multi_bit_programmable_bootstrap_lwe_ciphertext_vector_64(
     void const *lwe_input_indexes, void const *bootstrapping_key,
     int8_t *buffer, uint32_t lwe_dimension, uint32_t glwe_dimension,
     uint32_t polynomial_size, uint32_t grouping_factor, uint32_t base_log,
-    uint32_t level_count, uint32_t num_samples, uint32_t num_many_lut,
+    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,

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

@@ -1,358 +0,0 @@
-#ifndef CNCRT_FAST_KS_CUH
-#define CNCRT_FAST_KS_CUH
-
-#undef NDEBUG
-#include <assert.h>
-
-#include "device.h"
-#include "gadget.cuh"
-#include "helper_multi_gpu.h"
-#include "keyswitch.cuh"
-#include "polynomial/functions.cuh"
-#include "polynomial/polynomial_math.cuh"
-#include "torus.cuh"
-#include "utils/helper.cuh"
-#include "utils/kernel_dimensions.cuh"
-#include <thread>
-#include <vector>
-
-#define CEIL_DIV(M, N) ((M) + (N)-1) / (N)
-
-const int BLOCK_SIZE_GEMM = 64;
-const int THREADS_GEMM = 8;
-const int BLOCK_SIZE_DECOMP = 8;
-
-template <typename Torus> uint64_t get_shared_mem_size_tgemm() {
-  return BLOCK_SIZE_GEMM * THREADS_GEMM * 2 * sizeof(Torus);
-}
-
-__host__ inline bool can_use_pks_fast_path(uint32_t lwe_dimension,
-                                           uint32_t num_lwe,
-                                           uint32_t polynomial_size,
-                                           uint32_t level_count,
-                                           uint32_t glwe_dimension) {
-  // TODO: activate it back, fix tests and extend to level_count > 1
-  return false;
-}
-
-// Initialize decomposition by performing rounding
-// and decomposing one level of an array of Torus LWEs. Only
-// decomposes the mask elements of the incoming LWEs.
-template <typename Torus, typename TorusVec>
-__global__ void decompose_vectorize_init(Torus const *lwe_in, Torus *lwe_out,
-                                         uint32_t lwe_dimension,
-                                         uint32_t num_lwe, uint32_t base_log,
-                                         uint32_t level_count) {
-
-  // index of this LWE ct in the buffer
-  auto lwe_idx = blockIdx.x * blockDim.x + threadIdx.x;
-  // index of the LWE sample in the LWE ct
-  auto lwe_sample_idx = blockIdx.y * blockDim.y + threadIdx.y;
-
-  if (lwe_idx >= num_lwe || lwe_sample_idx >= lwe_dimension)
-    return;
-
-  // Input LWE array is [mask_0, .., mask_lwe_dim, message] and
-  // we only decompose the mask. Thus the stride for reading
-  // is lwe_dimension + 1, while for writing it is lwe_dimension
-  auto read_val_idx = lwe_idx * (lwe_dimension + 1) + lwe_sample_idx;
-  auto write_val_idx = lwe_idx * lwe_dimension + lwe_sample_idx;
-
-  Torus a_i = lwe_in[read_val_idx];
-
-  Torus state = init_decomposer_state(a_i, base_log, level_count);
-
-  Torus mod_b_mask = (1ll << base_log) - 1ll;
-  lwe_out[write_val_idx] = decompose_one<Torus>(state, mod_b_mask, base_log);
-}
-
-// Continue decomposiion of an array of Torus elements in place. Supposes
-// that the array contains already decomposed elements and
-// computes the new decomposed level in place.
-template <typename Torus, typename TorusVec>
-__global__ void
-decompose_vectorize_step_inplace(Torus *buffer_in, uint32_t lwe_dimension,
-                                 uint32_t num_lwe, uint32_t base_log,
-                                 uint32_t level_count) {
-
-  // index of this LWE ct in the buffer
-  auto lwe_idx = blockIdx.x * blockDim.x + threadIdx.x;
-  // index of the LWE sample in the LWE ct
-  auto lwe_sample_idx = blockIdx.y * blockDim.y + threadIdx.y;
-
-  if (lwe_idx >= num_lwe || lwe_sample_idx >= lwe_dimension)
-    return;
-
-  auto val_idx = lwe_idx * lwe_dimension + lwe_sample_idx;
-
-  Torus state = buffer_in[val_idx];
-
-  Torus mod_b_mask = (1ll << base_log) - 1ll;
-
-  buffer_in[val_idx] = decompose_one<Torus>(state, mod_b_mask, base_log);
-}
-
-// Multiply matrices A, B of size (M, K), (K, N) respectively
-// with K as the inner dimension.
-//
-// A block of threads processeds blocks of size (BLOCK_SIZE_GEMM,
-// BLOCK_SIZE_GEMM) splitting them in multiple tiles: (BLOCK_SIZE_GEMM,
-// THREADS_GEMM)-shaped tiles of values from A, and a (THREADS_GEMM,
-// BLOCK_SIZE_GEMM)-shaped tiles of values from B.
-template <typename Torus, typename TorusVec>
-__global__ void tgemm(int M, int N, int K, const Torus *A, const Torus *B,
-                      int stride_B, Torus *C) {
-
-  const int BM = BLOCK_SIZE_GEMM;
-  const int BN = BLOCK_SIZE_GEMM;
-  const int BK = THREADS_GEMM;
-  const int TM = THREADS_GEMM;
-
-  const uint cRow = blockIdx.y;
-  const uint cCol = blockIdx.x;
-
-  const uint totalResultsBlocktile = BM * BN;
-  const int threadCol = threadIdx.x % BN;
-  const int threadRow = threadIdx.x / BN;
-
-  // Allocate space for the current block tile in shared memory
-  __shared__ Torus As[BM * BK];
-  __shared__ Torus Bs[BK * BN];
-
-  // Initialize the pointers to the input blocks from A, B
-  // Tiles from these blocks are loaded to shared memory
-  A += cRow * BM * K;
-  B += cCol * BN;
-
-  // Each thread will handle multiple sub-blocks
-  const uint innerColA = threadIdx.x % BK;
-  const uint innerRowA = threadIdx.x / BK;
-  const uint innerColB = threadIdx.x % BN;
-  const uint innerRowB = threadIdx.x / BN;
-
-  // allocate thread-local cache for results in registerfile
-  Torus threadResults[TM] = {0};
-
-  auto row_A = cRow * BM + innerRowA;
-  auto col_B = cCol * BN + innerColB;
-
-  // For each thread, loop over block tiles
-  for (uint bkIdx = 0; bkIdx < K; bkIdx += BK) {
-    auto col_A = bkIdx + innerColA;
-    auto row_B = bkIdx + innerRowB;
-
-    if (row_A < M && col_A < K) {
-      As[innerRowA * BK + innerColA] = A[innerRowA * K + innerColA];
-    } else {
-      As[innerRowA * BK + innerColA] = 0;
-    }
-
-    if (col_B < N && row_B < K) {
-      Bs[innerRowB * BN + innerColB] = B[innerRowB * stride_B + innerColB];
-    } else {
-      Bs[innerRowB * BN + innerColB] = 0;
-    }
-    __syncthreads();
-
-    // Advance blocktile for the next iteration of this loop
-    A += BK;
-    B += BK * stride_B;
-
-    // calculate per-thread results
-    for (uint dotIdx = 0; dotIdx < BK; ++dotIdx) {
-      // we make the dotproduct loop the outside loop, which facilitates
-      // reuse of the Bs entry, which we can cache in a tmp var.
-      Torus tmp = Bs[dotIdx * BN + threadCol];
-      for (uint resIdx = 0; resIdx < TM; ++resIdx) {
-        threadResults[resIdx] +=
-            As[(threadRow * TM + resIdx) * BK + dotIdx] * tmp;
-      }
-    }
-    __syncthreads();
-  }
-
-  // Initialize the pointer to the output block of size (BLOCK_SIZE_GEMM,
-  // BLOCK_SIZE_GEMM)
-  C += cRow * BM * N + cCol * BN;
-
-  // write out the results
-  for (uint resIdx = 0; resIdx < TM; ++resIdx) {
-    int outRow = cRow * BM + threadRow * TM + resIdx;
-    int outCol = cCol * BN + threadCol;
-
-    if (outRow >= M)
-      continue;
-    if (outCol >= N)
-      continue;
-
-    C[(threadRow * TM + resIdx) * N + threadCol] += threadResults[resIdx];
-  }
-}
-
-// Finish the keyswitching operation and prepare GLWEs for accumulation.
-// 1. Finish the keyswitching computation partially performed with a GEMM:
-//  - negate the dot product between the GLWE and KSK polynomial
-//  - add the GLWE message for the N-th polynomial coeff in the message poly
-// 2. Rotate each of the GLWE . KSK poly dot products to
-//    prepare them for accumulation into a single GLWE
-template <typename Torus>
-__global__ void polynomial_accumulate_monic_monomial_mul_many_neg_and_add_C(
-    Torus *in_glwe_buffer, Torus *out_glwe_buffer, Torus const *lwe_array,
-    uint32_t lwe_dimension, uint32_t num_glwes, uint32_t polynomial_size,
-    uint32_t glwe_dimension) {
-
-  uint32_t glwe_id = blockIdx.x * blockDim.x + threadIdx.x;
-  uint32_t degree = glwe_id; // lwe 0 rotate 0, lwe 1 rotate 1, .. , lwe
-                             // poly_size-1 rotate poly_size-1
-  uint32_t coeffIdx = blockIdx.y * blockDim.y + threadIdx.y;
-
-  if (glwe_id >= num_glwes)
-    return;
-  if (coeffIdx >= polynomial_size)
-    return;
-
-  auto in_poly =
-      in_glwe_buffer + glwe_id * polynomial_size * (glwe_dimension + 1);
-  auto out_result =
-      out_glwe_buffer + glwe_id * polynomial_size * (glwe_dimension + 1);
-  if (coeffIdx == 0) {
-    // Add the message value of the input LWE (`C`) to the N-th coefficient
-    // in the GLWE . KSK dot product
-
-    // The C is added to the first position of the last polynomial in the GLWE
-    // which has (glwe_dimension+1) polynomials
-    // The C value is extracted as the last value of the LWE ct. (of index
-    // glwe_id) the LWEs have (polynomial_size + 1) values
-    in_poly[polynomial_size * glwe_dimension] =
-        lwe_array[glwe_id * (lwe_dimension + 1) + lwe_dimension] -
-        in_poly[polynomial_size * glwe_dimension];
-
-    for (int gi = 1; gi < glwe_dimension; ++gi)
-      in_poly[coeffIdx + gi * polynomial_size] =
-          -in_poly[coeffIdx + gi * polynomial_size];
-
-  } else {
-    // Otherwise simply negate the input coefficient
-    for (int gi = 1; gi < glwe_dimension + 1; ++gi)
-      in_poly[coeffIdx + gi * polynomial_size] =
-          -in_poly[coeffIdx + gi * polynomial_size];
-  }
-  // Negate all the coefficients for rotation for the first poly
-  in_poly[coeffIdx] = -in_poly[coeffIdx];
-
-  // rotate the body
-  polynomial_accumulate_monic_monomial_mul<Torus>(
-      out_result, in_poly, degree, coeffIdx, polynomial_size, 1, true);
-  // rotate the mask too
-  for (int gi = 1; gi < glwe_dimension + 1; ++gi)
-    polynomial_accumulate_monic_monomial_mul<Torus>(
-        out_result + gi * polynomial_size, in_poly + gi * polynomial_size,
-        degree, coeffIdx, polynomial_size, 1, true);
-}
-
-template <typename Torus, typename TorusVec>
-__host__ void host_fast_packing_keyswitch_lwe_list_to_glwe(
-    cudaStream_t stream, uint32_t gpu_index, Torus *glwe_out,
-    Torus const *lwe_array_in, Torus const *fp_ksk_array, int8_t *fp_ks_buffer,
-    uint32_t lwe_dimension, uint32_t glwe_dimension, uint32_t polynomial_size,
-    uint32_t base_log, uint32_t level_count, uint32_t num_lwes) {
-
-  // Optimization of packing keyswitch when packing many LWEs
-
-  if (level_count > 1) {
-    PANIC("Fast path PKS only supports level_count==1");
-  }
-
-  cudaSetDevice(gpu_index);
-  check_cuda_error(cudaGetLastError());
-
-  int glwe_accumulator_size = (glwe_dimension + 1) * polynomial_size;
-
-  // The fast path of PKS uses the scratch buffer (d_mem) differently than the
-  // old path: it needs to store the decomposed masks in the first half of this
-  // buffer and the keyswitched GLWEs in the second half of the buffer. Thus the
-  // scratch buffer for the fast path must determine the half-size of the
-  // scratch buffer as the max between the size of the GLWE and the size of the
-  // LWE-mask
-  int memory_unit = glwe_accumulator_size > lwe_dimension
-                        ? glwe_accumulator_size
-                        : lwe_dimension;
-
-  // ping pong the buffer between successive calls
-  // split the buffer in two parts of this size
-  auto d_mem_0 = (Torus *)fp_ks_buffer;
-  auto d_mem_1 = d_mem_0 + num_lwes * memory_unit;
-
-  // Set the scratch buffer to 0 as it is used to accumulate
-  // decomposition temporary results
-  cuda_memset_async(d_mem_1, 0, num_lwes * memory_unit * sizeof(Torus), stream,
-                    gpu_index);
-  check_cuda_error(cudaGetLastError());
-
-  // decompose LWEs
-  // don't decompose LWE body - the LWE has lwe_size + 1 elements. The last
-  // element, the body is ignored by rounding down the number of blocks assuming
-  // here that the LWE dimension is a multiple of the block size
-  dim3 grid_decomp(CEIL_DIV(num_lwes, BLOCK_SIZE_DECOMP),
-                   CEIL_DIV(lwe_dimension, BLOCK_SIZE_DECOMP));
-  dim3 threads_decomp(BLOCK_SIZE_DECOMP, BLOCK_SIZE_DECOMP);
-
-  // decompose first level
-  decompose_vectorize_init<Torus, TorusVec>
-      <<<grid_decomp, threads_decomp, 0, stream>>>(lwe_array_in, d_mem_0,
-                                                   lwe_dimension, num_lwes,
-                                                   base_log, level_count);
-  check_cuda_error(cudaGetLastError());
-
-  // gemm to ks the individual LWEs to GLWEs
-  dim3 grid_gemm(CEIL_DIV(glwe_accumulator_size, BLOCK_SIZE_GEMM),
-                 CEIL_DIV(num_lwes, BLOCK_SIZE_GEMM));
-  dim3 threads_gemm(BLOCK_SIZE_GEMM * THREADS_GEMM);
-
-  auto stride_KSK_buffer = glwe_accumulator_size;
-
-  uint32_t shared_mem_size = get_shared_mem_size_tgemm<Torus>();
-  tgemm<Torus, TorusVec><<<grid_gemm, threads_gemm, shared_mem_size, stream>>>(
-      num_lwes, glwe_accumulator_size, lwe_dimension, d_mem_0, fp_ksk_array,
-      stride_KSK_buffer, d_mem_1);
-  check_cuda_error(cudaGetLastError());
-
-  /*
-    TODO: transpose key to generalize to level_count > 1
-
-    for (int li = 1; li < level_count; ++li) {
-      decompose_vectorize_step_inplace<Torus, TorusVec>
-          <<<grid_decomp, threads_decomp, 0, stream>>>(
-              d_mem_0, lwe_dimension, num_lwes, base_log, level_count);
-      check_cuda_error(cudaGetLastError());
-
-      tgemm<Torus, TorusVec><<<grid_gemm, threads_gemm, shared_mem_size,
-    stream>>>( num_lwes, glwe_accumulator_size, lwe_dimension, d_mem_0,
-          fp_ksk_array + li * ksk_block_size, stride_KSK_buffer, d_mem_1);
-      check_cuda_error(cudaGetLastError());
-    }
-  */
-
-  // should we include the mask in the rotation ??
-  dim3 grid_rotate(CEIL_DIV(num_lwes, BLOCK_SIZE_DECOMP),
-                   CEIL_DIV(polynomial_size, BLOCK_SIZE_DECOMP));
-  dim3 threads_rotate(BLOCK_SIZE_DECOMP, BLOCK_SIZE_DECOMP);
-  // rotate the GLWEs
-  polynomial_accumulate_monic_monomial_mul_many_neg_and_add_C<Torus>
-      <<<grid_rotate, threads_rotate, 0, stream>>>(
-          d_mem_1, d_mem_0, lwe_array_in, lwe_dimension, num_lwes,
-          polynomial_size, glwe_dimension);
-  check_cuda_error(cudaGetLastError());
-
-  dim3 grid_accumulate(
-      CEIL_DIV(polynomial_size * (glwe_dimension + 1), BLOCK_SIZE_DECOMP));
-  dim3 threads_accum(BLOCK_SIZE_DECOMP);
-
-  // accumulate to a single glwe
-  accumulate_glwes<Torus><<<grid_accumulate, threads_accum, 0, stream>>>(
-      glwe_out, d_mem_0, glwe_dimension, polynomial_size, num_lwes);
-
-  check_cuda_error(cudaGetLastError());
-}
-
-#endif

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

@@ -1,7 +1,6 @@
 #ifndef CNCRT_CRYPTO_CUH
 #define CNCRT_CRPYTO_CUH
 
-#include "crypto/torus.cuh"
 #include "device.h"
 #include <cstdint>
 
@@ -22,6 +21,7 @@ private:
   uint32_t base_log;
   uint32_t mask;
   uint32_t num_poly;
+  int current_level;
   T mask_mod_b;
   T *state;
 
@@ -32,6 +32,13 @@ public:
         state(state) {
 
     mask_mod_b = (1ll << base_log) - 1ll;
+    current_level = level_count;
+    int tid = threadIdx.x;
+    for (int i = 0; i < num_poly * params::opt; i++) {
+      state[tid] >>= (sizeof(T) * 8 - base_log * level_count);
+      tid += params::degree / params::opt;
+    }
+    synchronize_threads_in_block();
   }
 
   // Decomposes all polynomials at once
@@ -45,30 +52,28 @@ public:
   // Decomposes a single polynomial
   __device__ void decompose_and_compress_next_polynomial(double2 *result,
                                                          int j) {
-    uint32_t tid = threadIdx.x;
-    auto state_slice = &state[j * params::degree];
+    if (j == 0)
+      current_level -= 1;
+
+    int tid = threadIdx.x;
+    auto state_slice = state + j * params::degree;
     for (int i = 0; i < params::opt / 2; i++) {
-      auto input1 = &state_slice[tid];
-      auto input2 = &state_slice[tid + params::degree / 2];
-      T res_re = *input1 & mask_mod_b;
-      T res_im = *input2 & mask_mod_b;
-
-      *input1 >>= base_log; // Update state
-      *input2 >>= base_log; // Update state
-
-      T carry_re = ((res_re - 1ll) | *input1) & res_re;
-      T carry_im = ((res_im - 1ll) | *input2) & res_im;
+      T res_re = state_slice[tid] & mask_mod_b;
+      T res_im = state_slice[tid + params::degree / 2] & mask_mod_b;
+      state_slice[tid] >>= base_log;
+      state_slice[tid + params::degree / 2] >>= base_log;
+      T carry_re = ((res_re - 1ll) | state_slice[tid]) & res_re;
+      T carry_im =
+          ((res_im - 1ll) | state_slice[tid + params::degree / 2]) & res_im;
       carry_re >>= (base_log - 1);
       carry_im >>= (base_log - 1);
-
-      *input1 += carry_re; // Update state
-      *input2 += carry_im; // Update state
-
+      state_slice[tid] += carry_re;
+      state_slice[tid + params::degree / 2] += carry_im;
       res_re -= carry_re << base_log;
       res_im -= carry_im << base_log;
 
-      typecast_torus_to_double(res_re, result[tid].x);
-      typecast_torus_to_double(res_im, result[tid].y);
+      result[tid].x = (int32_t)res_re;
+      result[tid].y = (int32_t)res_im;
 
       tid += params::degree / params::opt;
     }

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

@@ -1,8 +1,6 @@
-#include "fast_packing_keyswitch.cuh"
 #include "keyswitch.cuh"
 #include "keyswitch.h"
 #include <cstdint>
-#include <stdio.h>
 
 /* Perform keyswitch on a batch of 32 bits input LWE ciphertexts.
  * Head out to the equivalent operation on 64 bits for more details.
@@ -55,17 +53,15 @@ void cuda_keyswitch_lwe_ciphertext_vector_64(
 
 void scratch_packing_keyswitch_lwe_list_to_glwe_64(
     void *stream, uint32_t gpu_index, int8_t **fp_ks_buffer,
-    uint32_t lwe_dimension, uint32_t glwe_dimension, uint32_t polynomial_size,
-    uint32_t num_lwes, bool allocate_gpu_memory) {
+    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, lwe_dimension,
+      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 const *lwe_array_in, void const *fp_ksk_array, int8_t *fp_ks_buffer,
@@ -73,24 +69,13 @@ void cuda_packing_keyswitch_lwe_list_to_glwe_64(
     uint32_t output_polynomial_size, uint32_t base_log, uint32_t level_count,
     uint32_t num_lwes) {
 
-  if (can_use_pks_fast_path(input_lwe_dimension, num_lwes,
-                            output_polynomial_size, level_count,
-                            output_glwe_dimension)) {
-    host_fast_packing_keyswitch_lwe_list_to_glwe<uint64_t, ulonglong4>(
-        static_cast<cudaStream_t>(stream), gpu_index,
-        static_cast<uint64_t *>(glwe_array_out),
-        static_cast<const uint64_t *>(lwe_array_in),
-        static_cast<const uint64_t *>(fp_ksk_array), fp_ks_buffer,
-        input_lwe_dimension, output_glwe_dimension, output_polynomial_size,
-        base_log, level_count, num_lwes);
-  } else
-    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<const uint64_t *>(lwe_array_in),
-        static_cast<const uint64_t *>(fp_ksk_array), fp_ks_buffer,
-        input_lwe_dimension, output_glwe_dimension, output_polynomial_size,
-        base_log, level_count, 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<const uint64_t *>(lwe_array_in),
+      static_cast<const 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,

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

@@ -71,10 +71,12 @@ keyswitch(Torus *lwe_array_out, const Torus *__restrict__ lwe_output_indexes,
 
     // This loop distribution seems to benefit the global mem reads
     for (int i = start_i; i < end_i; i++) {
-      Torus state =
-          init_decomposer_state(block_lwe_array_in[i], base_log, level_count);
+      Torus a_i = round_to_closest_multiple(block_lwe_array_in[i], base_log,
+                                            level_count);
+      Torus state = a_i >> (sizeof(Torus) * 8 - base_log * level_count);
 
-      for (int j = 0; j < level_count; j++) {
+      for (int j = level_count - 1; j >= 0; j--) {
+        // Levels are stored in reverse order
         auto ksk_block =
             get_ith_block(ksk, i, j, lwe_dimension_out, level_count);
         Torus decomposed = decompose_one<Torus>(state, mask_mod_b, base_log);
@@ -158,20 +160,16 @@ void execute_keyswitch_async(cudaStream_t const *streams,
 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 lwe_dimension, uint32_t glwe_dimension, uint32_t polynomial_size,
-    uint32_t num_lwes, bool allocate_gpu_memory) {
+    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;
 
-  int memory_unit = glwe_accumulator_size > lwe_dimension
-                        ? glwe_accumulator_size
-                        : lwe_dimension;
-
-  if (allocate_gpu_memory) {
+  if (allocate_gpu_memory)
     *fp_ks_buffer = (int8_t *)cuda_malloc_async(
-        2 * num_lwes * memory_unit * sizeof(Torus), stream, gpu_index);
-  }
+        2 * num_lwes * glwe_accumulator_size * sizeof(Torus), stream,
+        gpu_index);
 }
 
 // public functional packing keyswitch for a single LWE ciphertext
@@ -204,13 +202,15 @@ __device__ void packing_keyswitch_lwe_ciphertext_into_glwe_ciphertext(
     // Iterate through all lwe elements
     for (int i = 0; i < lwe_dimension_in; i++) {
       // Round and prepare decomposition
-      Torus state = init_decomposer_state(lwe_in[i], base_log, level_count);
+      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++) {
+      for (int j = level_count - 1; j >= 0; j--) {
+        // Levels are stored in reverse order
         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];
@@ -245,7 +245,6 @@ __global__ void packing_keyswitch_lwe_list_to_glwe(
   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<Torus>(
       ks_glwe_out, lwe_in, fp_ksk, lwe_dimension_in, glwe_dimension,
@@ -298,18 +297,8 @@ __host__ void host_packing_keyswitch_lwe_list_to_glwe(
   dim3 grid(num_blocks, num_lwes);
   dim3 threads(num_threads);
 
-  // The fast path of PKS uses the scratch buffer (d_mem) differently:
-  // it needs to store the decomposed masks in the first half of this buffer
-  // and the keyswitched GLWEs in the second half of the buffer. Thus the
-  // scratch buffer for the fast path must determine the half-size of the
-  // scratch buffer as the max between the size of the GLWE and the size of the
-  // LWE-mask
-  int memory_unit = glwe_accumulator_size > lwe_dimension_in
-                        ? glwe_accumulator_size
-                        : lwe_dimension_in;
-
   auto d_mem = (Torus *)fp_ks_buffer;
-  auto d_tmp_glwe_array_out = d_mem + num_lwes * memory_unit;
+  auto d_tmp_glwe_array_out = d_mem + num_lwes * glwe_accumulator_size;
 
   // individually keyswitch each lwe
   packing_keyswitch_lwe_list_to_glwe<Torus><<<grid, threads, 0, stream>>>(

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

@@ -1,7 +1,6 @@
 #ifndef CNCRT_TORUS_CUH
 #define CNCRT_TORUS_CUH
 
-#include "device.h"
 #include "polynomial/parameters.cuh"
 #include "types/int128.cuh"
 #include "utils/kernel_dimensions.cuh"
@@ -12,11 +11,6 @@ __host__ __device__ __forceinline__ constexpr double get_two_pow_torus_bits() {
   return (sizeof(T) == 4) ? 4294967296.0 : 18446744073709551616.0;
 }
 
-template <typename T>
-__host__ __device__ __forceinline__ constexpr T scalar_max() {
-  return std::numeric_limits<T>::max();
-}
-
 template <typename T>
 __device__ inline void typecast_double_to_torus(double x, T &r) {
   r = T(x);
@@ -50,36 +44,14 @@ __device__ inline void typecast_double_round_to_torus(double x, T &r) {
 }
 
 template <typename T>
-__device__ inline void typecast_torus_to_double(T x, double &r);
-
-template <>
-__device__ inline void typecast_torus_to_double<uint32_t>(uint32_t x,
-                                                          double &r) {
-  r = __int2double_rn(x);
-}
-
-template <>
-__device__ inline void typecast_torus_to_double<uint64_t>(uint64_t x,
-                                                          double &r) {
-  r = __ll2double_rn(x);
-}
-
-template <typename T>
-__device__ inline T init_decomposer_state(T input, uint32_t base_log,
-                                          uint32_t level_count) {
-  const T rep_bit_count = level_count * base_log;
-  const T non_rep_bit_count = sizeof(T) * 8 - rep_bit_count;
-  T res = input >> (non_rep_bit_count - 1);
-  T rounding_bit = res & (T)(1);
-  res++;
-  res >>= 1;
-  T torus_max = scalar_max<T>();
-  T mod_mask = torus_max >> non_rep_bit_count;
-  res &= mod_mask;
-  T shifted_random = rounding_bit << (rep_bit_count - 1);
-  T need_balance =
-      (((res - (T)(1)) | shifted_random) & res) >> (rep_bit_count - 1);
-  return res - (need_balance << rep_bit_count);
+__device__ inline T round_to_closest_multiple(T x, uint32_t base_log,
+                                              uint32_t level_count) {
+  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 += 1;
+  res &= (T)(-2);
+  return res << shift;
 }
 
 template <typename T>

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

@@ -2,30 +2,6 @@
 #include <cstdint>
 #include <cuda_runtime.h>
 
-cudaEvent_t cuda_create_event(uint32_t gpu_index) {
-  check_cuda_error(cudaSetDevice(gpu_index));
-  cudaEvent_t event;
-  check_cuda_error(cudaEventCreate(&event));
-  return event;
-}
-
-void cuda_event_record(cudaEvent_t event, cudaStream_t stream,
-                       uint32_t gpu_index) {
-  check_cuda_error(cudaSetDevice(gpu_index));
-  check_cuda_error(cudaEventRecord(event, stream));
-}
-
-void cuda_stream_wait_event(cudaStream_t stream, cudaEvent_t event,
-                            uint32_t gpu_index) {
-  check_cuda_error(cudaSetDevice(gpu_index));
-  check_cuda_error(cudaStreamWaitEvent(stream, event, 0));
-}
-
-void cuda_event_destroy(cudaEvent_t event, uint32_t gpu_index) {
-  check_cuda_error(cudaSetDevice(gpu_index));
-  check_cuda_error(cudaEventDestroy(event));
-}
-
 /// Unsafe function to create a CUDA stream, must check first that GPU exists
 cudaStream_t cuda_create_stream(uint32_t gpu_index) {
   check_cuda_error(cudaSetDevice(gpu_index));
@@ -45,9 +21,6 @@ void cuda_synchronize_stream(cudaStream_t stream, uint32_t gpu_index) {
   check_cuda_error(cudaStreamSynchronize(stream));
 }
 
-// Determine if a CUDA device is available at runtime
-uint32_t cuda_is_available() { return cudaSetDevice(0) == cudaSuccess; }
-
 /// Unsafe function that will try to allocate even if gpu_index is invalid
 /// or if there's not enough memory. A safe wrapper around it must call
 /// cuda_check_valid_malloc() first
@@ -295,20 +268,17 @@ 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) {
   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 = 100000;
-#elif CUDA_ARCH == 860
-  max_shared_memory = 100000;
 #elif CUDA_ARCH == 800
   max_shared_memory = 163000;
 #elif CUDA_ARCH == 700
   max_shared_memory = 95000;
-#else
-  cudaDeviceGetAttribute(&max_shared_memory, cudaDevAttrMaxSharedMemoryPerBlock,
-                         gpu_index);
-  check_cuda_error(cudaGetLastError());
 #endif
   return max_shared_memory;
 }

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

@@ -1,43 +0,0 @@
-#include "integer/abs.cuh"
-
-void scratch_cuda_integer_abs_inplace_radix_ciphertext_kb_64(
-    void *const *streams, uint32_t const *gpu_indexes, uint32_t gpu_count,
-    int8_t **mem_ptr, bool is_signed, uint32_t glwe_dimension,
-    uint32_t polynomial_size, uint32_t big_lwe_dimension,
-    uint32_t small_lwe_dimension, uint32_t ks_level, uint32_t ks_base_log,
-    uint32_t pbs_level, uint32_t pbs_base_log, uint32_t grouping_factor,
-    uint32_t num_blocks, uint32_t message_modulus, uint32_t carry_modulus,
-    PBS_TYPE pbs_type, bool allocate_gpu_memory) {
-
-  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_abs_kb<uint64_t>(
-      (cudaStream_t *)(streams), gpu_indexes, gpu_count,
-      (int_abs_buffer<uint64_t> **)mem_ptr, is_signed, num_blocks, params,
-      allocate_gpu_memory);
-}
-
-void cuda_integer_abs_inplace_radix_ciphertext_kb_64(
-    void *const *streams, uint32_t const *gpu_indexes, uint32_t gpu_count,
-    void *ct, int8_t *mem_ptr, bool is_signed, void *const *bsks,
-    void *const *ksks, uint32_t num_blocks) {
-
-  auto mem = (int_abs_buffer<uint64_t> *)mem_ptr;
-
-  host_integer_abs_kb<uint64_t>((cudaStream_t *)(streams), gpu_indexes,
-                                gpu_count, static_cast<uint64_t *>(ct), bsks,
-                                (uint64_t **)(ksks), mem, is_signed,
-                                num_blocks);
-}
-
-void cleanup_cuda_integer_abs_inplace(void *const *streams,
-                                      uint32_t const *gpu_indexes,
-                                      uint32_t gpu_count,
-                                      int8_t **mem_ptr_void) {
-  int_abs_buffer<uint64_t> *mem_ptr =
-      (int_abs_buffer<uint64_t> *)(*mem_ptr_void);
-  mem_ptr->release((cudaStream_t *)(streams), gpu_indexes, gpu_count);
-}

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

@@ -1,71 +0,0 @@
-#ifndef TFHE_RS_ABS_CUH
-#define TFHE_RS_ABS_CUH
-
-#include "crypto/keyswitch.cuh"
-#include "device.h"
-#include "integer/bitwise_ops.cuh"
-#include "integer/comparison.cuh"
-#include "integer/integer.cuh"
-#include "integer/integer_utilities.h"
-#include "integer/negation.cuh"
-#include "integer/scalar_shifts.cuh"
-#include "linear_algebra.h"
-#include "pbs/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>
-__host__ void scratch_cuda_integer_abs_kb(
-    cudaStream_t const *streams, uint32_t const *gpu_indexes,
-    uint32_t gpu_count, int_abs_buffer<Torus> **mem_ptr, bool is_signed,
-    uint32_t num_blocks, int_radix_params params, bool allocate_gpu_memory) {
-
-  if (is_signed)
-    *mem_ptr =
-        new int_abs_buffer<Torus>(streams, gpu_indexes, gpu_count, params,
-                                  num_blocks, allocate_gpu_memory);
-}
-
-template <typename Torus>
-__host__ void
-host_integer_abs_kb(cudaStream_t const *streams, uint32_t const *gpu_indexes,
-                    uint32_t gpu_count, Torus *ct, void *const *bsks,
-                    uint64_t *const *ksks, int_abs_buffer<uint64_t> *mem_ptr,
-                    bool is_signed, uint32_t num_blocks) {
-  if (!is_signed)
-    return;
-
-  auto radix_params = mem_ptr->params;
-  auto mask = mem_ptr->mask;
-
-  auto big_lwe_dimension = radix_params.big_lwe_dimension;
-  auto big_lwe_size = big_lwe_dimension + 1;
-  auto big_lwe_size_bytes = big_lwe_size * sizeof(Torus);
-  uint32_t num_bits_in_ciphertext =
-      (31 - __builtin_clz(radix_params.message_modulus)) * num_blocks;
-
-  cuda_memcpy_async_gpu_to_gpu(mask, ct, num_blocks * big_lwe_size_bytes,
-                               streams[0], gpu_indexes[0]);
-
-  host_integer_radix_arithmetic_scalar_shift_kb_inplace(
-      streams, gpu_indexes, gpu_count, mask, num_bits_in_ciphertext - 1,
-      mem_ptr->arithmetic_scalar_shift_mem, bsks, ksks, num_blocks);
-  host_addition<Torus>(streams[0], gpu_indexes[0], ct, mask, ct,
-                       radix_params.big_lwe_dimension, num_blocks);
-
-  uint32_t requested_flag = outputFlag::FLAG_NONE;
-  uint32_t uses_carry = 0;
-  host_propagate_single_carry<Torus>(
-      streams, gpu_indexes, gpu_count, ct, nullptr, nullptr, mem_ptr->scp_mem,
-      bsks, ksks, num_blocks, requested_flag, uses_carry);
-
-  host_integer_radix_bitop_kb(streams, gpu_indexes, gpu_count, ct, mask, ct,
-                              mem_ptr->bitxor_mem, bsks, ksks, num_blocks);
-}
-
-#endif // TFHE_RS_ABS_CUH

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

@@ -0,0 +1,50 @@
+#include "integer/addition.cuh"
+
+void scratch_cuda_signed_overflowing_add_or_sub_radix_ciphertext_kb_64(
+    void *const *streams, uint32_t const *gpu_indexes, uint32_t gpu_count,
+    int8_t **mem_ptr, uint32_t glwe_dimension, uint32_t polynomial_size,
+    uint32_t big_lwe_dimension, uint32_t small_lwe_dimension, uint32_t ks_level,
+    uint32_t ks_base_log, uint32_t pbs_level, uint32_t pbs_base_log,
+    uint32_t grouping_factor, uint32_t num_blocks, int8_t signed_operation,
+    uint32_t message_modulus, uint32_t carry_modulus, PBS_TYPE pbs_type,
+    bool allocate_gpu_memory) {
+
+  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 *const *streams, uint32_t const *gpu_indexes, uint32_t gpu_count,
+    void *lhs, void const *rhs, void *overflowed, int8_t signed_operation,
+    int8_t *mem_ptr, void *const *bsks, void *const *ksks,
+    uint32_t num_blocks) {
+
+  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 const *>(rhs),
+      static_cast<uint64_t *>(overflowed), op, bsks, (uint64_t *const *)(ksks),
+      mem, num_blocks);
+}
+
+void cleanup_signed_overflowing_add_or_sub(void *const *streams,
+                                           uint32_t const *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 +1,139 @@
+#ifndef TFHE_RS_ADDITION_CUH
+#define TFHE_RS_ADDITION_CUH
+
+#include "crypto/keyswitch.cuh"
+#include "device.h"
+#include "integer/comparison.cuh"
+#include "integer/integer.cuh"
+#include "integer/integer_utilities.h"
+#include "integer/negation.cuh"
+#include "integer/scalar_shifts.cuh"
+#include "linear_algebra.h"
+#include "pbs/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 const *streams, uint32_t const *gpu_indexes,
+    uint32_t gpu_count, Torus *result, Torus *last_block_inner_propagation,
+    Torus const *last_block_input_carry, Torus *last_block_output_carry,
+    int_resolve_signed_overflow_memory<Torus> *mem, void *const *bsks,
+    Torus *const *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 const *streams, uint32_t const *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 const *streams, uint32_t const *gpu_indexes,
+    uint32_t gpu_count, Torus *lhs, Torus const *rhs, Torus *overflowed,
+    SIGNED_OPERATION op, void *const *bsks, uint64_t *const *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.cuh

@@ -14,14 +14,27 @@ __host__ void zero_out_if(cudaStream_t const *streams,
   cudaSetDevice(gpu_indexes[0]);
   auto params = mem_ptr->params;
 
+  int big_lwe_size = params.big_lwe_dimension + 1;
+
+  // Left message is shifted
+  int num_blocks = 0, num_threads = 0;
+  int num_entries = (params.big_lwe_dimension + 1);
+  getNumBlocksAndThreads(num_entries, 512, num_blocks, num_threads);
+
   // We can't use integer_radix_apply_bivariate_lookup_table_kb since the
-  // second operand is not an array
+  // second operand is fixed
   auto tmp_lwe_array_input = mem_ptr->tmp;
-  pack_bivariate_blocks_with_single_block<Torus>(
-      streams, gpu_indexes, gpu_count, tmp_lwe_array_input,
-      predicate->lwe_indexes_in, lwe_array_input, lwe_condition,
-      predicate->lwe_indexes_in, params.big_lwe_dimension,
-      params.message_modulus, num_radix_blocks);
+  for (int i = 0; i < num_radix_blocks; i++) {
+    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<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());
+  }
 
   integer_radix_apply_univariate_lookup_table_kb<Torus>(
       streams, gpu_indexes, gpu_count, lwe_array_out, tmp_lwe_array_input, bsks,
@@ -37,32 +50,39 @@ __host__ void host_integer_radix_cmux_kb(
     uint32_t num_radix_blocks) {
 
   auto params = mem_ptr->params;
-  Torus lwe_size = params.big_lwe_dimension + 1;
-  Torus radix_lwe_size = lwe_size * num_radix_blocks;
-  cuda_memcpy_async_gpu_to_gpu(mem_ptr->buffer_in, lwe_array_true,
-                               radix_lwe_size * sizeof(Torus), streams[0],
-                               gpu_indexes[0]);
-  cuda_memcpy_async_gpu_to_gpu(mem_ptr->buffer_in + radix_lwe_size,
-                               lwe_array_false, radix_lwe_size * sizeof(Torus),
-                               streams[0], gpu_indexes[0]);
-  for (uint i = 0; i < 2 * num_radix_blocks; i++) {
-    cuda_memcpy_async_gpu_to_gpu(mem_ptr->condition_array + i * lwe_size,
-                                 lwe_condition, lwe_size * sizeof(Torus),
-                                 streams[0], gpu_indexes[0]);
+
+  // Since our CPU threads will be working on different streams we shall assert
+  // the work in the main stream is completed
+  auto true_streams = mem_ptr->zero_if_true_buffer->true_streams;
+  auto false_streams = mem_ptr->zero_if_false_buffer->false_streams;
+  for (uint j = 0; j < gpu_count; j++) {
+    cuda_synchronize_stream(streams[j], gpu_indexes[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]);
   }
-  integer_radix_apply_bivariate_lookup_table_kb<Torus>(
-      streams, gpu_indexes, gpu_count, mem_ptr->buffer_out, mem_ptr->buffer_in,
-      mem_ptr->condition_array, bsks, ksks, 2 * num_radix_blocks,
-      mem_ptr->predicate_lut, params.message_modulus);
 
   // If the condition was true, true_ct will have kept its value and false_ct
   // will be 0 If the condition was false, true_ct will be 0 and false_ct will
   // have kept its value
-  auto mem_true = mem_ptr->buffer_out;
-  auto mem_false = &mem_ptr->buffer_out[radix_lwe_size];
-  auto added_cts = mem_true;
-  host_addition<Torus>(streams[0], gpu_indexes[0], added_cts, mem_true,
-                       mem_false, params.big_lwe_dimension, num_radix_blocks);
+  auto added_cts = mem_ptr->tmp_true_ct;
+  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.cu

@@ -58,9 +58,6 @@ void cuda_comparison_integer_radix_ciphertext_kb_64(
   case GE:
   case LT:
   case LE:
-    if (num_radix_blocks % 2 != 0)
-      PANIC("Cuda error (comparisons): the number of radix blocks has to be "
-            "even.")
     host_integer_radix_difference_check_kb<uint64_t>(
         (cudaStream_t *)(streams), gpu_indexes, gpu_count,
         static_cast<uint64_t *>(lwe_array_out),
@@ -71,8 +68,6 @@ void cuda_comparison_integer_radix_ciphertext_kb_64(
     break;
   case MAX:
   case MIN:
-    if (num_radix_blocks % 2 != 0)
-      PANIC("Cuda error (max/min): the number of radix blocks has to be even.")
     host_integer_radix_maxmin_kb<uint64_t>(
         (cudaStream_t *)(streams), gpu_indexes, gpu_count,
         static_cast<uint64_t *>(lwe_array_out),
@@ -94,91 +89,3 @@ void cleanup_cuda_integer_comparison(void *const *streams,
       (int_comparison_buffer<uint64_t> *)(*mem_ptr_void);
   mem_ptr->release((cudaStream_t *)(streams), gpu_indexes, gpu_count);
 }
-
-void scratch_cuda_integer_are_all_comparisons_block_true_kb_64(
-    void *const *streams, uint32_t const *gpu_indexes, uint32_t gpu_count,
-    int8_t **mem_ptr, uint32_t glwe_dimension, uint32_t polynomial_size,
-    uint32_t big_lwe_dimension, uint32_t small_lwe_dimension, uint32_t ks_level,
-    uint32_t ks_base_log, uint32_t pbs_level, uint32_t pbs_base_log,
-    uint32_t grouping_factor, uint32_t num_radix_blocks,
-    uint32_t message_modulus, uint32_t carry_modulus, PBS_TYPE pbs_type,
-    bool allocate_gpu_memory) {
-
-  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_radix_comparison_check_kb<uint64_t>(
-      (cudaStream_t *)(streams), gpu_indexes, gpu_count,
-      (int_comparison_buffer<uint64_t> **)mem_ptr, num_radix_blocks, params, EQ,
-      false, allocate_gpu_memory);
-}
-
-void cuda_integer_are_all_comparisons_block_true_kb_64(
-    void *const *streams, uint32_t const *gpu_indexes, uint32_t gpu_count,
-    void *lwe_array_out, void const *lwe_array_in, int8_t *mem_ptr,
-    void *const *bsks, void *const *ksks, uint32_t num_radix_blocks) {
-
-  int_comparison_buffer<uint64_t> *buffer =
-      (int_comparison_buffer<uint64_t> *)mem_ptr;
-
-  host_integer_are_all_comparisons_block_true_kb<uint64_t>(
-      (cudaStream_t *)(streams), gpu_indexes, gpu_count,
-      static_cast<uint64_t *>(lwe_array_out),
-      static_cast<const uint64_t *>(lwe_array_in), buffer, bsks,
-      (uint64_t **)(ksks), num_radix_blocks);
-}
-
-void cleanup_cuda_integer_are_all_comparisons_block_true(
-    void *const *streams, uint32_t const *gpu_indexes, uint32_t gpu_count,
-    int8_t **mem_ptr_void) {
-
-  int_comparison_buffer<uint64_t> *mem_ptr =
-      (int_comparison_buffer<uint64_t> *)(*mem_ptr_void);
-  mem_ptr->release((cudaStream_t *)(streams), gpu_indexes, gpu_count);
-}
-
-void scratch_cuda_integer_is_at_least_one_comparisons_block_true_kb_64(
-    void *const *streams, uint32_t const *gpu_indexes, uint32_t gpu_count,
-    int8_t **mem_ptr, uint32_t glwe_dimension, uint32_t polynomial_size,
-    uint32_t big_lwe_dimension, uint32_t small_lwe_dimension, uint32_t ks_level,
-    uint32_t ks_base_log, uint32_t pbs_level, uint32_t pbs_base_log,
-    uint32_t grouping_factor, uint32_t num_radix_blocks,
-    uint32_t message_modulus, uint32_t carry_modulus, PBS_TYPE pbs_type,
-    bool allocate_gpu_memory) {
-
-  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_radix_comparison_check_kb<uint64_t>(
-      (cudaStream_t *)(streams), gpu_indexes, gpu_count,
-      (int_comparison_buffer<uint64_t> **)mem_ptr, num_radix_blocks, params, EQ,
-      false, allocate_gpu_memory);
-}
-
-void cuda_integer_is_at_least_one_comparisons_block_true_kb_64(
-    void *const *streams, uint32_t const *gpu_indexes, uint32_t gpu_count,
-    void *lwe_array_out, void const *lwe_array_in, int8_t *mem_ptr,
-    void *const *bsks, void *const *ksks, uint32_t num_radix_blocks) {
-
-  int_comparison_buffer<uint64_t> *buffer =
-      (int_comparison_buffer<uint64_t> *)mem_ptr;
-
-  host_integer_is_at_least_one_comparisons_block_true_kb<uint64_t>(
-      (cudaStream_t *)(streams), gpu_indexes, gpu_count,
-      static_cast<uint64_t *>(lwe_array_out),
-      static_cast<const uint64_t *>(lwe_array_in), buffer, bsks,
-      (uint64_t **)(ksks), num_radix_blocks);
-}
-
-void cleanup_cuda_integer_is_at_least_one_comparisons_block_true(
-    void *const *streams, uint32_t const *gpu_indexes, uint32_t gpu_count,
-    int8_t **mem_ptr_void) {
-
-  int_comparison_buffer<uint64_t> *mem_ptr =
-      (int_comparison_buffer<uint64_t> *)(*mem_ptr_void);
-  mem_ptr->release((cudaStream_t *)(streams), gpu_indexes, gpu_count);
-}

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

@@ -58,7 +58,7 @@ __host__ void accumulate_all_blocks(cudaStream_t stream, uint32_t gpu_index,
 template <typename Torus>
 __host__ void are_all_comparisons_block_true(
     cudaStream_t const *streams, uint32_t const *gpu_indexes,
-    uint32_t gpu_count, Torus *lwe_array_out, Torus const *lwe_array_in,
+    uint32_t gpu_count, Torus *lwe_array_out, Torus *lwe_array_in,
     int_comparison_buffer<Torus> *mem_ptr, void *const *bsks,
     Torus *const *ksks, uint32_t num_radix_blocks) {
 
@@ -85,19 +85,16 @@ __host__ void are_all_comparisons_block_true(
 
   while (remaining_blocks > 0) {
     // Split in max_value chunks
-    int num_chunks = (remaining_blocks + max_value - 1) / max_value;
+    uint32_t chunk_length = std::min(max_value, remaining_blocks);
+    int num_chunks = remaining_blocks / chunk_length;
 
     // Since all blocks encrypt either 0 or 1, we can sum max_value of them
     // as in the worst case we will be adding `max_value` ones
     auto input_blocks = tmp_out;
     auto accumulator = are_all_block_true_buffer->tmp_block_accumulated;
-    auto is_max_value_lut = are_all_block_true_buffer->is_max_value;
-    uint32_t chunk_lengths[num_chunks];
-    auto begin_remaining_blocks = remaining_blocks;
+    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++) {
-      uint32_t chunk_length =
-          std::min(max_value, begin_remaining_blocks - i * max_value);
-      chunk_lengths[i] = chunk_length;
       accumulate_all_blocks<Torus>(streams[0], gpu_indexes[0], accumulator,
                                    input_blocks, big_lwe_dimension,
                                    chunk_length);
@@ -114,33 +111,29 @@ __host__ void are_all_comparisons_block_true(
       // is_non_zero_lut_buffer LUT
       lut = mem_ptr->eq_buffer->is_non_zero_lut;
     } else {
-      if (chunk_lengths[num_chunks - 1] != max_value) {
+      if ((*is_equal_to_num_blocks_map).find(chunk_length) !=
+          (*is_equal_to_num_blocks_map).end()) {
+        // The LUT is already computed
+        lut = (*is_equal_to_num_blocks_map)[chunk_length];
+      } else {
         // LUT needs to be computed
-        uint32_t chunk_length = chunk_lengths[num_chunks - 1];
+        auto new_lut =
+            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 = [chunk_length](Torus x) -> Torus {
           return x == chunk_length;
         };
         generate_device_accumulator<Torus>(
-            streams[0], gpu_indexes[0], is_max_value_lut->get_lut(0, 1),
+            streams[0], gpu_indexes[0], new_lut->get_lut(gpu_indexes[0], 0),
             glwe_dimension, polynomial_size, message_modulus, carry_modulus,
             is_equal_to_num_blocks_lut_f);
 
-        Torus *h_lut_indexes = (Torus *)malloc(num_chunks * sizeof(Torus));
-        for (int index = 0; index < num_chunks; index++) {
-          if (index == num_chunks - 1) {
-            h_lut_indexes[index] = 1;
-          } else {
-            h_lut_indexes[index] = 0;
-          }
-        }
-        cuda_memcpy_async_to_gpu(is_max_value_lut->get_lut_indexes(0, 0),
-                                 h_lut_indexes, num_chunks * sizeof(Torus),
-                                 streams[0], gpu_indexes[0]);
-        is_max_value_lut->broadcast_lut(streams, gpu_indexes, 0);
-        cuda_synchronize_stream(streams[0], gpu_indexes[0]);
-        free(h_lut_indexes);
+        new_lut->broadcast_lut(streams, gpu_indexes, gpu_indexes[0]);
+
+        (*is_equal_to_num_blocks_map)[chunk_length] = new_lut;
+        lut = new_lut;
       }
-      lut = is_max_value_lut;
     }
 
     // Applies the LUT
@@ -167,7 +160,7 @@ __host__ void are_all_comparisons_block_true(
 template <typename Torus>
 __host__ void is_at_least_one_comparisons_block_true(
     cudaStream_t const *streams, uint32_t const *gpu_indexes,
-    uint32_t gpu_count, Torus *lwe_array_out, Torus const *lwe_array_in,
+    uint32_t gpu_count, Torus *lwe_array_out, Torus *lwe_array_in,
     int_comparison_buffer<Torus> *mem_ptr, void *const *bsks,
     Torus *const *ksks, uint32_t num_radix_blocks) {
 
@@ -189,18 +182,14 @@ __host__ void is_at_least_one_comparisons_block_true(
   uint32_t remaining_blocks = num_radix_blocks;
   while (remaining_blocks > 0) {
     // Split in max_value chunks
-    int num_chunks = (remaining_blocks + max_value - 1) / max_value;
+    uint32_t chunk_length = std::min(max_value, remaining_blocks);
+    int num_chunks = remaining_blocks / chunk_length;
 
     // Since all blocks encrypt either 0 or 1, we can sum max_value of them
     // as in the worst case we will be adding `max_value` ones
     auto input_blocks = mem_ptr->tmp_lwe_array_out;
     auto accumulator = buffer->tmp_block_accumulated;
-    uint32_t chunk_lengths[num_chunks];
-    auto begin_remaining_blocks = remaining_blocks;
     for (int i = 0; i < num_chunks; i++) {
-      uint32_t chunk_length =
-          std::min(max_value, begin_remaining_blocks - i * max_value);
-      chunk_lengths[i] = chunk_length;
       accumulate_all_blocks<Torus>(streams[0], gpu_indexes[0], accumulator,
                                    input_blocks, big_lwe_dimension,
                                    chunk_length);
@@ -460,9 +449,9 @@ __host__ void tree_sign_reduction(
     f = sign_handler_f;
   }
   generate_device_accumulator<Torus>(
-      streams[0], gpu_indexes[0], last_lut->get_lut(0, 0), glwe_dimension,
-      polynomial_size, message_modulus, carry_modulus, f);
-  last_lut->broadcast_lut(streams, gpu_indexes, 0);
+      streams[0], gpu_indexes[0], last_lut->get_lut(gpu_indexes[0], 0),
+      glwe_dimension, polynomial_size, message_modulus, carry_modulus, f);
+  last_lut->broadcast_lut(streams, gpu_indexes, gpu_indexes[0]);
 
   // Last leaf
   integer_radix_apply_univariate_lookup_table_kb<Torus>(
@@ -492,9 +481,8 @@ __host__ void host_integer_radix_difference_check_kb(
   if (carry_modulus >= message_modulus) {
     // Packing is possible
     // Pack inputs
-    Torus *packed_left = diff_buffer->tmp_packed;
-    Torus *packed_right =
-        diff_buffer->tmp_packed + num_radix_blocks / 2 * big_lwe_size;
+    Torus *packed_left = diff_buffer->tmp_packed_left;
+    Torus *packed_right = diff_buffer->tmp_packed_right;
     // In case the ciphertext is signed, the sign block and the one before it
     // are handled separately
     if (mem_ptr->is_signed) {
@@ -513,7 +501,10 @@ __host__ void host_integer_radix_difference_check_kb(
     auto identity_lut = mem_ptr->identity_lut;
     integer_radix_apply_univariate_lookup_table_kb<Torus>(
         streams, gpu_indexes, gpu_count, packed_left, packed_left, bsks, ksks,
-        2 * packed_num_radix_blocks, identity_lut);
+        packed_num_radix_blocks, identity_lut);
+    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);
 
     lhs = packed_left;
     rhs = packed_right;
@@ -542,13 +533,11 @@ __host__ void host_integer_radix_difference_check_kb(
 
       // Compare the last block before the sign block separately
       auto identity_lut = mem_ptr->identity_lut;
-      Torus *packed_left = diff_buffer->tmp_packed;
-      Torus *packed_right =
-          diff_buffer->tmp_packed + num_radix_blocks / 2 * big_lwe_size;
       Torus *last_left_block_before_sign_block =
-          packed_left + packed_num_radix_blocks * big_lwe_size;
+          diff_buffer->tmp_packed_left + packed_num_radix_blocks * big_lwe_size;
       Torus *last_right_block_before_sign_block =
-          packed_right + packed_num_radix_blocks * big_lwe_size;
+          diff_buffer->tmp_packed_right +
+          packed_num_radix_blocks * big_lwe_size;
       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,
@@ -626,35 +615,4 @@ __host__ void host_integer_radix_maxmin_kb(
       mem_ptr->cmux_buffer, bsks, ksks, total_num_radix_blocks);
 }
 
-template <typename Torus>
-__host__ void host_integer_are_all_comparisons_block_true_kb(
-    cudaStream_t const *streams, uint32_t const *gpu_indexes,
-    uint32_t gpu_count, Torus *lwe_array_out, Torus const *lwe_array_in,
-    int_comparison_buffer<Torus> *mem_ptr, void *const *bsks,
-    Torus *const *ksks, uint32_t num_radix_blocks) {
-
-  auto eq_buffer = mem_ptr->eq_buffer;
-
-  // It returns a block encrypting 1 if all input blocks are 1
-  // otherwise the block encrypts 0
-  are_all_comparisons_block_true<Torus>(streams, gpu_indexes, gpu_count,
-                                        lwe_array_out, lwe_array_in, mem_ptr,
-                                        bsks, ksks, num_radix_blocks);
-}
-
-template <typename Torus>
-__host__ void host_integer_is_at_least_one_comparisons_block_true_kb(
-    cudaStream_t const *streams, uint32_t const *gpu_indexes,
-    uint32_t gpu_count, Torus *lwe_array_out, Torus const *lwe_array_in,
-    int_comparison_buffer<Torus> *mem_ptr, void *const *bsks,
-    Torus *const *ksks, uint32_t num_radix_blocks) {
-
-  auto eq_buffer = mem_ptr->eq_buffer;
-
-  // It returns a block encrypting 1 if all input blocks are 1
-  // otherwise the block encrypts 0
-  is_at_least_one_comparisons_block_true<Torus>(
-      streams, gpu_indexes, gpu_count, lwe_array_out, lwe_array_in, mem_ptr,
-      bsks, ksks, num_radix_blocks);
-}
 #endif

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

@@ -2,7 +2,6 @@
 #define CUDA_INTEGER_COMPRESSION_CUH
 
 #include "ciphertext.h"
-#include "crypto/fast_packing_keyswitch.cuh"
 #include "crypto/keyswitch.cuh"
 #include "device.h"
 #include "integer/compression/compression.h"
@@ -117,21 +116,11 @@ host_integer_compress(cudaStream_t const *streams, uint32_t const *gpu_indexes,
   while (rem_lwes > 0) {
     auto chunk_size = min(rem_lwes, mem_ptr->lwe_per_glwe);
 
-    if (can_use_pks_fast_path(
-            input_lwe_dimension, chunk_size, compression_params.polynomial_size,
-            compression_params.ks_level, compression_params.glwe_dimension)) {
-      host_fast_packing_keyswitch_lwe_list_to_glwe<Torus, ulonglong4>(
-          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);
-    } else {
-      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);
-    }
+    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;
@@ -306,7 +295,7 @@ __host__ void host_integer_decompress(
   extracted_lwe = h_mem_ptr->tmp_extracted_lwe;
 
   // In the case of extracting a single LWE these parameters are dummy
-  uint32_t num_many_lut = 1;
+  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
@@ -322,7 +311,7 @@ __host__ void host_integer_decompress(
         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, num_many_lut, lut_stride);
+        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
@@ -349,7 +338,7 @@ __host__ void host_integer_decompress(
         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, num_many_lut, lut_stride);
+        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>(

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

@@ -2,12 +2,11 @@
 
 void scratch_cuda_integer_div_rem_radix_ciphertext_kb_64(
     void *const *streams, uint32_t const *gpu_indexes, uint32_t gpu_count,
-    bool is_signed, int8_t **mem_ptr, uint32_t glwe_dimension,
-    uint32_t polynomial_size, uint32_t big_lwe_dimension,
-    uint32_t small_lwe_dimension, uint32_t ks_level, uint32_t ks_base_log,
-    uint32_t pbs_level, uint32_t pbs_base_log, uint32_t grouping_factor,
-    uint32_t num_blocks, uint32_t message_modulus, uint32_t carry_modulus,
-    PBS_TYPE pbs_type, bool allocate_gpu_memory) {
+    int8_t **mem_ptr, uint32_t glwe_dimension, uint32_t polynomial_size,
+    uint32_t big_lwe_dimension, uint32_t small_lwe_dimension, uint32_t ks_level,
+    uint32_t ks_base_log, uint32_t pbs_level, uint32_t pbs_base_log,
+    uint32_t grouping_factor, uint32_t num_blocks, uint32_t message_modulus,
+    uint32_t carry_modulus, PBS_TYPE pbs_type, bool allocate_gpu_memory) {
 
   int_radix_params params(pbs_type, glwe_dimension, polynomial_size,
                           big_lwe_dimension, small_lwe_dimension, ks_level,
@@ -15,7 +14,7 @@ void scratch_cuda_integer_div_rem_radix_ciphertext_kb_64(
                           message_modulus, carry_modulus);
 
   scratch_cuda_integer_div_rem_kb<uint64_t>(
-      (cudaStream_t *)(streams), gpu_indexes, gpu_count, is_signed,
+      (cudaStream_t *)(streams), gpu_indexes, gpu_count,
       (int_div_rem_memory<uint64_t> **)mem_ptr, num_blocks, params,
       allocate_gpu_memory);
 }
@@ -23,7 +22,7 @@ void scratch_cuda_integer_div_rem_radix_ciphertext_kb_64(
 void cuda_integer_div_rem_radix_ciphertext_kb_64(
     void *const *streams, uint32_t const *gpu_indexes, uint32_t gpu_count,
     void *quotient, void *remainder, void const *numerator, void const *divisor,
-    bool is_signed, int8_t *mem_ptr, void *const *bsks, void *const *ksks,
+    int8_t *mem_ptr, void *const *bsks, void *const *ksks,
     uint32_t num_blocks) {
 
   auto mem = (int_div_rem_memory<uint64_t> *)mem_ptr;
@@ -32,8 +31,8 @@ void cuda_integer_div_rem_radix_ciphertext_kb_64(
       (cudaStream_t *)(streams), gpu_indexes, gpu_count,
       static_cast<uint64_t *>(quotient), static_cast<uint64_t *>(remainder),
       static_cast<const uint64_t *>(numerator),
-      static_cast<const uint64_t *>(divisor), is_signed, bsks,
-      (uint64_t **)(ksks), mem, num_blocks);
+      static_cast<const uint64_t *>(divisor), bsks, (uint64_t **)(ksks), mem,
+      num_blocks);
 }
 
 void cleanup_cuda_integer_div_rem(void *const *streams,

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

@@ -3,7 +3,6 @@
 
 #include "crypto/keyswitch.cuh"
 #include "device.h"
-#include "integer/abs.cuh"
 #include "integer/comparison.cuh"
 #include "integer/integer.cuh"
 #include "integer/integer_utilities.h"
@@ -162,21 +161,22 @@ template <typename Torus> struct lwe_ciphertext_list {
 template <typename Torus>
 __host__ void scratch_cuda_integer_div_rem_kb(
     cudaStream_t const *streams, uint32_t const *gpu_indexes,
-    uint32_t gpu_count, bool is_signed, int_div_rem_memory<Torus> **mem_ptr,
+    uint32_t gpu_count, int_div_rem_memory<Torus> **mem_ptr,
     uint32_t num_blocks, int_radix_params params, bool allocate_gpu_memory) {
 
-  *mem_ptr =
-      new int_div_rem_memory<Torus>(streams, gpu_indexes, gpu_count, params,
-                                    is_signed, num_blocks, allocate_gpu_memory);
+  *mem_ptr = new int_div_rem_memory<Torus>(
+      streams, gpu_indexes, gpu_count, params, num_blocks, allocate_gpu_memory);
 }
 
 template <typename Torus>
-__host__ void host_unsigned_integer_div_rem_kb(
-    cudaStream_t const *streams, uint32_t const *gpu_indexes,
-    uint32_t gpu_count, Torus *quotient, Torus *remainder,
-    Torus const *numerator, Torus const *divisor, void *const *bsks,
-    uint64_t *const *ksks, unsigned_int_div_rem_memory<uint64_t> *mem_ptr,
-    uint32_t num_blocks) {
+__host__ void host_integer_div_rem_kb(cudaStream_t const *streams,
+                                      uint32_t const *gpu_indexes,
+                                      uint32_t gpu_count, Torus *quotient,
+                                      Torus *remainder, Torus const *numerator,
+                                      Torus const *divisor, void *const *bsks,
+                                      uint64_t *const *ksks,
+                                      int_div_rem_memory<uint64_t> *mem_ptr,
+                                      uint32_t num_blocks) {
 
   auto radix_params = mem_ptr->params;
 
@@ -425,24 +425,11 @@ __host__ void host_unsigned_integer_div_rem_kb(
     auto do_overflowing_sub = [&](cudaStream_t const *streams,
                                   uint32_t const *gpu_indexes,
                                   uint32_t gpu_count) {
-      uint32_t compute_borrow = 1;
-      uint32_t uses_input_borrow = 0;
-      auto first_indexes = mem_ptr->first_indexes_for_overflow_sub
-                               [merged_interesting_remainder.len - 1];
-      auto second_indexes = mem_ptr->second_indexes_for_overflow_sub
-                                [merged_interesting_remainder.len - 1];
-      auto scalar_indexes =
-          mem_ptr
-              ->scalars_for_overflow_sub[merged_interesting_remainder.len - 1];
-      mem_ptr->overflow_sub_mem->update_lut_indexes(
-          streams, gpu_indexes, first_indexes, second_indexes, scalar_indexes,
-          merged_interesting_remainder.len);
-      host_integer_overflowing_sub<uint64_t>(
+      host_integer_overflowing_sub_kb<Torus>(
           streams, gpu_indexes, gpu_count, new_remainder.data,
-          (uint64_t *)merged_interesting_remainder.data,
-          interesting_divisor.data, subtraction_overflowed.data,
-          (const Torus *)nullptr, mem_ptr->overflow_sub_mem, bsks, ksks,
-          merged_interesting_remainder.len, compute_borrow, uses_input_borrow);
+          subtraction_overflowed.data, merged_interesting_remainder.data,
+          interesting_divisor.data, bsks, ksks, mem_ptr->overflow_sub_mem,
+          merged_interesting_remainder.len);
     };
 
     // fills:
@@ -607,108 +594,4 @@ __host__ void host_unsigned_integer_div_rem_kb(
   }
 }
 
-template <typename Torus>
-__host__ void host_integer_div_rem_kb(cudaStream_t const *streams,
-                                      uint32_t const *gpu_indexes,
-                                      uint32_t gpu_count, Torus *quotient,
-                                      Torus *remainder, Torus const *numerator,
-                                      Torus const *divisor, bool is_signed,
-                                      void *const *bsks, uint64_t *const *ksks,
-                                      int_div_rem_memory<uint64_t> *int_mem_ptr,
-                                      uint32_t num_blocks) {
-
-  if (is_signed) {
-    auto radix_params = int_mem_ptr->params;
-    uint32_t big_lwe_size = radix_params.big_lwe_dimension + 1;
-
-    // temporary memory
-    lwe_ciphertext_list<Torus> positive_numerator(
-        int_mem_ptr->positive_numerator, radix_params, num_blocks);
-    lwe_ciphertext_list<Torus> positive_divisor(int_mem_ptr->positive_divisor,
-                                                radix_params, num_blocks);
-
-    positive_numerator.clone_from((Torus *)numerator, 0, num_blocks - 1,
-                                  streams[0], gpu_indexes[0]);
-    positive_divisor.clone_from((Torus *)divisor, 0, num_blocks - 1, streams[0],
-                                gpu_indexes[0]);
-
-    for (uint j = 0; j < gpu_count; j++) {
-      cuda_synchronize_stream(streams[j], gpu_indexes[j]);
-    }
-
-    host_integer_abs_kb<Torus>(int_mem_ptr->sub_streams_1, gpu_indexes,
-                               gpu_count, positive_numerator.data, bsks, ksks,
-                               int_mem_ptr->abs_mem_1, true, num_blocks);
-    host_integer_abs_kb<Torus>(int_mem_ptr->sub_streams_2, gpu_indexes,
-                               gpu_count, positive_divisor.data, bsks, ksks,
-                               int_mem_ptr->abs_mem_2, true, num_blocks);
-    for (uint j = 0; j < int_mem_ptr->active_gpu_count; j++) {
-      cuda_synchronize_stream(int_mem_ptr->sub_streams_1[j], gpu_indexes[j]);
-      cuda_synchronize_stream(int_mem_ptr->sub_streams_2[j], gpu_indexes[j]);
-    }
-
-    host_unsigned_integer_div_rem_kb<Torus>(
-        int_mem_ptr->sub_streams_1, gpu_indexes, gpu_count, quotient, remainder,
-        positive_numerator.data, positive_divisor.data, bsks, ksks,
-        int_mem_ptr->unsigned_mem, num_blocks);
-
-    integer_radix_apply_bivariate_lookup_table_kb<Torus>(
-        int_mem_ptr->sub_streams_2, gpu_indexes, gpu_count,
-        int_mem_ptr->sign_bits_are_different,
-        &numerator[big_lwe_size * (num_blocks - 1)],
-        &divisor[big_lwe_size * (num_blocks - 1)], bsks, ksks, 1,
-        int_mem_ptr->compare_signed_bits_lut,
-        int_mem_ptr->compare_signed_bits_lut->params.message_modulus);
-
-    for (uint j = 0; j < int_mem_ptr->active_gpu_count; j++) {
-      cuda_synchronize_stream(int_mem_ptr->sub_streams_1[j], gpu_indexes[j]);
-      cuda_synchronize_stream(int_mem_ptr->sub_streams_2[j], gpu_indexes[j]);
-    }
-
-    host_integer_radix_negation(
-        int_mem_ptr->sub_streams_1, gpu_indexes, gpu_count,
-        int_mem_ptr->negated_quotient, quotient, radix_params.big_lwe_dimension,
-        num_blocks, radix_params.message_modulus, radix_params.carry_modulus);
-
-    uint32_t requested_flag = outputFlag::FLAG_NONE;
-    uint32_t uses_carry = 0;
-    host_propagate_single_carry<Torus>(
-        int_mem_ptr->sub_streams_1, gpu_indexes, gpu_count,
-        int_mem_ptr->negated_quotient, nullptr, nullptr, int_mem_ptr->scp_mem_1,
-        bsks, ksks, num_blocks, requested_flag, uses_carry);
-
-    host_integer_radix_negation(int_mem_ptr->sub_streams_2, gpu_indexes,
-                                gpu_count, int_mem_ptr->negated_remainder,
-                                remainder, radix_params.big_lwe_dimension,
-                                num_blocks, radix_params.message_modulus,
-                                radix_params.carry_modulus);
-
-    host_propagate_single_carry<Torus>(
-        int_mem_ptr->sub_streams_2, gpu_indexes, gpu_count,
-        int_mem_ptr->negated_remainder, nullptr, nullptr,
-        int_mem_ptr->scp_mem_2, bsks, ksks, num_blocks, requested_flag,
-        uses_carry);
-
-    host_integer_radix_cmux_kb<Torus>(
-        int_mem_ptr->sub_streams_1, gpu_indexes, gpu_count, quotient,
-        int_mem_ptr->sign_bits_are_different, int_mem_ptr->negated_quotient,
-        quotient, int_mem_ptr->cmux_quotient_mem, bsks, ksks, num_blocks);
-
-    host_integer_radix_cmux_kb<Torus>(
-        int_mem_ptr->sub_streams_2, gpu_indexes, gpu_count, remainder,
-        &numerator[big_lwe_size * (num_blocks - 1)],
-        int_mem_ptr->negated_remainder, remainder,
-        int_mem_ptr->cmux_remainder_mem, bsks, ksks, num_blocks);
-
-    for (uint j = 0; j < int_mem_ptr->active_gpu_count; j++) {
-      cuda_synchronize_stream(int_mem_ptr->sub_streams_1[j], gpu_indexes[j]);
-      cuda_synchronize_stream(int_mem_ptr->sub_streams_2[j], gpu_indexes[j]);
-    }
-  } else {
-    host_unsigned_integer_div_rem_kb<Torus>(
-        streams, gpu_indexes, gpu_count, quotient, remainder, numerator,
-        divisor, bsks, ksks, int_mem_ptr->unsigned_mem, num_blocks);
-  }
-}
-
 #endif // TFHE_RS_DIV_REM_CUH

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

@@ -1,5 +1,4 @@
 #include "integer/integer.cuh"
-#include "integer/negation.cuh"
 #include <linear_algebra.h>
 
 void cuda_full_propagation_64_inplace(void *const *streams,
@@ -50,8 +49,7 @@ void scratch_cuda_propagate_single_carry_kb_64_inplace(
     uint32_t big_lwe_dimension, uint32_t small_lwe_dimension, uint32_t ks_level,
     uint32_t ks_base_log, uint32_t pbs_level, uint32_t pbs_base_log,
     uint32_t grouping_factor, uint32_t num_blocks, uint32_t message_modulus,
-    uint32_t carry_modulus, PBS_TYPE pbs_type, uint32_t requested_flag,
-    uint32_t uses_carry, bool allocate_gpu_memory) {
+    uint32_t carry_modulus, PBS_TYPE pbs_type, bool allocate_gpu_memory) {
 
   int_radix_params params(pbs_type, glwe_dimension, polynomial_size,
                           big_lwe_dimension, small_lwe_dimension, ks_level,
@@ -61,94 +59,30 @@ void scratch_cuda_propagate_single_carry_kb_64_inplace(
   scratch_cuda_propagate_single_carry_kb_inplace<uint64_t>(
       (cudaStream_t *)(streams), gpu_indexes, gpu_count,
       (int_sc_prop_memory<uint64_t> **)mem_ptr, num_blocks, params,
-      requested_flag, uses_carry, allocate_gpu_memory);
-}
-
-void scratch_cuda_add_and_propagate_single_carry_kb_64_inplace(
-    void *const *streams, uint32_t const *gpu_indexes, uint32_t gpu_count,
-    int8_t **mem_ptr, uint32_t glwe_dimension, uint32_t polynomial_size,
-    uint32_t big_lwe_dimension, uint32_t small_lwe_dimension, uint32_t ks_level,
-    uint32_t ks_base_log, uint32_t pbs_level, uint32_t pbs_base_log,
-    uint32_t grouping_factor, uint32_t num_blocks, uint32_t message_modulus,
-    uint32_t carry_modulus, PBS_TYPE pbs_type, uint32_t requested_flag,
-    uint32_t uses_carry, bool allocate_gpu_memory) {
-
-  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_propagate_single_carry_kb_inplace<uint64_t>(
-      (cudaStream_t *)(streams), gpu_indexes, gpu_count,
-      (int_sc_prop_memory<uint64_t> **)mem_ptr, num_blocks, params,
-      requested_flag, uses_carry, allocate_gpu_memory);
-}
-
-void scratch_cuda_integer_overflowing_sub_kb_64_inplace(
-    void *const *streams, uint32_t const *gpu_indexes, uint32_t gpu_count,
-    int8_t **mem_ptr, uint32_t glwe_dimension, uint32_t polynomial_size,
-    uint32_t big_lwe_dimension, uint32_t small_lwe_dimension, uint32_t ks_level,
-    uint32_t ks_base_log, uint32_t pbs_level, uint32_t pbs_base_log,
-    uint32_t grouping_factor, uint32_t num_blocks, uint32_t message_modulus,
-    uint32_t carry_modulus, PBS_TYPE pbs_type, uint32_t compute_overflow,
-    bool allocate_gpu_memory) {
-
-  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_overflowing_sub<uint64_t>(
-      (cudaStream_t *)(streams), gpu_indexes, gpu_count,
-      (int_borrow_prop_memory<uint64_t> **)mem_ptr, num_blocks, params,
-      compute_overflow, allocate_gpu_memory);
+      allocate_gpu_memory);
 }
 
 void cuda_propagate_single_carry_kb_64_inplace(
     void *const *streams, uint32_t const *gpu_indexes, uint32_t gpu_count,
-    void *lwe_array, void *carry_out, const void *carry_in, int8_t *mem_ptr,
-    void *const *bsks, void *const *ksks, uint32_t num_blocks,
-    uint32_t requested_flag, uint32_t uses_carry) {
-
+    void *lwe_array, void *carry_out, int8_t *mem_ptr, void *const *bsks,
+    void *const *ksks, uint32_t num_blocks) {
   host_propagate_single_carry<uint64_t>(
       (cudaStream_t *)(streams), gpu_indexes, gpu_count,
       static_cast<uint64_t *>(lwe_array), static_cast<uint64_t *>(carry_out),
-      static_cast<const uint64_t *>(carry_in),
-      (int_sc_prop_memory<uint64_t> *)mem_ptr, bsks, (uint64_t **)(ksks),
-      num_blocks, requested_flag, uses_carry);
+      nullptr, (int_sc_prop_memory<uint64_t> *)mem_ptr, bsks,
+      (uint64_t **)(ksks), num_blocks);
 }
 
-void cuda_add_and_propagate_single_carry_kb_64_inplace(
+void cuda_propagate_single_carry_get_input_carries_kb_64_inplace(
     void *const *streams, uint32_t const *gpu_indexes, uint32_t gpu_count,
-    void *lhs_array, const void *rhs_array, void *carry_out,
-    const void *carry_in, int8_t *mem_ptr, void *const *bsks, void *const *ksks,
-    uint32_t num_blocks, uint32_t requested_flag, uint32_t uses_carry) {
-
-  host_add_and_propagate_single_carry<uint64_t>(
+    void *lwe_array, void *carry_out, void *input_carries, int8_t *mem_ptr,
+    void *const *bsks, void *const *ksks, uint32_t num_blocks) {
+  host_propagate_single_carry<uint64_t>(
       (cudaStream_t *)(streams), gpu_indexes, gpu_count,
-      static_cast<uint64_t *>(lhs_array),
-      static_cast<const uint64_t *>(rhs_array),
-      static_cast<uint64_t *>(carry_out),
-      static_cast<const uint64_t *>(carry_in),
+      static_cast<uint64_t *>(lwe_array), static_cast<uint64_t *>(carry_out),
+      static_cast<uint64_t *>(input_carries),
       (int_sc_prop_memory<uint64_t> *)mem_ptr, bsks, (uint64_t **)(ksks),
-      num_blocks, requested_flag, uses_carry);
-}
-
-void cuda_integer_overflowing_sub_kb_64_inplace(
-    void *const *streams, uint32_t const *gpu_indexes, uint32_t gpu_count,
-    void *lhs_array, const void *rhs_array, void *overflow_block,
-    const void *input_borrow, int8_t *mem_ptr, void *const *bsks,
-    void *const *ksks, uint32_t num_blocks, uint32_t compute_overflow,
-    uint32_t uses_input_borrow) {
-
-  host_integer_overflowing_sub<uint64_t>(
-      (cudaStream_t const *)streams, gpu_indexes, gpu_count,
-      static_cast<uint64_t *>(lhs_array), static_cast<uint64_t *>(lhs_array),
-      static_cast<const uint64_t *>(rhs_array),
-      static_cast<uint64_t *>(overflow_block),
-      static_cast<const uint64_t *>(input_borrow),
-      (int_borrow_prop_memory<uint64_t> *)mem_ptr, bsks, (uint64_t **)ksks,
-      num_blocks, compute_overflow, uses_input_borrow);
+      num_blocks);
 }
 
 void cleanup_cuda_propagate_single_carry(void *const *streams,
@@ -160,23 +94,6 @@ void cleanup_cuda_propagate_single_carry(void *const *streams,
   mem_ptr->release((cudaStream_t *)(streams), gpu_indexes, gpu_count);
 }
 
-void cleanup_cuda_add_and_propagate_single_carry(void *const *streams,
-                                                 uint32_t const *gpu_indexes,
-                                                 uint32_t gpu_count,
-                                                 int8_t **mem_ptr_void) {
-  int_sc_prop_memory<uint64_t> *mem_ptr =
-      (int_sc_prop_memory<uint64_t> *)(*mem_ptr_void);
-  mem_ptr->release((cudaStream_t *)(streams), gpu_indexes, gpu_count);
-}
-void cleanup_cuda_integer_overflowing_sub(void *const *streams,
-                                          uint32_t const *gpu_indexes,
-                                          uint32_t gpu_count,
-                                          int8_t **mem_ptr_void) {
-  int_borrow_prop_memory<uint64_t> *mem_ptr =
-      (int_borrow_prop_memory<uint64_t> *)(*mem_ptr_void);
-  mem_ptr->release((cudaStream_t *)(streams), gpu_indexes, gpu_count);
-}
-
 void scratch_cuda_apply_univariate_lut_kb_64(
     void *const *streams, uint32_t const *gpu_indexes, uint32_t gpu_count,
     int8_t **mem_ptr, void const *input_lut, uint32_t lwe_dimension,
@@ -198,27 +115,6 @@ void scratch_cuda_apply_univariate_lut_kb_64(
       allocate_gpu_memory);
 }
 
-void scratch_cuda_apply_many_univariate_lut_kb_64(
-    void *const *streams, uint32_t const *gpu_indexes, uint32_t gpu_count,
-    int8_t **mem_ptr, void const *input_lut, uint32_t lwe_dimension,
-    uint32_t glwe_dimension, uint32_t polynomial_size, uint32_t ks_level,
-    uint32_t ks_base_log, uint32_t pbs_level, uint32_t pbs_base_log,
-    uint32_t grouping_factor, uint32_t num_radix_blocks,
-    uint32_t message_modulus, uint32_t carry_modulus, PBS_TYPE pbs_type,
-    uint32_t num_many_lut, bool allocate_gpu_memory) {
-
-  int_radix_params params(pbs_type, glwe_dimension, polynomial_size,
-                          glwe_dimension * polynomial_size, lwe_dimension,
-                          ks_level, ks_base_log, pbs_level, pbs_base_log,
-                          grouping_factor, message_modulus, carry_modulus);
-
-  scratch_cuda_apply_many_univariate_lut_kb<uint64_t>(
-      (cudaStream_t *)(streams), gpu_indexes, gpu_count,
-      (int_radix_lut<uint64_t> **)mem_ptr,
-      static_cast<const uint64_t *>(input_lut), num_radix_blocks, params,
-      num_many_lut, allocate_gpu_memory);
-}
-
 void cuda_apply_univariate_lut_kb_64(void *const *streams,
                                      uint32_t const *gpu_indexes,
                                      uint32_t gpu_count, void *output_radix_lwe,
@@ -246,19 +142,19 @@ void cuda_apply_many_univariate_lut_kb_64(
     void *const *streams, uint32_t const *gpu_indexes, uint32_t gpu_count,
     void *output_radix_lwe, void const *input_radix_lwe, int8_t *mem_ptr,
     void *const *ksks, void *const *bsks, uint32_t num_blocks,
-    uint32_t num_many_lut, uint32_t lut_stride) {
+    uint32_t lut_count, uint32_t lut_stride) {
 
   host_apply_many_univariate_lut_kb<uint64_t>(
       (cudaStream_t *)(streams), gpu_indexes, gpu_count,
       static_cast<uint64_t *>(output_radix_lwe),
       static_cast<const uint64_t *>(input_radix_lwe),
       (int_radix_lut<uint64_t> *)mem_ptr, (uint64_t **)(ksks), bsks, num_blocks,
-      num_many_lut, lut_stride);
+      lut_count, lut_stride);
 }
 
 void scratch_cuda_apply_bivariate_lut_kb_64(
     void *const *streams, uint32_t const *gpu_indexes, uint32_t gpu_count,
-    int8_t **mem_ptr, void const *input_lut, uint32_t lwe_dimension,
+    int8_t **mem_ptr, void *input_lut, uint32_t lwe_dimension,
     uint32_t glwe_dimension, uint32_t polynomial_size, uint32_t ks_level,
     uint32_t ks_base_log, uint32_t pbs_level, uint32_t pbs_base_log,
     uint32_t grouping_factor, uint32_t num_radix_blocks,
@@ -272,9 +168,8 @@ void scratch_cuda_apply_bivariate_lut_kb_64(
 
   scratch_cuda_apply_bivariate_lut_kb<uint64_t>(
       (cudaStream_t *)(streams), gpu_indexes, gpu_count,
-      (int_radix_lut<uint64_t> **)mem_ptr,
-      static_cast<const uint64_t *>(input_lut), num_radix_blocks, params,
-      allocate_gpu_memory);
+      (int_radix_lut<uint64_t> **)mem_ptr, static_cast<uint64_t *>(input_lut),
+      num_radix_blocks, params, allocate_gpu_memory);
 }
 
 void cuda_apply_bivariate_lut_kb_64(

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

@@ -67,12 +67,11 @@ void generate_ids_update_degrees(int *terms_degree, size_t *h_lwe_idx_in,
  */
 void scratch_cuda_integer_mult_radix_ciphertext_kb_64(
     void *const *streams, uint32_t const *gpu_indexes, uint32_t gpu_count,
-    int8_t **mem_ptr, bool const is_boolean_left, bool const is_boolean_right,
-    uint32_t message_modulus, uint32_t carry_modulus, uint32_t glwe_dimension,
-    uint32_t lwe_dimension, uint32_t polynomial_size, uint32_t pbs_base_log,
-    uint32_t pbs_level, uint32_t ks_base_log, uint32_t ks_level,
-    uint32_t grouping_factor, uint32_t num_radix_blocks, PBS_TYPE pbs_type,
-    bool allocate_gpu_memory) {
+    int8_t **mem_ptr, uint32_t message_modulus, uint32_t carry_modulus,
+    uint32_t glwe_dimension, uint32_t lwe_dimension, uint32_t polynomial_size,
+    uint32_t pbs_base_log, uint32_t pbs_level, uint32_t ks_base_log,
+    uint32_t ks_level, uint32_t grouping_factor, uint32_t num_radix_blocks,
+    PBS_TYPE pbs_type, bool allocate_gpu_memory) {
 
   int_radix_params params(pbs_type, glwe_dimension, polynomial_size,
                           polynomial_size * glwe_dimension, lwe_dimension,
@@ -89,8 +88,8 @@ void scratch_cuda_integer_mult_radix_ciphertext_kb_64(
   case 16384:
     scratch_cuda_integer_mult_radix_ciphertext_kb<uint64_t>(
         (cudaStream_t const *)(streams), gpu_indexes, gpu_count,
-        (int_mul_memory<uint64_t> **)mem_ptr, is_boolean_left, is_boolean_right,
-        num_radix_blocks, params, allocate_gpu_memory);
+        (int_mul_memory<uint64_t> **)mem_ptr, num_radix_blocks, params,
+        allocate_gpu_memory);
     break;
   default:
     PANIC("Cuda error (integer multiplication): unsupported polynomial size. "
@@ -127,66 +126,65 @@ void scratch_cuda_integer_mult_radix_ciphertext_kb_64(
  */
 void cuda_integer_mult_radix_ciphertext_kb_64(
     void *const *streams, uint32_t const *gpu_indexes, uint32_t gpu_count,
-    void *radix_lwe_out, void const *radix_lwe_left, bool const is_bool_left,
-    void const *radix_lwe_right, bool const is_bool_right, void *const *bsks,
-    void *const *ksks, int8_t *mem_ptr, uint32_t polynomial_size,
-    uint32_t num_blocks) {
+    void *radix_lwe_out, void const *radix_lwe_left,
+    void const *radix_lwe_right, void *const *bsks, void *const *ksks,
+    int8_t *mem_ptr, uint32_t polynomial_size, uint32_t num_blocks) {
 
   switch (polynomial_size) {
   case 256:
     host_integer_mult_radix_kb<uint64_t, AmortizedDegree<256>>(
         (cudaStream_t *)(streams), gpu_indexes, gpu_count,
         static_cast<uint64_t *>(radix_lwe_out),
-        static_cast<const uint64_t *>(radix_lwe_left), is_bool_left,
-        static_cast<const uint64_t *>(radix_lwe_right), is_bool_right, bsks,
+        static_cast<const uint64_t *>(radix_lwe_left),
+        static_cast<const uint64_t *>(radix_lwe_right), bsks,
         (uint64_t **)(ksks), (int_mul_memory<uint64_t> *)mem_ptr, num_blocks);
     break;
   case 512:
     host_integer_mult_radix_kb<uint64_t, AmortizedDegree<512>>(
         (cudaStream_t *)(streams), gpu_indexes, gpu_count,
         static_cast<uint64_t *>(radix_lwe_out),
-        static_cast<const uint64_t *>(radix_lwe_left), is_bool_left,
-        static_cast<const uint64_t *>(radix_lwe_right), is_bool_right, bsks,
+        static_cast<const uint64_t *>(radix_lwe_left),
+        static_cast<const uint64_t *>(radix_lwe_right), bsks,
         (uint64_t **)(ksks), (int_mul_memory<uint64_t> *)mem_ptr, num_blocks);
     break;
   case 1024:
     host_integer_mult_radix_kb<uint64_t, AmortizedDegree<1024>>(
         (cudaStream_t *)(streams), gpu_indexes, gpu_count,
         static_cast<uint64_t *>(radix_lwe_out),
-        static_cast<const uint64_t *>(radix_lwe_left), is_bool_left,
-        static_cast<const uint64_t *>(radix_lwe_right), is_bool_right, bsks,
+        static_cast<const uint64_t *>(radix_lwe_left),
+        static_cast<const uint64_t *>(radix_lwe_right), bsks,
         (uint64_t **)(ksks), (int_mul_memory<uint64_t> *)mem_ptr, num_blocks);
     break;
   case 2048:
     host_integer_mult_radix_kb<uint64_t, AmortizedDegree<2048>>(
         (cudaStream_t *)(streams), gpu_indexes, gpu_count,
         static_cast<uint64_t *>(radix_lwe_out),
-        static_cast<const uint64_t *>(radix_lwe_left), is_bool_left,
-        static_cast<const uint64_t *>(radix_lwe_right), is_bool_right, bsks,
+        static_cast<const uint64_t *>(radix_lwe_left),
+        static_cast<const uint64_t *>(radix_lwe_right), bsks,
         (uint64_t **)(ksks), (int_mul_memory<uint64_t> *)mem_ptr, num_blocks);
     break;
   case 4096:
     host_integer_mult_radix_kb<uint64_t, AmortizedDegree<4096>>(
         (cudaStream_t *)(streams), gpu_indexes, gpu_count,
         static_cast<uint64_t *>(radix_lwe_out),
-        static_cast<const uint64_t *>(radix_lwe_left), is_bool_left,
-        static_cast<const uint64_t *>(radix_lwe_right), is_bool_right, bsks,
+        static_cast<const uint64_t *>(radix_lwe_left),
+        static_cast<const uint64_t *>(radix_lwe_right), bsks,
         (uint64_t **)(ksks), (int_mul_memory<uint64_t> *)mem_ptr, num_blocks);
     break;
   case 8192:
     host_integer_mult_radix_kb<uint64_t, AmortizedDegree<8192>>(
         (cudaStream_t *)(streams), gpu_indexes, gpu_count,
         static_cast<uint64_t *>(radix_lwe_out),
-        static_cast<const uint64_t *>(radix_lwe_left), is_bool_left,
-        static_cast<const uint64_t *>(radix_lwe_right), is_bool_right, bsks,
+        static_cast<const uint64_t *>(radix_lwe_left),
+        static_cast<const uint64_t *>(radix_lwe_right), bsks,
         (uint64_t **)(ksks), (int_mul_memory<uint64_t> *)mem_ptr, num_blocks);
     break;
   case 16384:
     host_integer_mult_radix_kb<uint64_t, AmortizedDegree<16384>>(
         (cudaStream_t *)(streams), gpu_indexes, gpu_count,
         static_cast<uint64_t *>(radix_lwe_out),
-        static_cast<const uint64_t *>(radix_lwe_left), is_bool_left,
-        static_cast<const uint64_t *>(radix_lwe_right), is_bool_right, bsks,
+        static_cast<const uint64_t *>(radix_lwe_left),
+        static_cast<const uint64_t *>(radix_lwe_right), bsks,
         (uint64_t **)(ksks), (int_mul_memory<uint64_t> *)mem_ptr, num_blocks);
     break;
   default:

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

@@ -9,7 +9,6 @@
 #include "crypto/keyswitch.cuh"
 #include "device.h"
 #include "helper_multi_gpu.h"
-#include "integer/cmux.cuh"
 #include "integer/integer.cuh"
 #include "integer/integer_utilities.h"
 #include "linear_algebra.h"
@@ -209,7 +208,7 @@ __host__ void host_integer_partial_sum_ciphertexts_vec_kb(
   auto small_lwe_size = small_lwe_dimension + 1;
 
   // In the case of extracting a single LWE this parameters are dummy
-  uint32_t num_many_lut = 1;
+  uint32_t lut_count = 1;
   uint32_t lut_stride = 0;
 
   if (num_radix_in_vec == 0)
@@ -267,8 +266,8 @@ __host__ void host_integer_partial_sum_ciphertexts_vec_kb(
         streams, gpu_indexes, gpu_count, mem_ptr->params, 2,
         2 * ch_amount * num_blocks, reused_lut);
   }
-  auto message_acc = luts_message_carry->get_lut(0, 0);
-  auto carry_acc = luts_message_carry->get_lut(0, 1);
+  auto message_acc = luts_message_carry->get_lut(gpu_indexes[0], 0);
+  auto carry_acc = luts_message_carry->get_lut(gpu_indexes[0], 1);
 
   // define functions for each accumulator
   auto lut_f_message = [message_modulus](Torus x) -> Torus {
@@ -285,7 +284,7 @@ __host__ void host_integer_partial_sum_ciphertexts_vec_kb(
   generate_device_accumulator<Torus>(
       streams[0], gpu_indexes[0], carry_acc, glwe_dimension, polynomial_size,
       message_modulus, carry_modulus, lut_f_carry);
-  luts_message_carry->broadcast_lut(streams, gpu_indexes, 0);
+  luts_message_carry->broadcast_lut(streams, gpu_indexes, gpu_indexes[0]);
 
   while (r > 2) {
     size_t cur_total_blocks = r * num_blocks;
@@ -334,10 +333,10 @@ __host__ void host_integer_partial_sum_ciphertexts_vec_kb(
     if (carry_count > 0)
       cuda_set_value_async<Torus>(
           streams[0], gpu_indexes[0],
-          luts_message_carry->get_lut_indexes(0, message_count), 1,
+          luts_message_carry->get_lut_indexes(gpu_indexes[0], message_count), 1,
           carry_count);
 
-    luts_message_carry->broadcast_lut(streams, gpu_indexes, 0);
+    luts_message_carry->broadcast_lut(streams, gpu_indexes, gpu_indexes[0]);
 
     /// For multi GPU execution we create vectors of pointers for inputs and
     /// outputs
@@ -370,7 +369,7 @@ __host__ void host_integer_partial_sum_ciphertexts_vec_kb(
           glwe_dimension, small_lwe_dimension, polynomial_size,
           mem_ptr->params.pbs_base_log, mem_ptr->params.pbs_level,
           mem_ptr->params.grouping_factor, total_count,
-          mem_ptr->params.pbs_type, num_many_lut, lut_stride);
+          mem_ptr->params.pbs_type, lut_count, lut_stride);
     } else {
       cuda_synchronize_stream(streams[0], gpu_indexes[0]);
 
@@ -418,7 +417,7 @@ __host__ void host_integer_partial_sum_ciphertexts_vec_kb(
           glwe_dimension, small_lwe_dimension, polynomial_size,
           mem_ptr->params.pbs_base_log, mem_ptr->params.pbs_level,
           mem_ptr->params.grouping_factor, total_count,
-          mem_ptr->params.pbs_type, num_many_lut, lut_stride);
+          mem_ptr->params.pbs_type, lut_count, lut_stride);
 
       multi_gpu_gather_lwe_async<Torus>(
           streams, gpu_indexes, active_gpu_count, new_blocks, lwe_after_pbs_vec,
@@ -454,8 +453,7 @@ template <typename Torus, class params>
 __host__ void host_integer_mult_radix_kb(
     cudaStream_t const *streams, uint32_t const *gpu_indexes,
     uint32_t gpu_count, uint64_t *radix_lwe_out, uint64_t const *radix_lwe_left,
-    bool const is_bool_left, uint64_t const *radix_lwe_right,
-    bool const is_bool_right, void *const *bsks, uint64_t *const *ksks,
+    uint64_t const *radix_lwe_right, void *const *bsks, uint64_t *const *ksks,
     int_mul_memory<Torus> *mem_ptr, uint32_t num_blocks) {
 
   auto glwe_dimension = mem_ptr->params.glwe_dimension;
@@ -466,20 +464,6 @@ __host__ void host_integer_mult_radix_kb(
 
   int big_lwe_dimension = glwe_dimension * polynomial_size;
 
-  if (is_bool_right) {
-    zero_out_if<Torus>(streams, gpu_indexes, gpu_count, radix_lwe_out,
-                       radix_lwe_left, radix_lwe_right, mem_ptr->zero_out_mem,
-                       mem_ptr->zero_out_predicate_lut, bsks, ksks, num_blocks);
-    return;
-  }
-
-  if (is_bool_left) {
-    zero_out_if<Torus>(streams, gpu_indexes, gpu_count, radix_lwe_out,
-                       radix_lwe_right, radix_lwe_left, mem_ptr->zero_out_mem,
-                       mem_ptr->zero_out_predicate_lut, bsks, ksks, num_blocks);
-    return;
-  }
-
   // 'vector_result_lsb' contains blocks from all possible right shifts of
   // radix_lwe_left, only nonzero blocks are kept
   int lsb_vector_block_count = num_blocks * (num_blocks + 1) / 2;
@@ -578,26 +562,19 @@ __host__ void host_integer_mult_radix_kb(
       terms_degree, bsks, ksks, mem_ptr->sum_ciphertexts_mem, num_blocks,
       2 * num_blocks, mem_ptr->luts_array);
 
-  uint32_t block_modulus = message_modulus * carry_modulus;
-  uint32_t num_bits_in_block = log2_int(block_modulus);
-
-  auto scp_mem_ptr = mem_ptr->sc_prop_mem;
-  uint32_t requested_flag = outputFlag::FLAG_NONE;
-  uint32_t uses_carry = 0;
-  host_propagate_single_carry<Torus>(
-      streams, gpu_indexes, gpu_count, radix_lwe_out, nullptr, nullptr,
-      scp_mem_ptr, bsks, ksks, num_blocks, requested_flag, uses_carry);
+  auto scp_mem_ptr = mem_ptr->sum_ciphertexts_mem->scp_mem;
+  host_propagate_single_carry<Torus>(streams, gpu_indexes, gpu_count,
+                                     radix_lwe_out, nullptr, nullptr,
+                                     scp_mem_ptr, bsks, ksks, num_blocks);
 }
 
 template <typename Torus>
 __host__ void scratch_cuda_integer_mult_radix_ciphertext_kb(
     cudaStream_t const *streams, uint32_t const *gpu_indexes,
     uint32_t gpu_count, int_mul_memory<Torus> **mem_ptr,
-    bool const is_boolean_left, bool const is_boolean_right,
     uint32_t num_radix_blocks, int_radix_params params,
     bool allocate_gpu_memory) {
   *mem_ptr = new int_mul_memory<Torus>(streams, gpu_indexes, gpu_count, params,
-                                       is_boolean_left, is_boolean_right,
                                        num_radix_blocks, allocate_gpu_memory);
 }