chore(zk): improve parallelism in prove_impl

Prior to this commit, all generated artifacts would be identified
as integer benchmarks.

.cargo/audit.toml

@@ -2,6 +2,8 @@
 ignore = [
     # Ignoring unmaintained 'paste' advisory as it is a widely used, low-risk build dependency.
     "RUSTSEC-2024-0436",
+    # Ignoring unmaintained 'bincode' crate. Getting rid of it would be too complex on the short term.
+    "RUSTSEC-2025-0141",
 ]
 
 [output]

.github/actions/gpu_setup/action.yml

@@ -23,6 +23,8 @@ runs:
         echo "${CMAKE_SCRIPT_SHA} cmake-${CMAKE_VERSION}-linux-x86_64.sh" > checksum
         sha256sum -c checksum
         sudo bash cmake-"${CMAKE_VERSION}"-linux-x86_64.sh --skip-license --prefix=/usr/ --exclude-subdir
+        sudo apt-get clean
+        sudo rm -rf /var/lib/apt/lists/*
         sudo apt update
         sudo apt remove -y unattended-upgrades
         sudo apt install -y cmake-format libclang-dev

.github/workflows/aws_tfhe_backward_compat_tests.yml

@@ -66,7 +66,7 @@ jobs:
       cancel-in-progress: ${{ github.ref != 'refs/heads/main' }}
     steps:
       - name: Checkout tfhe-rs
-        uses: actions/checkout@1af3b93b6815bc44a9784bd300feb67ff0d1eeb3
+        uses: actions/checkout@8e8c483db84b4bee98b60c0593521ed34d9990e8
         with:
           persist-credentials: 'true' # Needed to pull lfs data
           token: ${{ env.CHECKOUT_TOKEN }}
@@ -80,7 +80,7 @@ jobs:
 
       - name: Retrieve data from cache
         id: retrieve-data-cache
-        uses: actions/cache/restore@0057852bfaa89a56745cba8c7296529d2fc39830 #v4.3.0
+        uses: actions/cache/restore@9255dc7a253b0ccc959486e2bca901246202afeb #v5.0.1
         with:
           path: |
             utils/tfhe-backward-compat-data/**/*.cbor
@@ -109,7 +109,7 @@ jobs:
       - name: Store data in cache
         if: steps.retrieve-data-cache.outputs.cache-hit != 'true'
         continue-on-error: true
-        uses: actions/cache/save@0057852bfaa89a56745cba8c7296529d2fc39830 #v4.3.0
+        uses: actions/cache/save@9255dc7a253b0ccc959486e2bca901246202afeb #v5.0.1
         with:
           path: |
             utils/tfhe-backward-compat-data/**/*.cbor

.github/workflows/aws_tfhe_fast_tests.yml

@@ -63,7 +63,7 @@ jobs:
       any_file_changed: ${{ env.IS_PULL_REQUEST == 'false' || steps.aggregated-changes.outputs.any_changed }}
     steps:
       - name: Checkout tfhe-rs
-        uses: actions/checkout@1af3b93b6815bc44a9784bd300feb67ff0d1eeb3
+        uses: actions/checkout@8e8c483db84b4bee98b60c0593521ed34d9990e8
         with:
           fetch-depth: 0
           persist-credentials: 'false'
@@ -71,7 +71,7 @@ jobs:
 
       - name: Check for file changes
         id: changed-files
-        uses: tj-actions/changed-files@24d32ffd492484c1d75e0c0b894501ddb9d30d62 # v47.0.0
+        uses: tj-actions/changed-files@e0021407031f5be11a464abee9a0776171c79891 # v47.0.1
         with:
           files_yaml: |
             dependencies:
@@ -171,7 +171,7 @@ jobs:
     runs-on: ${{ needs.setup-instance.outputs.runner-name }}
     steps:
       - name: Checkout tfhe-rs
-        uses: actions/checkout@1af3b93b6815bc44a9784bd300feb67ff0d1eeb3
+        uses: actions/checkout@8e8c483db84b4bee98b60c0593521ed34d9990e8
         with:
           persist-credentials: 'false'
           token: ${{ env.CHECKOUT_TOKEN }}
@@ -219,7 +219,7 @@ jobs:
 
       - name: Node cache restoration
         id: node-cache
-        uses: actions/cache/restore@0057852bfaa89a56745cba8c7296529d2fc39830 #v4.3.0
+        uses: actions/cache/restore@9255dc7a253b0ccc959486e2bca901246202afeb #v5.0.1
         with:
           path: |
             ~/.nvm
@@ -232,7 +232,7 @@ jobs:
           make install_node
 
       - name: Node cache save
-        uses: actions/cache/save@0057852bfaa89a56745cba8c7296529d2fc39830 #v4.3.0
+        uses: actions/cache/save@9255dc7a253b0ccc959486e2bca901246202afeb #v5.0.1
         if: steps.node-cache.outputs.cache-hit != 'true'
         with:
           path: |

.github/workflows/aws_tfhe_integer_tests.yml

@@ -50,7 +50,7 @@ jobs:
         steps.changed-files.outputs.integer_any_changed }}
     steps:
       - name: Checkout tfhe-rs
-        uses: actions/checkout@1af3b93b6815bc44a9784bd300feb67ff0d1eeb3
+        uses: actions/checkout@8e8c483db84b4bee98b60c0593521ed34d9990e8
         with:
           fetch-depth: 0
           persist-credentials: 'false'
@@ -58,7 +58,7 @@ jobs:
 
       - name: Check for file changes
         id: changed-files
-        uses: tj-actions/changed-files@24d32ffd492484c1d75e0c0b894501ddb9d30d62 # v47.0.0
+        uses: tj-actions/changed-files@e0021407031f5be11a464abee9a0776171c79891 # v47.0.1
         with:
           files_yaml: |
             integer:
@@ -112,7 +112,7 @@ jobs:
     timeout-minutes: 480 # 8 hours
     steps:
       - name: Checkout tfhe-rs
-        uses: actions/checkout@1af3b93b6815bc44a9784bd300feb67ff0d1eeb3
+        uses: actions/checkout@8e8c483db84b4bee98b60c0593521ed34d9990e8
         with:
           persist-credentials: "false"
           token: ${{ env.CHECKOUT_TOKEN }}

.github/workflows/aws_tfhe_noise_checks.yml

@@ -60,7 +60,7 @@ jobs:
     timeout-minutes: 1440
     steps:
       - name: Checkout tfhe-rs
-        uses: actions/checkout@1af3b93b6815bc44a9784bd300feb67ff0d1eeb3
+        uses: actions/checkout@8e8c483db84b4bee98b60c0593521ed34d9990e8
         with:
           persist-credentials: 'false'
           token: ${{ env.CHECKOUT_TOKEN }}

.github/workflows/aws_tfhe_signed_integer_tests.yml

@@ -51,7 +51,7 @@ jobs:
         steps.changed-files.outputs.integer_any_changed }}
     steps:
       - name: Checkout tfhe-rs
-        uses: actions/checkout@1af3b93b6815bc44a9784bd300feb67ff0d1eeb3
+        uses: actions/checkout@8e8c483db84b4bee98b60c0593521ed34d9990e8
         with:
           fetch-depth: 0
           persist-credentials: 'false'
@@ -59,7 +59,7 @@ jobs:
 
       - name: Check for file changes
         id: changed-files
-        uses: tj-actions/changed-files@24d32ffd492484c1d75e0c0b894501ddb9d30d62 # v47.0.0
+        uses: tj-actions/changed-files@e0021407031f5be11a464abee9a0776171c79891 # v47.0.1
         with:
           files_yaml: |
             integer:
@@ -112,7 +112,7 @@ jobs:
     runs-on: ${{ needs.setup-instance.outputs.runner-name }}
     steps:
       - name: Checkout tfhe-rs
-        uses: actions/checkout@1af3b93b6815bc44a9784bd300feb67ff0d1eeb3
+        uses: actions/checkout@8e8c483db84b4bee98b60c0593521ed34d9990e8
         with:
           persist-credentials: "false"
           token: ${{ env.CHECKOUT_TOKEN }}

.github/workflows/aws_tfhe_tests.yml

@@ -72,7 +72,7 @@ jobs:
       any_file_changed: ${{ env.IS_PULL_REQUEST == 'false' || steps.aggregated-changes.outputs.any_changed }}
     steps:
       - name: Checkout tfhe-rs
-        uses: actions/checkout@1af3b93b6815bc44a9784bd300feb67ff0d1eeb3
+        uses: actions/checkout@8e8c483db84b4bee98b60c0593521ed34d9990e8
         with:
           fetch-depth: 0
           persist-credentials: 'false'
@@ -80,7 +80,7 @@ jobs:
 
       - name: Check for file changes
         id: changed-files
-        uses: tj-actions/changed-files@24d32ffd492484c1d75e0c0b894501ddb9d30d62 # v47.0.0
+        uses: tj-actions/changed-files@e0021407031f5be11a464abee9a0776171c79891 # v47.0.1
         with:
           files_yaml: |
             dependencies:
@@ -182,7 +182,7 @@ jobs:
     runs-on: ${{ needs.setup-instance.outputs.runner-name }}
     steps:
       - name: Checkout tfhe-rs
-        uses: actions/checkout@1af3b93b6815bc44a9784bd300feb67ff0d1eeb3
+        uses: actions/checkout@8e8c483db84b4bee98b60c0593521ed34d9990e8
         with:
           persist-credentials: 'false'
           token: ${{ env.CHECKOUT_TOKEN }}

.github/workflows/aws_tfhe_wasm_tests.yml

@@ -64,7 +64,7 @@ jobs:
     runs-on: ${{ needs.setup-instance.outputs.runner-name }}
     steps:
       - name: Checkout tfhe-rs
-        uses: actions/checkout@1af3b93b6815bc44a9784bd300feb67ff0d1eeb3
+        uses: actions/checkout@8e8c483db84b4bee98b60c0593521ed34d9990e8
         with:
           persist-credentials: 'false'
           token: ${{ env.CHECKOUT_TOKEN }}
@@ -80,7 +80,7 @@ jobs:
 
       - name: Node cache restoration
         id: node-cache
-        uses: actions/cache/restore@0057852bfaa89a56745cba8c7296529d2fc39830 #v4.3.0
+        uses: actions/cache/restore@9255dc7a253b0ccc959486e2bca901246202afeb #v5.0.1
         with:
           path: |
             ~/.nvm
@@ -93,7 +93,7 @@ jobs:
           make install_node
 
       - name: Node cache save
-        uses: actions/cache/save@0057852bfaa89a56745cba8c7296529d2fc39830 #v4.3.0
+        uses: actions/cache/save@9255dc7a253b0ccc959486e2bca901246202afeb #v5.0.1
         if: steps.node-cache.outputs.cache-hit != 'true'
         with:
           path: |

.github/workflows/benchmark_cpu_common.yml

@@ -149,7 +149,7 @@ jobs:
         params_type: ${{ fromJSON(needs.prepare-matrix.outputs.params_type) }}
     steps:
       - name: Checkout tfhe-rs repo with tags
-        uses: actions/checkout@1af3b93b6815bc44a9784bd300feb67ff0d1eeb3
+        uses: actions/checkout@8e8c483db84b4bee98b60c0593521ed34d9990e8
         with:
           fetch-depth: 0
           persist-credentials: 'false'
@@ -223,13 +223,13 @@ jobs:
           results_type: ${{ inputs.additional_results_type }}
 
       - name: Upload parsed results artifact
-        uses: actions/upload-artifact@330a01c490aca151604b8cf639adc76d48f6c5d4
+        uses: actions/upload-artifact@b7c566a772e6b6bfb58ed0dc250532a479d7789f
         with:
           name: ${{ github.sha }}_${{ matrix.command }}_${{ matrix.op_flavor }}_${{ matrix.bench_type }}_${{ matrix.params_type }}
           path: ${{ env.RESULTS_FILENAME }}
 
       - name: Checkout Slab repo
-        uses: actions/checkout@1af3b93b6815bc44a9784bd300feb67ff0d1eeb3
+        uses: actions/checkout@8e8c483db84b4bee98b60c0593521ed34d9990e8
         with:
           repository: zama-ai/slab
           path: slab

.github/workflows/benchmark_ct_key_sizes.yml

@@ -49,7 +49,7 @@ jobs:
     runs-on: ${{ needs.setup-instance.outputs.runner-name }}
     steps:
       - name: Checkout tfhe-rs repo with tags
-        uses: actions/checkout@1af3b93b6815bc44a9784bd300feb67ff0d1eeb3
+        uses: actions/checkout@8e8c483db84b4bee98b60c0593521ed34d9990e8
         with:
           fetch-depth: 0
           persist-credentials: 'false'
@@ -99,13 +99,13 @@ jobs:
           --append-results
 
       - name: Upload parsed results artifact
-        uses: actions/upload-artifact@330a01c490aca151604b8cf639adc76d48f6c5d4
+        uses: actions/upload-artifact@b7c566a772e6b6bfb58ed0dc250532a479d7789f
         with:
           name: ${{ github.sha }}_ct_key_sizes
           path: ${{ env.RESULTS_FILENAME }}
 
       - name: Checkout Slab repo
-        uses: actions/checkout@1af3b93b6815bc44a9784bd300feb67ff0d1eeb3
+        uses: actions/checkout@8e8c483db84b4bee98b60c0593521ed34d9990e8
         with:
           repository: zama-ai/slab
           path: slab

.github/workflows/benchmark_documentation.yml

@@ -8,8 +8,13 @@ on:
         description: "Run CPU benchmarks"
         type: boolean
         default: true
-      run-gpu-benchmarks:
-        description: "Run GPU benchmarks"
+      # GPU benchmarks are split because of resource scarcity.
+      run-gpu-integer-benchmarks:
+        description: "Run GPU integer benchmarks"
+        type: boolean
+        default: true
+      run-gpu-core-crypto-benchmarks:
+        description: "Run GPU core-crypto benchmarks"
         type: boolean
         default: true
       run-hpu-benchmarks:
@@ -52,7 +57,7 @@ jobs:
   run-benchmarks-gpu-integer:
     name: benchmark_documentation/run-benchmarks-gpu-integer
     uses: ./.github/workflows/benchmark_gpu_common.yml
-    if: inputs.run-gpu-benchmarks
+    if: inputs.run-gpu-integer-benchmarks
     with:
       profile: multi-h100-sxm5
       hardware_name: n3-H100-SXM5x8
@@ -113,7 +118,7 @@ jobs:
   run-benchmarks-gpu-core-crypto:
     name: benchmark_documentation/run-benchmarks-gpu-core-crypto
     uses: ./.github/workflows/benchmark_gpu_common.yml
-    if: inputs.run-gpu-benchmarks
+    if: inputs.run-gpu-core-crypto-benchmarks
     with:
       profile: multi-h100-sxm5
       hardware_name: n3-H100-SXM5x8
@@ -133,7 +138,7 @@ jobs:
   generate-svgs-with-benchmarks-run:
     name: benchmark-documentation/generate-svgs-with-benchmarks-run
     if: ${{ always() &&
-      (inputs.run-cpu-benchmarks || inputs.run-gpu-benchmarks ||inputs.run-hpu-benchmarks) &&
+      (inputs.run-cpu-benchmarks || inputs.run-gpu-integer-benchmarks || inputs.run-gpu-core-crypto-benchmarks ||inputs.run-hpu-benchmarks) &&
       inputs.generate-svgs }}
     needs: [
       run-benchmarks-cpu-integer, run-benchmarks-gpu-integer, run-benchmarks-hpu-integer,
@@ -143,7 +148,7 @@ jobs:
     with:
       time_span_days: 5
       generate-cpu-svgs: ${{ inputs.run-cpu-benchmarks }}
-      generate-gpu-svgs: ${{ inputs.run-gpu-benchmarks }}
+      generate-gpu-svgs: ${{ inputs.run-gpu-integer-benchmarks || inputs.run-gpu-core-crypto-benchmarks }}
       generate-hpu-svgs: ${{ inputs.run-hpu-benchmarks }}
     secrets:
       DATA_EXTRACTOR_DATABASE_USER: ${{ secrets.DATA_EXTRACTOR_DATABASE_USER }}
@@ -152,7 +157,7 @@ jobs:
 
   generate-svgs-without-benchmarks-run:
     name: benchmark-documentation/generate-svgs-without-benchmarks-run
-    if: ${{ !(inputs.run-cpu-benchmarks || inputs.run-gpu-benchmarks || inputs.run-hpu-benchmarks) &&
+    if: ${{ !(inputs.run-cpu-benchmarks || inputs.run-gpu-integer-benchmarks || inputs.run-gpu-core-crypto-benchmarks || inputs.run-hpu-benchmarks) &&
       inputs.generate-svgs }}
     uses: ./.github/workflows/generate_svgs.yml
     with:
@@ -175,12 +180,12 @@ jobs:
       PATH_TO_DOC_ASSETS: tfhe/docs/.gitbook/assets
     steps:
       - name: Checkout tfhe-rs
-        uses: actions/checkout@1af3b93b6815bc44a9784bd300feb67ff0d1eeb3
+        uses: actions/checkout@8e8c483db84b4bee98b60c0593521ed34d9990e8
         with:
           persist-credentials: 'false'
 
       - name: Download SVG tables
-        uses: actions/download-artifact@018cc2cf5baa6db3ef3c5f8a56943fffe632ef53 # v6.0.0
+        uses: actions/download-artifact@37930b1c2abaa49bbe596cd826c3c89aef350131 # v7.0.0
         with:
           path: svg_tables
           merge-multiple: 'true'
@@ -198,7 +203,7 @@ jobs:
           echo "date=$(date '+%g_%m_%d_%Hh%Mm%Ss')" >> "${GITHUB_OUTPUT}"
 
       - name: Create pull-request
-        uses: peter-evans/create-pull-request@84ae59a2cdc2258d6fa0732dd66352dddae2a412 # v7.0.9
+        uses: peter-evans/create-pull-request@98357b18bf14b5342f975ff684046ec3b2a07725 # v8.0.0
         with:
           sign-commits: true # Commit will be signed by github-actions bot
           add-paths: ${{ env.PATH_TO_DOC_ASSETS }}/*.svg

.github/workflows/benchmark_gpu_4090.yml

@@ -40,7 +40,7 @@ jobs:
     timeout-minutes: 1440 # 24 hours
     steps:
       - name: Checkout tfhe-rs
-        uses: actions/checkout@1af3b93b6815bc44a9784bd300feb67ff0d1eeb3
+        uses: actions/checkout@8e8c483db84b4bee98b60c0593521ed34d9990e8
         with:
           fetch-depth: 0
           persist-credentials: 'false'
@@ -63,7 +63,7 @@ jobs:
           toolchain: nightly
 
       - name: Checkout Slab repo
-        uses: actions/checkout@1af3b93b6815bc44a9784bd300feb67ff0d1eeb3
+        uses: actions/checkout@8e8c483db84b4bee98b60c0593521ed34d9990e8
         with:
           repository: zama-ai/slab
           path: slab
@@ -89,7 +89,7 @@ jobs:
           REF_NAME: ${{ github.ref_name }}
 
       - name: Upload parsed results artifact
-        uses: actions/upload-artifact@330a01c490aca151604b8cf639adc76d48f6c5d4
+        uses: actions/upload-artifact@b7c566a772e6b6bfb58ed0dc250532a479d7789f
         with:
           name: ${{ github.sha }}_integer_multi_bit_gpu_default
           path: ${{ env.RESULTS_FILENAME }}
@@ -123,7 +123,7 @@ jobs:
 
     steps:
       - name: Checkout tfhe-rs
-        uses: actions/checkout@1af3b93b6815bc44a9784bd300feb67ff0d1eeb3
+        uses: actions/checkout@8e8c483db84b4bee98b60c0593521ed34d9990e8
         with:
           fetch-depth: 0
           persist-credentials: 'false'
@@ -146,7 +146,7 @@ jobs:
           toolchain: nightly
 
       - name: Checkout Slab repo
-        uses: actions/checkout@1af3b93b6815bc44a9784bd300feb67ff0d1eeb3
+        uses: actions/checkout@8e8c483db84b4bee98b60c0593521ed34d9990e8
         with:
           repository: zama-ai/slab
           path: slab
@@ -173,7 +173,7 @@ jobs:
           REF_NAME: ${{ github.ref_name }}
 
       - name: Upload parsed results artifact
-        uses: actions/upload-artifact@330a01c490aca151604b8cf639adc76d48f6c5d4
+        uses: actions/upload-artifact@b7c566a772e6b6bfb58ed0dc250532a479d7789f
         with:
           name: ${{ github.sha }}_core_crypto
           path: ${{ env.RESULTS_FILENAME }}

.github/workflows/benchmark_gpu_common.yml

@@ -175,7 +175,7 @@ jobs:
             gcc: 11
     steps:
       - name: Checkout tfhe-rs repo
-        uses: actions/checkout@1af3b93b6815bc44a9784bd300feb67ff0d1eeb3
+        uses: actions/checkout@8e8c483db84b4bee98b60c0593521ed34d9990e8
         with:
           fetch-depth: 0
           persist-credentials: 'false'
@@ -209,7 +209,7 @@ jobs:
       CUDA_PATH: /usr/local/cuda-${{ matrix.cuda }}
     steps:
       - name: Checkout tfhe-rs repo with tags
-        uses: actions/checkout@1af3b93b6815bc44a9784bd300feb67ff0d1eeb3
+        uses: actions/checkout@8e8c483db84b4bee98b60c0593521ed34d9990e8
         with:
           fetch-depth: 0
           persist-credentials: 'false'
@@ -281,13 +281,13 @@ jobs:
           BENCH_TYPE: ${{ matrix.bench_type }}
 
       - name: Upload parsed results artifact
-        uses: actions/upload-artifact@330a01c490aca151604b8cf639adc76d48f6c5d4
+        uses: actions/upload-artifact@b7c566a772e6b6bfb58ed0dc250532a479d7789f
         with:
           name: ${{ github.sha }}_${{ matrix.command }}_${{ matrix.op_flavor }}_${{ inputs.profile }}_${{ matrix.bench_type }}_${{ matrix.params_type }}
           path: ${{ env.RESULTS_FILENAME }}
 
       - name: Checkout Slab repo
-        uses: actions/checkout@1af3b93b6815bc44a9784bd300feb67ff0d1eeb3
+        uses: actions/checkout@8e8c483db84b4bee98b60c0593521ed34d9990e8
         with:
           repository: zama-ai/slab
           path: slab

.github/workflows/benchmark_gpu_coprocessor.yml

@@ -130,7 +130,7 @@ jobs:
           git lfs install
 
       - name: Checkout tfhe-rs
-        uses: actions/checkout@1af3b93b6815bc44a9784bd300feb67ff0d1eeb3 # v6.0.0
+        uses: actions/checkout@8e8c483db84b4bee98b60c0593521ed34d9990e8 # v6.0.1
         with:
           path: tfhe-rs
           persist-credentials: false
@@ -141,7 +141,7 @@ jobs:
           ls
 
       - name: Checkout fhevm
-        uses: actions/checkout@1af3b93b6815bc44a9784bd300feb67ff0d1eeb3 # v6.0.0
+        uses: actions/checkout@8e8c483db84b4bee98b60c0593521ed34d9990e8 # v6.0.1
         with:
           repository: zama-ai/fhevm
           persist-credentials: 'false'
@@ -192,10 +192,10 @@ jobs:
           cargo install sqlx-cli
 
       - name: Install foundry
-        uses: foundry-rs/foundry-toolchain@50d5a8956f2e319df19e6b57539d7e2acb9f8c1e
+        uses: foundry-rs/foundry-toolchain@8b0419c685ef46cb79ec93fbdc131174afceb730
 
       - name: Cache cargo
-        uses: actions/cache@0057852bfaa89a56745cba8c7296529d2fc39830 # v4.3.0
+        uses: actions/cache@9255dc7a253b0ccc959486e2bca901246202afeb # v5.0.1
         with:
           path: |
             ~/.cargo/registry
@@ -223,7 +223,7 @@ jobs:
         working-directory: fhevm/coprocessor/fhevm-engine/tfhe-worker
 
       - name: Use Node.js
-        uses: actions/setup-node@2028fbc5c25fe9cf00d9f06a71cc4710d4507903 # v6.0.0
+        uses: actions/setup-node@395ad3262231945c25e8478fd5baf05154b1d79f # v6.1.0
         with:
           node-version: 20.x
 
@@ -262,7 +262,7 @@ jobs:
       - name: Upload profile artifact
         env:
           REPORT_NAME: ${{ steps.nsys_profile_name.outputs.profile }}
-        uses: actions/upload-artifact@330a01c490aca151604b8cf639adc76d48f6c5d4
+        uses: actions/upload-artifact@b7c566a772e6b6bfb58ed0dc250532a479d7789f
         with:
           name: ${{ env.REPORT_NAME }}
           path: fhevm/coprocessor/fhevm-engine/tfhe-worker/${{ env.REPORT_NAME }}
@@ -293,13 +293,13 @@ jobs:
         working-directory: fhevm/
 
       - name: Upload parsed results artifact
-        uses: actions/upload-artifact@330a01c490aca151604b8cf639adc76d48f6c5d4
+        uses: actions/upload-artifact@b7c566a772e6b6bfb58ed0dc250532a479d7789f
         with:
           name: ${COMMIT_SHA}_${BENCHMARKS}_${{ needs.parse-inputs.outputs.profile }}
           path: fhevm/$${{ env.RESULTS_FILENAME }}
 
       - name: Checkout Slab repo
-        uses: actions/checkout@1af3b93b6815bc44a9784bd300feb67ff0d1eeb3
+        uses: actions/checkout@8e8c483db84b4bee98b60c0593521ed34d9990e8
         with:
           repository: zama-ai/slab
           path: slab

.github/workflows/benchmark_hpu_common.yml

@@ -126,7 +126,7 @@ jobs:
           ssh-private-key: ${{ secrets.SSH_PRIVATE_KEY }}
 
       - name: Checkout tfhe-rs repo with tags
-        uses: actions/checkout@1af3b93b6815bc44a9784bd300feb67ff0d1eeb3
+        uses: actions/checkout@8e8c483db84b4bee98b60c0593521ed34d9990e8
         with:
           fetch-depth: 0
           persist-credentials: 'false'
@@ -185,13 +185,13 @@ jobs:
           BENCH_TYPE: ${{ matrix.bench_type }}
 
       - name: Upload parsed results artifact
-        uses: actions/upload-artifact@330a01c490aca151604b8cf639adc76d48f6c5d4
+        uses: actions/upload-artifact@b7c566a772e6b6bfb58ed0dc250532a479d7789f
         with:
-          name: ${{ github.sha }}_${{ matrix.bench_type }}_integer_benchmarks
+          name: ${{ github.sha }}_${{ matrix.bench_type }}_${{ matrix.command }}_benchmarks
           path: ${{ env.RESULTS_FILENAME }}
 
       - name: Checkout Slab repo
-        uses: actions/checkout@1af3b93b6815bc44a9784bd300feb67ff0d1eeb3
+        uses: actions/checkout@8e8c483db84b4bee98b60c0593521ed34d9990e8
         with:
           repository: zama-ai/slab
           path: slab

.github/workflows/benchmark_perf_regression.yml

@@ -50,7 +50,7 @@ jobs:
       pull-requests: write # Needed to write a comment in a pull-request
     steps:
       - name: Checkout tfhe-rs repo
-        uses: actions/checkout@1af3b93b6815bc44a9784bd300feb67ff0d1eeb3
+        uses: actions/checkout@8e8c483db84b4bee98b60c0593521ed34d9990e8
         with:
           persist-credentials: 'false'
           token: ${{ secrets.REPO_CHECKOUT_TOKEN }}
@@ -164,7 +164,7 @@ jobs:
             gcc: 11
     steps:
       - name: Checkout tfhe-rs repo
-        uses: actions/checkout@1af3b93b6815bc44a9784bd300feb67ff0d1eeb3
+        uses: actions/checkout@8e8c483db84b4bee98b60c0593521ed34d9990e8
         with:
           persist-credentials: 'false'
           token: ${{ secrets.REPO_CHECKOUT_TOKEN }}
@@ -191,7 +191,7 @@ jobs:
         command: ${{ fromJson(needs.prepare-benchmarks.outputs.commands) }}
     steps:
       - name: Checkout tfhe-rs repo
-        uses: actions/checkout@1af3b93b6815bc44a9784bd300feb67ff0d1eeb3
+        uses: actions/checkout@8e8c483db84b4bee98b60c0593521ed34d9990e8
         with:
           fetch-depth: 0  # Needed to get commit hash
           persist-credentials: 'false'
@@ -245,7 +245,7 @@ jobs:
           toolchain: nightly
 
       - name: Checkout Slab repo
-        uses: actions/checkout@1af3b93b6815bc44a9784bd300feb67ff0d1eeb3
+        uses: actions/checkout@8e8c483db84b4bee98b60c0593521ed34d9990e8
         with:
           repository: zama-ai/slab
           path: slab
@@ -280,7 +280,7 @@ jobs:
           BENCH_TYPE: ${{ env.__TFHE_RS_BENCH_TYPE }}
 
       - name: Upload parsed results artifact
-        uses: actions/upload-artifact@330a01c490aca151604b8cf639adc76d48f6c5d4
+        uses: actions/upload-artifact@b7c566a772e6b6bfb58ed0dc250532a479d7789f
         with:
           name: ${{ github.sha }}_regression_${{ env.RESULTS_FILE_SHA }} # RESULT_FILE_SHA is needed to avoid collision between matrix.command runs
           path: ${{ env.RESULTS_FILENAME }}
@@ -305,7 +305,7 @@ jobs:
       REF_NAME: ${{ github.head_ref || github.ref_name }}
     steps:
       - name: Checkout tfhe-rs repo
-        uses: actions/checkout@1af3b93b6815bc44a9784bd300feb67ff0d1eeb3
+        uses: actions/checkout@8e8c483db84b4bee98b60c0593521ed34d9990e8
         with:
           persist-credentials: 'false'
           token: ${{ secrets.REPO_CHECKOUT_TOKEN }}

.github/workflows/benchmark_tfhe_fft.yml

@@ -55,7 +55,7 @@ jobs:
     runs-on: ${{ needs.setup-instance.outputs.runner-name }}
     steps:
       - name: Checkout tfhe-rs repo with tags
-        uses: actions/checkout@1af3b93b6815bc44a9784bd300feb67ff0d1eeb3
+        uses: actions/checkout@8e8c483db84b4bee98b60c0593521ed34d9990e8
         with:
           fetch-depth: 0
           persist-credentials: 'false'
@@ -96,13 +96,13 @@ jobs:
           REF_NAME: ${{ github.ref_name }}
 
       - name: Upload parsed results artifact
-        uses: actions/upload-artifact@330a01c490aca151604b8cf639adc76d48f6c5d4
+        uses: actions/upload-artifact@b7c566a772e6b6bfb58ed0dc250532a479d7789f
         with:
           name: ${{ github.sha }}_fft
           path: ${{ env.RESULTS_FILENAME }}
 
       - name: Checkout Slab repo
-        uses: actions/checkout@1af3b93b6815bc44a9784bd300feb67ff0d1eeb3
+        uses: actions/checkout@8e8c483db84b4bee98b60c0593521ed34d9990e8
         with:
           repository: zama-ai/slab
           path: slab

.github/workflows/benchmark_tfhe_ntt.yml

@@ -55,7 +55,7 @@ jobs:
     runs-on: ${{ needs.setup-instance.outputs.runner-name }}
     steps:
       - name: Checkout tfhe-rs repo with tags
-        uses: actions/checkout@1af3b93b6815bc44a9784bd300feb67ff0d1eeb3
+        uses: actions/checkout@8e8c483db84b4bee98b60c0593521ed34d9990e8
         with:
           fetch-depth: 0
           persist-credentials: 'false'
@@ -96,13 +96,13 @@ jobs:
           REF_NAME: ${{ github.ref_name }}
 
       - name: Upload parsed results artifact
-        uses: actions/upload-artifact@330a01c490aca151604b8cf639adc76d48f6c5d4
+        uses: actions/upload-artifact@b7c566a772e6b6bfb58ed0dc250532a479d7789f
         with:
           name: ${{ github.sha }}_ntt
           path: ${{ env.RESULTS_FILENAME }}
 
       - name: Checkout Slab repo
-        uses: actions/checkout@1af3b93b6815bc44a9784bd300feb67ff0d1eeb3
+        uses: actions/checkout@8e8c483db84b4bee98b60c0593521ed34d9990e8
         with:
           repository: zama-ai/slab
           path: slab

.github/workflows/benchmark_wasm_client.yml

@@ -39,7 +39,7 @@ jobs:
       wasm_bench: ${{ steps.changed-files.outputs.wasm_bench_any_changed }}
     steps:
       - name: Checkout tfhe-rs
-        uses: actions/checkout@1af3b93b6815bc44a9784bd300feb67ff0d1eeb3
+        uses: actions/checkout@8e8c483db84b4bee98b60c0593521ed34d9990e8
         with:
           fetch-depth: 0
           persist-credentials: 'false'
@@ -47,7 +47,7 @@ jobs:
 
       - name: Check for file changes
         id: changed-files
-        uses: tj-actions/changed-files@24d32ffd492484c1d75e0c0b894501ddb9d30d62 # v47.0.0
+        uses: tj-actions/changed-files@e0021407031f5be11a464abee9a0776171c79891 # v47.0.1
         with:
           files_yaml: |
             wasm_bench:
@@ -91,7 +91,7 @@ jobs:
         browser: [ chrome, firefox ]
     steps:
       - name: Checkout tfhe-rs repo with tags
-        uses: actions/checkout@1af3b93b6815bc44a9784bd300feb67ff0d1eeb3
+        uses: actions/checkout@8e8c483db84b4bee98b60c0593521ed34d9990e8
         with:
           fetch-depth: 0
           persist-credentials: 'false'
@@ -119,7 +119,7 @@ jobs:
 
       - name: Node cache restoration
         id: node-cache
-        uses: actions/cache/restore@0057852bfaa89a56745cba8c7296529d2fc39830 #v4.3.0
+        uses: actions/cache/restore@9255dc7a253b0ccc959486e2bca901246202afeb #v5.0.1
         with:
           path: |
             ~/.nvm
@@ -132,7 +132,7 @@ jobs:
           make install_node
 
       - name: Node cache save
-        uses: actions/cache/save@0057852bfaa89a56745cba8c7296529d2fc39830 #v4.3.0
+        uses: actions/cache/save@9255dc7a253b0ccc959486e2bca901246202afeb #v5.0.1
         if: steps.node-cache.outputs.cache-hit != 'true'
         with:
           path: |
@@ -153,6 +153,12 @@ jobs:
         env:
           BROWSER: ${{ matrix.browser }}
 
+      - name: Run benchmarks (unsafe coop)
+        run: |
+          make bench_web_js_api_unsafe_coop_"${BROWSER}"_ci
+        env:
+          BROWSER: ${{ matrix.browser }}
+
       - name: Parse results
         run: |
           make parse_wasm_benchmarks
@@ -169,13 +175,13 @@ jobs:
           REF_NAME: ${{ github.ref_name }}
 
       - name: Upload parsed results artifact
-        uses: actions/upload-artifact@330a01c490aca151604b8cf639adc76d48f6c5d4
+        uses: actions/upload-artifact@b7c566a772e6b6bfb58ed0dc250532a479d7789f
         with:
           name: ${{ github.sha }}_wasm_${{ matrix.browser }}
           path: ${{ env.RESULTS_FILENAME }}
 
       - name: Checkout Slab repo
-        uses: actions/checkout@1af3b93b6815bc44a9784bd300feb67ff0d1eeb3
+        uses: actions/checkout@8e8c483db84b4bee98b60c0593521ed34d9990e8
         with:
           repository: zama-ai/slab
           path: slab

.github/workflows/cargo_audit.yml

@@ -26,7 +26,7 @@ jobs:
     name: cargo_audit/audit
     runs-on: ubuntu-latest
     steps:
-      - uses: actions/checkout@1af3b93b6815bc44a9784bd300feb67ff0d1eeb3
+      - uses: actions/checkout@8e8c483db84b4bee98b60c0593521ed34d9990e8
         with:
           persist-credentials: 'false'
           token: ${{ env.CHECKOUT_TOKEN }}

.github/workflows/cargo_build.yml

@@ -24,7 +24,7 @@ jobs:
     outputs:
       matrix_command: ${{ steps.set-pcc-commands-matrix.outputs.commands }}
     steps:
-      - uses: actions/checkout@1af3b93b6815bc44a9784bd300feb67ff0d1eeb3
+      - uses: actions/checkout@8e8c483db84b4bee98b60c0593521ed34d9990e8
         with:
           persist-credentials: "false"
           token: ${{ env.CHECKOUT_TOKEN }}

.github/workflows/cargo_build_common.yml

@@ -140,7 +140,7 @@ jobs:
       result: ${{ steps.set_builds_result.outputs.result }}
     steps:
       - name: Checkout tfhe-rs repo
-        uses: actions/checkout@1af3b93b6815bc44a9784bd300feb67ff0d1eeb3
+        uses: actions/checkout@8e8c483db84b4bee98b60c0593521ed34d9990e8
         with:
           persist-credentials: 'false'
           token: ${{ env.CHECKOUT_TOKEN }}

.github/workflows/cargo_build_tfhe_fft.yml

@@ -26,7 +26,7 @@ jobs:
       fail-fast: false
 
     steps:
-      - uses: actions/checkout@1af3b93b6815bc44a9784bd300feb67ff0d1eeb3
+      - uses: actions/checkout@8e8c483db84b4bee98b60c0593521ed34d9990e8
         with:
           persist-credentials: 'false'
           token: ${{ env.CHECKOUT_TOKEN }}

.github/workflows/cargo_build_tfhe_ntt.yml

@@ -24,7 +24,7 @@ jobs:
         os: [ubuntu-latest, macos-latest, windows-latest]
       fail-fast: false
     steps:
-      - uses: actions/checkout@1af3b93b6815bc44a9784bd300feb67ff0d1eeb3
+      - uses: actions/checkout@8e8c483db84b4bee98b60c0593521ed34d9990e8
         with:
           persist-credentials: 'false'
           token: ${{ env.CHECKOUT_TOKEN }}

.github/workflows/cargo_test_fft.yml

@@ -29,7 +29,7 @@ jobs:
       fft_test: ${{ env.IS_PULL_REQUEST == 'false' || steps.changed-files.outputs.fft_any_changed }}
     steps:
       - name: Checkout tfhe-rs
-        uses: actions/checkout@1af3b93b6815bc44a9784bd300feb67ff0d1eeb3
+        uses: actions/checkout@8e8c483db84b4bee98b60c0593521ed34d9990e8
         with:
           fetch-depth: 0
           persist-credentials: 'false'
@@ -37,7 +37,7 @@ jobs:
 
       - name: Check for file changes
         id: changed-files
-        uses: tj-actions/changed-files@24d32ffd492484c1d75e0c0b894501ddb9d30d62 # v47.0.0
+        uses: tj-actions/changed-files@e0021407031f5be11a464abee9a0776171c79891 # v47.0.1
         with:
           files_yaml: |
             fft:
@@ -56,7 +56,7 @@ jobs:
         runner_type: [ ubuntu-latest, macos-latest, windows-latest ]
       fail-fast: false
     steps:
-      - uses: actions/checkout@1af3b93b6815bc44a9784bd300feb67ff0d1eeb3
+      - uses: actions/checkout@8e8c483db84b4bee98b60c0593521ed34d9990e8
         with:
           persist-credentials: 'false'
           token: ${{ env.CHECKOUT_TOKEN }}
@@ -92,7 +92,7 @@ jobs:
     if: needs.should-run.outputs.fft_test == 'true'
     runs-on: ubuntu-latest
     steps:
-      - uses: actions/checkout@1af3b93b6815bc44a9784bd300feb67ff0d1eeb3
+      - uses: actions/checkout@8e8c483db84b4bee98b60c0593521ed34d9990e8
         with:
           persist-credentials: 'false'
           token: ${{ env.CHECKOUT_TOKEN }}

.github/workflows/cargo_test_ntt.yml

@@ -31,7 +31,7 @@ jobs:
       ntt_test: ${{ env.IS_PULL_REQUEST == 'false' || steps.changed-files.outputs.ntt_any_changed }}
     steps:
       - name: Checkout tfhe-rs
-        uses: actions/checkout@1af3b93b6815bc44a9784bd300feb67ff0d1eeb3
+        uses: actions/checkout@8e8c483db84b4bee98b60c0593521ed34d9990e8
         with:
           fetch-depth: 0
           persist-credentials: "false"
@@ -39,7 +39,7 @@ jobs:
 
       - name: Check for file changes
         id: changed-files
-        uses: tj-actions/changed-files@24d32ffd492484c1d75e0c0b894501ddb9d30d62 # v47.0.0
+        uses: tj-actions/changed-files@e0021407031f5be11a464abee9a0776171c79891 # v47.0.1
         with:
           files_yaml: |
             ntt:
@@ -87,7 +87,7 @@ jobs:
         os: ${{fromJson(needs.setup-instance.outputs.matrix_os)}}
       fail-fast: false
     steps:
-      - uses: actions/checkout@1af3b93b6815bc44a9784bd300feb67ff0d1eeb3
+      - uses: actions/checkout@8e8c483db84b4bee98b60c0593521ed34d9990e8
         with:
           persist-credentials: "false"
           token: ${{ env.CHECKOUT_TOKEN }}

.github/workflows/ci_lint.yml

@@ -20,7 +20,7 @@ jobs:
     runs-on: ubuntu-latest
     steps:
       - name: Checkout tfhe-rs
-        uses: actions/checkout@1af3b93b6815bc44a9784bd300feb67ff0d1eeb3
+        uses: actions/checkout@8e8c483db84b4bee98b60c0593521ed34d9990e8
         with:
           persist-credentials: 'false'
           token: ${{ env.CHECKOUT_TOKEN }}
@@ -50,7 +50,7 @@ jobs:
           version: ${{ steps.get_zizmor.outputs.version }}
 
       - name: Ensure SHA pinned actions
-        uses: zgosalvez/github-actions-ensure-sha-pinned-actions@9e9574ef04ea69da568d6249bd69539ccc704e74 # v4.0.0
+        uses: zgosalvez/github-actions-ensure-sha-pinned-actions@6124774845927d14c601359ab8138699fa5b70c3 # v4.0.1
         with:
           allowlist: |
             slsa-framework/slsa-github-generator

.github/workflows/code_coverage.yml

@@ -50,7 +50,7 @@ jobs:
     timeout-minutes: 5760 # 4 days
     steps:
       - name: Checkout tfhe-rs
-        uses: actions/checkout@1af3b93b6815bc44a9784bd300feb67ff0d1eeb3
+        uses: actions/checkout@8e8c483db84b4bee98b60c0593521ed34d9990e8
         with:
           persist-credentials: 'false'
           token: ${{ env.CHECKOUT_TOKEN }}
@@ -62,7 +62,7 @@ jobs:
 
       - name: Check for file changes
         id: changed-files
-        uses: tj-actions/changed-files@24d32ffd492484c1d75e0c0b894501ddb9d30d62 # v47.0.0
+        uses: tj-actions/changed-files@e0021407031f5be11a464abee9a0776171c79891 # v47.0.1
         with:
           files_yaml: |
             tfhe:
@@ -92,7 +92,7 @@ jobs:
           make test_shortint_cov
 
       - name: Upload tfhe coverage to Codecov
-        uses: codecov/codecov-action@5a1091511ad55cbe89839c7260b706298ca349f7
+        uses: codecov/codecov-action@671740ac38dd9b0130fbe1cec585b89eea48d3de
         if: steps.changed-files.outputs.tfhe_any_changed == 'true'
         with:
           token: ${{ secrets.CODECOV_TOKEN }}
@@ -106,7 +106,7 @@ jobs:
           make test_integer_cov
 
       - name: Upload tfhe coverage to Codecov
-        uses: codecov/codecov-action@5a1091511ad55cbe89839c7260b706298ca349f7
+        uses: codecov/codecov-action@671740ac38dd9b0130fbe1cec585b89eea48d3de
         if: steps.changed-files.outputs.tfhe_any_changed == 'true'
         with:
           token: ${{ secrets.CODECOV_TOKEN }}

.github/workflows/csprng_randomness_tests.yml

@@ -62,7 +62,7 @@ jobs:
     runs-on: ${{ needs.setup-instance.outputs.runner-name }}
     steps:
       - name: Checkout tfhe-rs
-        uses: actions/checkout@1af3b93b6815bc44a9784bd300feb67ff0d1eeb3
+        uses: actions/checkout@8e8c483db84b4bee98b60c0593521ed34d9990e8
         with:
           persist-credentials: 'false'
           token: ${{ env.CHECKOUT_TOKEN }}

.github/workflows/generate_svg_common.yml

@@ -43,7 +43,7 @@ jobs:
     runs-on: ubuntu-latest
     steps:
       - name: Checkout tfhe-rs
-        uses: actions/checkout@1af3b93b6815bc44a9784bd300feb67ff0d1eeb3
+        uses: actions/checkout@8e8c483db84b4bee98b60c0593521ed34d9990e8
         with:
           persist-credentials: 'false'
 
@@ -75,13 +75,22 @@ jobs:
           DATA_EXTRACTOR_DATABASE_HOST: ${{ secrets.DATA_EXTRACTOR_DATABASE_HOST }}
           DATA_EXTRACTOR_DATABASE_PASSWORD: ${{ secrets.DATA_EXTRACTOR_DATABASE_PASSWORD }}
 
+      - name: Upload tables
+        if: inputs.backend_comparison == false
+        uses: actions/upload-artifact@b7c566a772e6b6bfb58ed0dc250532a479d7789f
+        with:
+          name: ${{ github.sha }}_${{ inputs.backend }}_${{ inputs.layer }}_${{ inputs.pbs_kind }}_${{ inputs.bench_type }}_tables
+          # This will upload all the file generated
+          path: ${{ inputs.output_filename }}*.svg
+          retention-days: 60
+
       - name: Produce backends comparison table from database
         if: inputs.backend_comparison == true
         run: |
           python3 -m pip install -r ci/data_extractor/requirements.txt
           python3 ci/data_extractor/src/data_extractor.py "${OUTPUT_FILENAME}" \
           --generate-svg \
-          --backend-comparison\
+          --backends-comparison \
           --time-span-days "${TIME_SPAN}"
         env:
           OUTPUT_FILENAME: ${{ inputs.output_filename }}
@@ -90,10 +99,11 @@ jobs:
           DATA_EXTRACTOR_DATABASE_HOST: ${{ secrets.DATA_EXTRACTOR_DATABASE_HOST }}
           DATA_EXTRACTOR_DATABASE_PASSWORD: ${{ secrets.DATA_EXTRACTOR_DATABASE_PASSWORD }}
 
-      - name: Upload tables
-        uses: actions/upload-artifact@330a01c490aca151604b8cf639adc76d48f6c5d4
+      - name: Upload comparison tables
+        if: inputs.backend_comparison == true
+        uses: actions/upload-artifact@b7c566a772e6b6bfb58ed0dc250532a479d7789f
         with:
-          name: ${{ github.sha }}_${{ inputs.backend }}_${{ inputs.layer }}_${{ inputs.pbs_kind }}_${{ inputs.bench_type }}_tables
+          name: ${{ github.sha }}_backends_comparison_tables
           # This will upload all the file generated
           path: ${{ inputs.output_filename }}*.svg
           retention-days: 60

.github/workflows/gpu_4090_tests.yml

@@ -41,7 +41,7 @@ jobs:
 
     steps:
       - name: Checkout tfhe-rs
-        uses: actions/checkout@1af3b93b6815bc44a9784bd300feb67ff0d1eeb3
+        uses: actions/checkout@8e8c483db84b4bee98b60c0593521ed34d9990e8
         with:
           persist-credentials: 'false'
           token: ${{ env.CHECKOUT_TOKEN }}

.github/workflows/gpu_code_validation_tests.yml

@@ -23,8 +23,8 @@ on:
   # Allows you to run this workflow manually from the Actions tab as an alternative.
   workflow_dispatch:
   schedule:
-    # every 3 months
-    - cron: "0 0 1 */3 *"
+    # every month
+    - cron: "0 0 1 * *"
 
 permissions:
   contents: read
@@ -50,7 +50,7 @@ jobs:
           slab-url: ${{ secrets.SLAB_BASE_URL }}
           job-secret: ${{ secrets.JOB_SECRET }}
           backend: hyperstack
-          profile: gpu-test
+          profile: single-h100
 
       # This instance will be spawned especially for pull-request from forked repository
       - name: Start GitHub instance
@@ -79,7 +79,7 @@ jobs:
             gcc: 11 
     steps:
       - name: Checkout tfhe-rs
-        uses: actions/checkout@1af3b93b6815bc44a9784bd300feb67ff0d1eeb3
+        uses: actions/checkout@8e8c483db84b4bee98b60c0593521ed34d9990e8
         with:
           persist-credentials: 'false'
           token: ${{ env.CHECKOUT_TOKEN }}

.github/workflows/gpu_fast_h100_tests.yml

@@ -40,7 +40,7 @@ jobs:
       gpu_test: ${{ env.IS_PULL_REQUEST == 'false' || steps.changed-files.outputs.gpu_any_changed }}
     steps:
       - name: Checkout tfhe-rs
-        uses: actions/checkout@1af3b93b6815bc44a9784bd300feb67ff0d1eeb3
+        uses: actions/checkout@8e8c483db84b4bee98b60c0593521ed34d9990e8
         with:
           fetch-depth: 0
           persist-credentials: 'false'
@@ -48,7 +48,7 @@ jobs:
 
       - name: Check for file changes
         id: changed-files
-        uses: tj-actions/changed-files@24d32ffd492484c1d75e0c0b894501ddb9d30d62 # v47.0.0
+        uses: tj-actions/changed-files@e0021407031f5be11a464abee9a0776171c79891 # v47.0.1
         with:
           files_yaml: |
             gpu:
@@ -129,7 +129,7 @@ jobs:
             gcc: 11 
     steps:
       - name: Checkout tfhe-rs
-        uses: actions/checkout@1af3b93b6815bc44a9784bd300feb67ff0d1eeb3
+        uses: actions/checkout@8e8c483db84b4bee98b60c0593521ed34d9990e8
         with:
           persist-credentials: 'false'
           token: ${{ env.CHECKOUT_TOKEN }}

.github/workflows/gpu_fast_tests.yml

@@ -39,7 +39,7 @@ jobs:
       gpu_test: ${{ env.IS_PULL_REQUEST == 'false' || steps.changed-files.outputs.gpu_any_changed }}
     steps:
       - name: Checkout tfhe-rs
-        uses: actions/checkout@1af3b93b6815bc44a9784bd300feb67ff0d1eeb3
+        uses: actions/checkout@8e8c483db84b4bee98b60c0593521ed34d9990e8
         with:
           fetch-depth: 0
           persist-credentials: 'false'
@@ -47,7 +47,7 @@ jobs:
 
       - name: Check for file changes
         id: changed-files
-        uses: tj-actions/changed-files@24d32ffd492484c1d75e0c0b894501ddb9d30d62 # v47.0.0
+        uses: tj-actions/changed-files@e0021407031f5be11a464abee9a0776171c79891 # v47.0.1
         with:
           files_yaml: |
             gpu:
@@ -114,7 +114,7 @@ jobs:
             gcc: 11 
     steps:
       - name: Checkout tfhe-rs
-        uses: actions/checkout@1af3b93b6815bc44a9784bd300feb67ff0d1eeb3
+        uses: actions/checkout@8e8c483db84b4bee98b60c0593521ed34d9990e8
         with:
           persist-credentials: 'false'
           token: ${{ env.CHECKOUT_TOKEN }}

.github/workflows/gpu_full_h100_tests.yml

@@ -68,7 +68,7 @@ jobs:
             gcc: 11 
     steps:
       - name: Checkout tfhe-rs
-        uses: actions/checkout@1af3b93b6815bc44a9784bd300feb67ff0d1eeb3
+        uses: actions/checkout@8e8c483db84b4bee98b60c0593521ed34d9990e8
         with:
           persist-credentials: 'false'
           token: ${{ secrets.REPO_CHECKOUT_TOKEN }}

.github/workflows/gpu_full_multi_gpu_tests.yml

@@ -40,7 +40,7 @@ jobs:
       gpu_test: ${{ env.IS_PULL_REQUEST == 'false' || steps.changed-files.outputs.gpu_any_changed }}
     steps:
       - name: Checkout tfhe-rs
-        uses: actions/checkout@1af3b93b6815bc44a9784bd300feb67ff0d1eeb3
+        uses: actions/checkout@8e8c483db84b4bee98b60c0593521ed34d9990e8
         with:
           fetch-depth: 0
           persist-credentials: 'false'
@@ -48,7 +48,7 @@ jobs:
 
       - name: Check for file changes
         id: changed-files
-        uses: tj-actions/changed-files@24d32ffd492484c1d75e0c0b894501ddb9d30d62 # v47.0.0
+        uses: tj-actions/changed-files@e0021407031f5be11a464abee9a0776171c79891 # v47.0.1
         with:
           files_yaml: |
             gpu:
@@ -116,7 +116,7 @@ jobs:
             gcc: 11 
     steps:
       - name: Checkout tfhe-rs
-        uses: actions/checkout@1af3b93b6815bc44a9784bd300feb67ff0d1eeb3
+        uses: actions/checkout@8e8c483db84b4bee98b60c0593521ed34d9990e8
         with:
           persist-credentials: 'false'
           token: ${{ env.CHECKOUT_TOKEN }}

.github/workflows/gpu_integer_long_run_tests.yml

@@ -65,7 +65,7 @@ jobs:
     timeout-minutes: 4320 # 72 hours
     steps:
       - name: Checkout tfhe-rs
-        uses: actions/checkout@1af3b93b6815bc44a9784bd300feb67ff0d1eeb3
+        uses: actions/checkout@8e8c483db84b4bee98b60c0593521ed34d9990e8
         with:
           persist-credentials: 'false'
           token: ${{ env.CHECKOUT_TOKEN }}

.github/workflows/gpu_memory_sanitizer.yml

@@ -78,7 +78,7 @@ jobs:
             gcc: 11 
     steps:
       - name: Checkout tfhe-rs
-        uses: actions/checkout@1af3b93b6815bc44a9784bd300feb67ff0d1eeb3
+        uses: actions/checkout@8e8c483db84b4bee98b60c0593521ed34d9990e8
         with:
           persist-credentials: 'false'
           token: ${{ env.CHECKOUT_TOKEN }}

.github/workflows/gpu_memory_sanitizer_h100.yml

@@ -78,7 +78,7 @@ jobs:
             gcc: 11 
     steps:
       - name: Checkout tfhe-rs
-        uses: actions/checkout@1af3b93b6815bc44a9784bd300feb67ff0d1eeb3
+        uses: actions/checkout@8e8c483db84b4bee98b60c0593521ed34d9990e8
         with:
           persist-credentials: 'false'
           token: ${{ env.CHECKOUT_TOKEN }}

.github/workflows/gpu_pcc.yml

@@ -74,7 +74,7 @@ jobs:
 
     steps:
       - name: Checkout tfhe-rs
-        uses: actions/checkout@1af3b93b6815bc44a9784bd300feb67ff0d1eeb3
+        uses: actions/checkout@8e8c483db84b4bee98b60c0593521ed34d9990e8
         with:
           persist-credentials: 'false'
           token: ${{ env.CHECKOUT_TOKEN }}

.github/workflows/gpu_signed_integer_classic_tests.yml

@@ -40,7 +40,7 @@ jobs:
       gpu_test: ${{ env.IS_PULL_REQUEST == 'false' || steps.changed-files.outputs.gpu_any_changed }}
     steps:
       - name: Checkout tfhe-rs
-        uses: actions/checkout@1af3b93b6815bc44a9784bd300feb67ff0d1eeb3
+        uses: actions/checkout@8e8c483db84b4bee98b60c0593521ed34d9990e8
         with:
           fetch-depth: 0
           persist-credentials: 'false'
@@ -48,7 +48,7 @@ jobs:
 
       - name: Check for file changes
         id: changed-files
-        uses: tj-actions/changed-files@24d32ffd492484c1d75e0c0b894501ddb9d30d62 # v47.0.0
+        uses: tj-actions/changed-files@e0021407031f5be11a464abee9a0776171c79891 # v47.0.1
         with:
           files_yaml: |
             gpu:
@@ -116,7 +116,7 @@ jobs:
             gcc: 11 
     steps:
       - name: Checkout tfhe-rs
-        uses: actions/checkout@1af3b93b6815bc44a9784bd300feb67ff0d1eeb3
+        uses: actions/checkout@8e8c483db84b4bee98b60c0593521ed34d9990e8
         with:
           persist-credentials: 'false'
           token: ${{ env.CHECKOUT_TOKEN }}

.github/workflows/gpu_signed_integer_h100_tests.yml

@@ -40,7 +40,7 @@ jobs:
       gpu_test: ${{ env.IS_PULL_REQUEST == 'false' || steps.changed-files.outputs.gpu_any_changed }}
     steps:
       - name: Checkout tfhe-rs
-        uses: actions/checkout@1af3b93b6815bc44a9784bd300feb67ff0d1eeb3
+        uses: actions/checkout@8e8c483db84b4bee98b60c0593521ed34d9990e8
         with:
           fetch-depth: 0
           persist-credentials: 'false'
@@ -48,7 +48,7 @@ jobs:
 
       - name: Check for file changes
         id: changed-files
-        uses: tj-actions/changed-files@24d32ffd492484c1d75e0c0b894501ddb9d30d62 # v47.0.0
+        uses: tj-actions/changed-files@e0021407031f5be11a464abee9a0776171c79891 # v47.0.1
         with:
           files_yaml: |
             gpu:
@@ -129,7 +129,7 @@ jobs:
             gcc: 11 
     steps:
       - name: Checkout tfhe-rs
-        uses: actions/checkout@1af3b93b6815bc44a9784bd300feb67ff0d1eeb3
+        uses: actions/checkout@8e8c483db84b4bee98b60c0593521ed34d9990e8
         with:
           persist-credentials: 'false'
           token: ${{ env.CHECKOUT_TOKEN }}

.github/workflows/gpu_signed_integer_tests.yml

@@ -41,7 +41,7 @@ jobs:
       gpu_test: ${{ env.IS_PULL_REQUEST == 'false' || steps.changed-files.outputs.gpu_any_changed }}
     steps:
       - name: Checkout tfhe-rs
-        uses: actions/checkout@1af3b93b6815bc44a9784bd300feb67ff0d1eeb3
+        uses: actions/checkout@8e8c483db84b4bee98b60c0593521ed34d9990e8
         with:
           fetch-depth: 0
           persist-credentials: 'false'
@@ -49,7 +49,7 @@ jobs:
 
       - name: Check for file changes
         id: changed-files
-        uses: tj-actions/changed-files@24d32ffd492484c1d75e0c0b894501ddb9d30d62 # v47.0.0
+        uses: tj-actions/changed-files@e0021407031f5be11a464abee9a0776171c79891 # v47.0.1
         with:
           files_yaml: |
             gpu:
@@ -117,7 +117,7 @@ jobs:
             gcc: 11
     steps:
       - name: Checkout tfhe-rs
-        uses: actions/checkout@1af3b93b6815bc44a9784bd300feb67ff0d1eeb3
+        uses: actions/checkout@8e8c483db84b4bee98b60c0593521ed34d9990e8
         with:
           persist-credentials: 'false'
           token: ${{ env.CHECKOUT_TOKEN }}

.github/workflows/gpu_unsigned_integer_classic_tests.yml

@@ -40,7 +40,7 @@ jobs:
       gpu_test: ${{ env.IS_PULL_REQUEST == 'false' || steps.changed-files.outputs.gpu_any_changed }}
     steps:
       - name: Checkout tfhe-rs
-        uses: actions/checkout@1af3b93b6815bc44a9784bd300feb67ff0d1eeb3
+        uses: actions/checkout@8e8c483db84b4bee98b60c0593521ed34d9990e8
         with:
           fetch-depth: 0
           persist-credentials: 'false'
@@ -48,7 +48,7 @@ jobs:
 
       - name: Check for file changes
         id: changed-files
-        uses: tj-actions/changed-files@24d32ffd492484c1d75e0c0b894501ddb9d30d62 # v47.0.0
+        uses: tj-actions/changed-files@e0021407031f5be11a464abee9a0776171c79891 # v47.0.1
         with:
           files_yaml: |
             gpu:
@@ -116,7 +116,7 @@ jobs:
             gcc: 11 
     steps:
       - name: Checkout tfhe-rs
-        uses: actions/checkout@1af3b93b6815bc44a9784bd300feb67ff0d1eeb3
+        uses: actions/checkout@8e8c483db84b4bee98b60c0593521ed34d9990e8
         with:
           persist-credentials: 'false'
           token: ${{ env.CHECKOUT_TOKEN }}

.github/workflows/gpu_unsigned_integer_h100_tests.yml

@@ -40,7 +40,7 @@ jobs:
       gpu_test: ${{ env.IS_PULL_REQUEST == 'false' || steps.changed-files.outputs.gpu_any_changed }}
     steps:
       - name: Checkout tfhe-rs
-        uses: actions/checkout@1af3b93b6815bc44a9784bd300feb67ff0d1eeb3
+        uses: actions/checkout@8e8c483db84b4bee98b60c0593521ed34d9990e8
         with:
           fetch-depth: 0
           persist-credentials: 'false'
@@ -48,7 +48,7 @@ jobs:
 
       - name: Check for file changes
         id: changed-files
-        uses: tj-actions/changed-files@24d32ffd492484c1d75e0c0b894501ddb9d30d62 # v47.0.0
+        uses: tj-actions/changed-files@e0021407031f5be11a464abee9a0776171c79891 # v47.0.1
         with:
           files_yaml: |
             gpu:
@@ -129,7 +129,7 @@ jobs:
             gcc: 11 
     steps:
       - name: Checkout tfhe-rs
-        uses: actions/checkout@1af3b93b6815bc44a9784bd300feb67ff0d1eeb3
+        uses: actions/checkout@8e8c483db84b4bee98b60c0593521ed34d9990e8
         with:
           persist-credentials: 'false'
           token: ${{ env.CHECKOUT_TOKEN }}

.github/workflows/gpu_unsigned_integer_tests.yml

@@ -41,7 +41,7 @@ jobs:
       gpu_test: ${{ env.IS_PULL_REQUEST == 'false' || steps.changed-files.outputs.gpu_any_changed }}
     steps:
       - name: Checkout tfhe-rs
-        uses: actions/checkout@1af3b93b6815bc44a9784bd300feb67ff0d1eeb3
+        uses: actions/checkout@8e8c483db84b4bee98b60c0593521ed34d9990e8
         with:
           fetch-depth: 0
           persist-credentials: 'false'
@@ -49,7 +49,7 @@ jobs:
 
       - name: Check for file changes
         id: changed-files
-        uses: tj-actions/changed-files@24d32ffd492484c1d75e0c0b894501ddb9d30d62 # v47.0.0
+        uses: tj-actions/changed-files@e0021407031f5be11a464abee9a0776171c79891 # v47.0.1
         with:
           files_yaml: |
             gpu:
@@ -117,7 +117,7 @@ jobs:
             gcc: 11
     steps:
       - name: Checkout tfhe-rs
-        uses: actions/checkout@1af3b93b6815bc44a9784bd300feb67ff0d1eeb3
+        uses: actions/checkout@8e8c483db84b4bee98b60c0593521ed34d9990e8
         with:
           persist-credentials: 'false'
           token: ${{ env.CHECKOUT_TOKEN }}

.github/workflows/hpu_hlapi_tests.yml

@@ -32,7 +32,7 @@ jobs:
       hpu_test: ${{ env.IS_PULL_REQUEST == 'false' || steps.changed-files.outputs.hpu_any_changed }}
     steps:
       - name: Checkout tfhe-rs
-        uses: actions/checkout@1af3b93b6815bc44a9784bd300feb67ff0d1eeb3
+        uses: actions/checkout@8e8c483db84b4bee98b60c0593521ed34d9990e8
         with:
           fetch-depth: 0
           persist-credentials: 'false'
@@ -40,7 +40,7 @@ jobs:
 
       - name: Check for file changes
         id: changed-files
-        uses: tj-actions/changed-files@24d32ffd492484c1d75e0c0b894501ddb9d30d62 # v47.0.0
+        uses: tj-actions/changed-files@e0021407031f5be11a464abee9a0776171c79891 # v47.0.1
         with:
           files_yaml: |
             hpu:
@@ -83,7 +83,7 @@ jobs:
     needs: setup-instance
     runs-on: ${{ needs.setup-instance.outputs.runner-name }}
     steps:
-      - uses: actions/checkout@1af3b93b6815bc44a9784bd300feb67ff0d1eeb3
+      - uses: actions/checkout@8e8c483db84b4bee98b60c0593521ed34d9990e8
         with:
           persist-credentials: 'false'
           token: ${{ env.CHECKOUT_TOKEN }}

.github/workflows/integer_long_run_tests.yml

@@ -53,7 +53,7 @@ jobs:
     timeout-minutes: 4320 # 72 hours
     steps:
       - name: Checkout tfhe-rs
-        uses: actions/checkout@1af3b93b6815bc44a9784bd300feb67ff0d1eeb3
+        uses: actions/checkout@8e8c483db84b4bee98b60c0593521ed34d9990e8
         with:
           persist-credentials: 'false'
           token: ${{ secrets.REPO_CHECKOUT_TOKEN }}

.github/workflows/m1_tests.yml

@@ -41,7 +41,7 @@ jobs:
     timeout-minutes: 720
 
     steps:
-      - uses: actions/checkout@1af3b93b6815bc44a9784bd300feb67ff0d1eeb3
+      - uses: actions/checkout@8e8c483db84b4bee98b60c0593521ed34d9990e8
         with:
           persist-credentials: "false"
           token: ${{ env.CHECKOUT_TOKEN }}

.github/workflows/make_release_common.yml

@@ -52,7 +52,7 @@ jobs:
       hash: ${{ steps.hash.outputs.hash }}
     steps:
       - name: Checkout
-        uses: actions/checkout@1af3b93b6815bc44a9784bd300feb67ff0d1eeb3 # v6.0.0
+        uses: actions/checkout@8e8c483db84b4bee98b60c0593521ed34d9990e8 # v6.0.1
         with:
           fetch-depth: 0
           persist-credentials: 'false'
@@ -62,7 +62,7 @@ jobs:
           PACKAGE: ${{ inputs.package-name }}
         run: |
           cargo package -p "${PACKAGE}"
-      - uses: actions/upload-artifact@330a01c490aca151604b8cf639adc76d48f6c5d4 # v5.0.0
+      - uses: actions/upload-artifact@b7c566a772e6b6bfb58ed0dc250532a479d7789f # v6.0.0
         with:
           name: crate-${{ inputs.package-name }}
           path: target/package/*.crate
@@ -93,14 +93,14 @@ jobs:
       id-token: write # Needed for OIDC token exchange on crates.io
     steps:
       - name: Checkout
-        uses: actions/checkout@1af3b93b6815bc44a9784bd300feb67ff0d1eeb3 # v6.0.0
+        uses: actions/checkout@8e8c483db84b4bee98b60c0593521ed34d9990e8 # v6.0.1
         with:
           fetch-depth: 0
           persist-credentials: 'false'
           token: ${{ secrets.REPO_CHECKOUT_TOKEN }}
 
       - name: Download artifact
-        uses: actions/download-artifact@018cc2cf5baa6db3ef3c5f8a56943fffe632ef53 # v6.0.0
+        uses: actions/download-artifact@37930b1c2abaa49bbe596cd826c3c89aef350131 # v7.0.0
         with:
           name: crate-${{ inputs.package-name }}
           path: target/package

.github/workflows/make_release_cuda.yml

@@ -64,7 +64,7 @@ jobs:
       CUDA_PATH: /usr/local/cuda-${{ matrix.cuda }}
     steps:
       - name: Checkout
-        uses: actions/checkout@1af3b93b6815bc44a9784bd300feb67ff0d1eeb3 # v6.0.0
+        uses: actions/checkout@8e8c483db84b4bee98b60c0593521ed34d9990e8 # v6.0.1
         with:
           fetch-depth: 0
           persist-credentials: "false"
@@ -104,7 +104,7 @@ jobs:
         run: |
           cargo package -p tfhe-cuda-backend
 
-      - uses: actions/upload-artifact@330a01c490aca151604b8cf639adc76d48f6c5d4 # v5.0.0
+      - uses: actions/upload-artifact@b7c566a772e6b6bfb58ed0dc250532a479d7789f # v6.0.0
         with:
           name: crate-tfhe-cuda-backend
           path: target/package/*.crate
@@ -174,7 +174,7 @@ jobs:
           GCC_VERSION: ${{ matrix.gcc }}
 
       - name: Download artifact
-        uses: actions/download-artifact@018cc2cf5baa6db3ef3c5f8a56943fffe632ef53 # v6.0.0
+        uses: actions/download-artifact@37930b1c2abaa49bbe596cd826c3c89aef350131 # v7.0.0
         with:
           name: crate-tfhe-cuda-backend
           path: target/package

.github/workflows/make_release_tfhe.yml

@@ -68,7 +68,7 @@ jobs:
       id-token: write # also needed for OIDC token exchange on crates.io and npmjs.com
     steps:
       - name: Checkout
-        uses: actions/checkout@1af3b93b6815bc44a9784bd300feb67ff0d1eeb3 # v6.0.0
+        uses: actions/checkout@8e8c483db84b4bee98b60c0593521ed34d9990e8 # v6.0.1
         with:
           fetch-depth: 0
           persist-credentials: 'false'
@@ -85,7 +85,7 @@ jobs:
           make build_web_js_api_parallel
 
       - name: Authenticate on NPM
-        uses: actions/setup-node@2028fbc5c25fe9cf00d9f06a71cc4710d4507903 # v6.0.0
+        uses: actions/setup-node@395ad3262231945c25e8478fd5baf05154b1d79f # v6.1.0
         with:
           node-version: '24'
           registry-url: 'https://registry.npmjs.org'

.github/workflows/parameters_check.yml

@@ -60,7 +60,7 @@ jobs:
     runs-on: ${{ needs.setup-instance.outputs.runner-name }}
     steps:
       - name: Checkout tfhe-rs
-        uses: actions/checkout@1af3b93b6815bc44a9784bd300feb67ff0d1eeb3
+        uses: actions/checkout@8e8c483db84b4bee98b60c0593521ed34d9990e8
         with:
           persist-credentials: 'false'
           token: ${{ secrets.REPO_CHECKOUT_TOKEN }}
@@ -71,7 +71,7 @@ jobs:
           toolchain: stable
 
       - name: Checkout lattice-estimator
-        uses: actions/checkout@1af3b93b6815bc44a9784bd300feb67ff0d1eeb3
+        uses: actions/checkout@8e8c483db84b4bee98b60c0593521ed34d9990e8
         with:
           repository: malb/lattice-estimator
           path: lattice_estimator

.github/workflows/unverified_prs.yml

@@ -17,7 +17,7 @@ jobs:
       issues: read # Needed to fetch all issues
       pull-requests: write # Needed to write message and close the PR
     steps:
-      - uses: actions/stale@5f858e3efba33a5ca4407a664cc011ad407f2008 # v10.1.0
+      - uses: actions/stale@997185467fa4f803885201cee163a9f38240193d # v10.1.1
         with:
           stale-pr-message: 'This PR is unverified and has been open for 2 days, it will now be closed. If you want to contribute please sign the CLA as indicated by the bot.'
           days-before-stale: 2

.gitignore

@@ -10,6 +10,7 @@ target/
 **/*.rmeta
 **/Cargo.lock
 **/*.bin
+**/.DS_Store
 
 # Some of our bench outputs
 /tfhe/benchmarks_parameters

CODEOWNERS

@@ -11,7 +11,7 @@
 /tfhe/src/core_crypto/gpu               @agnesLeroy
 /tfhe/src/core_crypto/hpu               @zama-ai/hardware
 
-/tfhe/src/shortint/                     @mayeul-zama
+/tfhe/src/shortint/                     @mayeul-zama @nsarlin-zama
 
 /tfhe/src/integer/                      @tmontaigu
 /tfhe/src/integer/gpu                   @agnesLeroy
@@ -19,8 +19,12 @@
 
 /tfhe/src/high_level_api/               @tmontaigu
 
+/tfhe-zk-pok/                           @nsarlin-zama
+
 /tfhe-benchmark/                        @soonum
 
+/utils/                                 @nsarlin-zama
+
 /Makefile                               @IceTDrinker @soonum
 
 /mockups/tfhe-hpu-mockup                @zama-ai/hardware

Cargo.toml

@@ -36,6 +36,8 @@ rayon = "1.11"
 serde = { version = "1.0", default-features = false }
 wasm-bindgen = "0.2.101"
 getrandom = "0.2.8"
+# The project maintainers consider that this is the last version of the 1.3 branch, any newer version should not be trusted
+bincode = "=1.3.3"
 
 [profile.bench]
 lto = "fat"

Makefile

@@ -1300,13 +1300,14 @@ run_web_js_api_parallel: build_web_js_api_parallel setup_venv
 	--browser-path $(browser_path) \
 	--driver-path $(driver_path) \
 	--browser-kind  $(browser_kind) \
-	--server-cmd "npm run server" \
+	--server-cmd $(server_cmd) \
 	--server-workdir "$(WEB_SERVER_DIR)" \
 	--id-pattern $(filter)
 
 test_web_js_api_parallel_chrome: browser_path = "$(WEB_RUNNER_DIR)/chrome/chrome-linux64/chrome"
 test_web_js_api_parallel_chrome: driver_path = "$(WEB_RUNNER_DIR)/chrome/chromedriver-linux64/chromedriver"
 test_web_js_api_parallel_chrome: browser_kind = chrome
+test_web_js_api_parallel_chrome: server_cmd = "npm run server:multithreaded"
 test_web_js_api_parallel_chrome: filter = Test
 
 .PHONY: test_web_js_api_parallel_chrome # Run tests for the web wasm api on Chrome
@@ -1322,6 +1323,7 @@ test_web_js_api_parallel_chrome_ci: setup_venv
 test_web_js_api_parallel_firefox: browser_path = "$(WEB_RUNNER_DIR)/firefox/firefox/firefox"
 test_web_js_api_parallel_firefox: driver_path = "$(WEB_RUNNER_DIR)/firefox/geckodriver"
 test_web_js_api_parallel_firefox: browser_kind = firefox
+test_web_js_api_parallel_firefox: server_cmd = "npm run server:multithreaded"
 test_web_js_api_parallel_firefox: filter = Test
 
 .PHONY: test_web_js_api_parallel_firefox # Run tests for the web wasm api on Firefox
@@ -1571,6 +1573,7 @@ bench_pbs128_gpu: install_rs_check_toolchain
 bench_web_js_api_parallel_chrome: browser_path = "$(WEB_RUNNER_DIR)/chrome/chrome-linux64/chrome"
 bench_web_js_api_parallel_chrome: driver_path = "$(WEB_RUNNER_DIR)/chrome/chromedriver-linux64/chromedriver"
 bench_web_js_api_parallel_chrome: browser_kind = chrome
+bench_web_js_api_parallel_chrome: server_cmd = "npm run server:multithreaded"
 bench_web_js_api_parallel_chrome: filter = Bench
 
 .PHONY: bench_web_js_api_parallel_chrome # Run benchmarks for the web wasm api
@@ -1586,6 +1589,7 @@ bench_web_js_api_parallel_chrome_ci: setup_venv
 bench_web_js_api_parallel_firefox: browser_path = "$(WEB_RUNNER_DIR)/firefox/firefox/firefox"
 bench_web_js_api_parallel_firefox: driver_path = "$(WEB_RUNNER_DIR)/firefox/geckodriver"
 bench_web_js_api_parallel_firefox: browser_kind = firefox
+bench_web_js_api_parallel_firefox: server_cmd = "npm run server:multithreaded"
 bench_web_js_api_parallel_firefox: filter = Bench
 
 .PHONY: bench_web_js_api_parallel_firefox # Run benchmarks for the web wasm api
@@ -1598,6 +1602,38 @@ bench_web_js_api_parallel_firefox_ci: setup_venv
 	nvm use $(NODE_VERSION) && \
 	$(MAKE) bench_web_js_api_parallel_firefox
 
+bench_web_js_api_unsafe_coop_chrome: browser_path = "$(WEB_RUNNER_DIR)/chrome/chrome-linux64/chrome"
+bench_web_js_api_unsafe_coop_chrome: driver_path = "$(WEB_RUNNER_DIR)/chrome/chromedriver-linux64/chromedriver"
+bench_web_js_api_unsafe_coop_chrome: browser_kind = chrome
+bench_web_js_api_unsafe_coop_chrome: server_cmd = "npm run server:unsafe-coop"
+bench_web_js_api_unsafe_coop_chrome: filter = ZeroKnowledgeBench # Only bench zk with unsafe coop
+
+.PHONY: bench_web_js_api_unsafe_coop_chrome # Run benchmarks for the web wasm api without cross-origin isolation
+bench_web_js_api_unsafe_coop_chrome: run_web_js_api_parallel
+
+.PHONY: bench_web_js_api_unsafe_coop_chrome_ci # Run benchmarks for the web wasm api without cross-origin isolation
+bench_web_js_api_unsafe_coop_chrome_ci: setup_venv
+	source ~/.nvm/nvm.sh && \
+	nvm install $(NODE_VERSION) && \
+	nvm use $(NODE_VERSION) && \
+	$(MAKE) bench_web_js_api_unsafe_coop_chrome
+
+bench_web_js_api_unsafe_coop_firefox: browser_path = "$(WEB_RUNNER_DIR)/firefox/firefox/firefox"
+bench_web_js_api_unsafe_coop_firefox: driver_path = "$(WEB_RUNNER_DIR)/firefox/geckodriver"
+bench_web_js_api_unsafe_coop_firefox: browser_kind = firefox
+bench_web_js_api_unsafe_coop_firefox: server_cmd = "npm run server:unsafe-coop"
+bench_web_js_api_unsafe_coop_firefox: filter = ZeroKnowledgeBench # Only bench zk with unsafe coop
+
+.PHONY: bench_web_js_api_unsafe_coop_firefox # Run benchmarks for the web wasm api without cross-origin isolation
+bench_web_js_api_unsafe_coop_firefox: run_web_js_api_parallel
+
+.PHONY: bench_web_js_api_unsafe_coop_firefox_ci # Run benchmarks for the web wasm api without cross-origin isolation
+bench_web_js_api_unsafe_coop_firefox_ci: setup_venv
+	source ~/.nvm/nvm.sh && \
+	nvm install $(NODE_VERSION) && \
+	nvm use $(NODE_VERSION) && \
+	$(MAKE) bench_web_js_api_unsafe_coop_firefox
+
 .PHONY: bench_hlapi # Run benchmarks for integer operations
 bench_hlapi: install_rs_check_toolchain
 	RUSTFLAGS="$(RUSTFLAGS)" __TFHE_RS_BENCH_BIT_SIZES_SET=$(BIT_SIZES_SET) \

apps/test-vectors/checksums.sha256

@@ -1,24 +1,32 @@
+08f31a47c29cc4d72ad32c0b5411fa20b3deef5b84558dd2fb892d3cdf90528a  data/toy_params/glwe_after_id_br_karatsuba.cbor
 29b6e3e7d27700004b70dca24d225816500490e2d6ee49b9af05837fd421896b  data/valid_params_128/lwe_after_spec_pbs.cbor
 2c70d1d78cc3760733850a353ace2b9c4705e840141b75841739e90e51247e18  data/valid_params_128/small_lwe_secret_key.cbor
+2fb4bb45c259b8383da10fc8f9459c40a6972c49b1696eb107f0a75640724be5  data/toy_params/lwe_after_id_pbs_karatsuba.cbor
 36c9080b636475fcacca503ce041bbfeee800fd3e1890dee559ea18defff9fe8  data/toy_params/glwe_after_id_br.cbor
 377761beeb4216cf5aa2624a8b64b8259f5a75c32d28e850be8bced3a0cdd6f5  data/toy_params/ksk.cbor
 59dba26d457f96478eda130cab5301fce86f23c6a8807de42f2a1e78c4985ca7  data/valid_params_128/lwe_ks.cbor
+5d80dd93fefae4f4f89484dfcd65bbe99cc32e7e3b0a90c33dd0d77516c0a023  data/valid_params_128/glwe_after_id_br_karatsuba.cbor
 656f0009c7834c5bcb61621e222047516054b9bc5d0593d474ab8f1c086b67a6  data/valid_params_128/lwe_after_id_pbs.cbor
 699580ca92b9c2f9e1f57fb1e312c9e8cb29714f7acdef9d2ba05f798546751f  data/toy_params/lwe_sum.cbor
 6e54ab41056984595b077baff70236d934308cf5c0c33b4482fbfb129b3756c6  data/valid_params_128/glwe_after_id_br.cbor
 70f5e5728822de05b49071efb5ec28551b0f5cc87aa709a455d8e7f04b9c96ee  data/toy_params/lwe_after_id_pbs.cbor
+76a5c52cab7fec1dc167da676c6cd39479cda6b2bb9f4e0573cb7d99c2692faa  data/valid_params_128/lwe_after_id_pbs_karatsuba.cbor
 7cc6803f5fbc3d5a1bf597f2b979ce17eecd3d6baca12183dea21022a7b65c52  data/toy_params/bsk.cbor
 7f3c40a134623b44779a556212477fea26eaed22450f3b6faeb8721d63699972  data/valid_params_128/lwe_sum.cbor
 837b3bd3245d4d0534ed255fdef896fb4fa6998a258a14543dfdadd0bfc9b6dd  data/toy_params/lwe_prod.cbor
+9ece8ca9c1436258b94e8c5e629b8722f9b18fdd415dd5209b6167a9dde8491c  data/toy_params/glwe_after_spec_br_karatsuba.cbor
 aa44aea29efd6d9e4d35a21a625d9cba155672e3f7ed3eddee1e211e62ad146b  data/valid_params_128/lwe_ms.cbor
 b7a037b9eaa88d6385167579b93e26a0cb6976d9b8967416fd1173e113bda199  data/valid_params_128/large_lwe_secret_key.cbor
+b7b8e3586128887bd682120f3e3a43156139bce5e3fe0b03284f8753a864d647  data/toy_params/lwe_after_spec_pbs_karatsuba.cbor
 bd00a8ae7494e400de5753029552ee1647efe7e17409b863a26a13b081099b8c  data/toy_params/lwe_after_spec_pbs.cbor
 c6df98676de04fe54b5ffc2eb30a82ebb706c9d7d5a4e0ed509700fec88761f7  data/toy_params/lwe_ms.cbor
 c7d5a864d5616a7d8ad50bbf40416e41e6c9b60c546dc14d4aa8fc40a418baa7  data/toy_params/large_lwe_secret_key.cbor
 c806533b325b1009db38be2f9bef5f3b2fad6b77b4c71f2855ccc9d3b4162e98  data/valid_params_128/lwe_b.cbor
 c9eb75bd2993639348a679cf48c06e3c38d1a513f48e5b0ce0047cea8cff6bbc  data/toy_params/lwe_a.cbor
+d3391969acf26dc69de0927ba279139d8d79999944069addc8ff469ad6c5ae2d  data/valid_params_128/lwe_after_spec_pbs_karatsuba.cbor
 d6da5baef0e787f6be56e218d8354e26904652602db964844156fdff08350ce6  data/toy_params/lwe_ks.cbor
 e591ab9af1b6a0aede273f9a3abb65a4c387feb5fa06a6959e9314058ca0f7e5  data/valid_params_128/ksk.cbor
+e59b002df3a9b01ad321ec51cf076fa35131ab9dbef141d1c54b717d61426c92  data/valid_params_128/glwe_after_spec_br_karatsuba.cbor
 e628354c81508a2d888016e8282df363dd12f1e19190b6475d4eb9d7ab8ae007  data/valid_params_128/glwe_after_spec_br.cbor
 e69d2d2c064fc8c0460b39191ca65338146990349954f5ec5ebd01d93610e7eb  data/valid_params_128/lwe_a.cbor
 e76c24b2a0c9a842ad13dda35473c2514f9e7d20983b5ea0759c4521a91626d9  data/valid_params_128/lwe_prod.cbor

apps/test-vectors/data/README.md

@@ -39,6 +39,9 @@ The following values are generated:
 | `glwe_after_spec_br` | The glwe returned by the application of the spec blind rotation on the mod switched ciphertexts.             | `GlweCiphertext<Vec<u64>>` | rot spec LUT |
 | `lwe_after_spec_pbs` | The lwe returned by the application of the sample extract operation on the output of the spec blind rotation | `LweCiphertext<Vec<u64>>`  | `spec(A)`    |
 
+Ciphertexts with the `_karatsuba` suffix are generated using the Karatsuba polynomial multiplication algorithm in the blind rotation, while default ciphertexts are generated using an FFT multiplication.
+This makes it easier to reproduce bit exact results.
+
 ### Encodings
 #### Non native encoding
 Warning: TFHE-rs uses a specific encoding for non native (ie: u32, u64) power of two ciphertext modulus. This encoding puts the encoded value in the high bits of the native integer.

apps/test-vectors/data/toy_params/glwe_after_id_br_karatsuba.cbor

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:08f31a47c29cc4d72ad32c0b5411fa20b3deef5b84558dd2fb892d3cdf90528a
+size 4679

apps/test-vectors/data/toy_params/glwe_after_spec_br_karatsuba.cbor

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9ece8ca9c1436258b94e8c5e629b8722f9b18fdd415dd5209b6167a9dde8491c
+size 4679

apps/test-vectors/data/toy_params/lwe_after_id_pbs_karatsuba.cbor

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2fb4bb45c259b8383da10fc8f9459c40a6972c49b1696eb107f0a75640724be5
+size 2365

apps/test-vectors/data/toy_params/lwe_after_spec_pbs_karatsuba.cbor

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b7b8e3586128887bd682120f3e3a43156139bce5e3fe0b03284f8753a864d647
+size 2365

apps/test-vectors/data/valid_params_128/glwe_after_id_br_karatsuba.cbor

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5d80dd93fefae4f4f89484dfcd65bbe99cc32e7e3b0a90c33dd0d77516c0a023
+size 36935

apps/test-vectors/data/valid_params_128/glwe_after_spec_br_karatsuba.cbor

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e59b002df3a9b01ad321ec51cf076fa35131ab9dbef141d1c54b717d61426c92
+size 36935

apps/test-vectors/data/valid_params_128/lwe_after_id_pbs_karatsuba.cbor

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:76a5c52cab7fec1dc167da676c6cd39479cda6b2bb9f4e0573cb7d99c2692faa
+size 18493

apps/test-vectors/data/valid_params_128/lwe_after_spec_pbs_karatsuba.cbor

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d3391969acf26dc69de0927ba279139d8d79999944069addc8ff469ad6c5ae2d
+size 18493

apps/test-vectors/src/main.rs

@@ -265,6 +265,7 @@ fn generate_test_vectors<P: AsRef<Path>>(
 
     let mut id_lut = encoding.encode_lut(glwe_dimension, polynomial_size, ID_LUT);
     assert_data_not_zero(&id_lut);
+    let mut id_lut_karatsuba = id_lut.clone();
 
     blind_rotate_assign(&modswitched, &mut id_lut, &fourier_bsk);
     assert_data_not_zero(&id_lut);
@@ -287,8 +288,32 @@ fn generate_test_vectors<P: AsRef<Path>>(
     assert_data_not_zero(&lwe_pbs_id);
     store_data(path, &lwe_pbs_id, "lwe_after_id_pbs");
 
+    blind_rotate_karatsuba_assign(&modswitched, &mut id_lut_karatsuba, &bsk);
+    store_data(path, &id_lut_karatsuba, "glwe_after_id_br_karatsuba");
+
+    let mut lwe_pbs_karatsuba_id = LweCiphertext::new(
+        0u64,
+        glwe_dimension
+            .to_equivalent_lwe_dimension(polynomial_size)
+            .to_lwe_size(),
+        encoding.ciphertext_modulus,
+    );
+
+    extract_lwe_sample_from_glwe_ciphertext(
+        &id_lut_karatsuba,
+        &mut lwe_pbs_karatsuba_id,
+        MonomialDegree(0),
+    );
+
+    let decrypted_pbs_id = decrypt_lwe_ciphertext(&large_lwe_secret_key, &lwe_pbs_karatsuba_id);
+    let res = encoding.decode(decrypted_pbs_id);
+
+    assert_eq!(res, MSG_A);
+    store_data(path, &lwe_pbs_karatsuba_id, "lwe_after_id_pbs_karatsuba");
+
     let mut spec_lut = encoding.encode_lut(glwe_dimension, polynomial_size, SPEC_LUT);
     assert_data_not_zero(&spec_lut);
+    let mut spec_lut_karatsuba = spec_lut.clone();
 
     blind_rotate_assign(&modswitched, &mut spec_lut, &fourier_bsk);
     assert_data_not_zero(&spec_lut);
@@ -310,6 +335,33 @@ fn generate_test_vectors<P: AsRef<Path>>(
     assert_eq!(res, SPEC_LUT(MSG_A));
     assert_data_not_zero(&lwe_pbs_spec);
     store_data(path, &lwe_pbs_spec, "lwe_after_spec_pbs");
+
+    blind_rotate_karatsuba_assign(&modswitched, &mut spec_lut_karatsuba, &bsk);
+    store_data(path, &spec_lut_karatsuba, "glwe_after_spec_br_karatsuba");
+
+    let mut lwe_pbs_karatsuba_spec = LweCiphertext::new(
+        0u64,
+        glwe_dimension
+            .to_equivalent_lwe_dimension(polynomial_size)
+            .to_lwe_size(),
+        encoding.ciphertext_modulus,
+    );
+
+    extract_lwe_sample_from_glwe_ciphertext(
+        &spec_lut_karatsuba,
+        &mut lwe_pbs_karatsuba_spec,
+        MonomialDegree(0),
+    );
+
+    let decrypted_pbs_spec = decrypt_lwe_ciphertext(&large_lwe_secret_key, &lwe_pbs_karatsuba_spec);
+    let res = encoding.decode(decrypted_pbs_spec);
+
+    assert_eq!(res, SPEC_LUT(MSG_A));
+    store_data(
+        path,
+        &lwe_pbs_karatsuba_spec,
+        "lwe_after_spec_pbs_karatsuba",
+    );
 }
 
 fn rm_dir_except_readme<P: AsRef<Path>>(dir: P) {

backends/tfhe-cuda-backend/cuda/include/aes/aes_utilities.h

@@ -35,7 +35,8 @@ template <typename Torus> struct int_aes_lut_buffers {
         params.glwe_dimension, params.polynomial_size, params.message_modulus,
         params.carry_modulus, and_lambda, allocate_gpu_memory);
     auto active_streams_and_lut = streams.active_gpu_subset(
-        SBOX_MAX_AND_GATES * num_aes_inputs * sbox_parallelism);
+        SBOX_MAX_AND_GATES * num_aes_inputs * sbox_parallelism,
+        params.pbs_type);
     this->and_lut->broadcast_lut(active_streams_and_lut);
     this->and_lut->setup_gemm_batch_ks_temp_buffers(size_tracker);
 
@@ -50,8 +51,8 @@ template <typename Torus> struct int_aes_lut_buffers {
         this->flush_lut->get_degree(0), this->flush_lut->get_max_degree(0),
         params.glwe_dimension, params.polynomial_size, params.message_modulus,
         params.carry_modulus, flush_lambda, allocate_gpu_memory);
-    auto active_streams_flush_lut =
-        streams.active_gpu_subset(AES_STATE_BITS * num_aes_inputs);
+    auto active_streams_flush_lut = streams.active_gpu_subset(
+        AES_STATE_BITS * num_aes_inputs, params.pbs_type);
     this->flush_lut->broadcast_lut(active_streams_flush_lut);
     this->flush_lut->setup_gemm_batch_ks_temp_buffers(size_tracker);
 
@@ -65,7 +66,8 @@ template <typename Torus> struct int_aes_lut_buffers {
         this->carry_lut->get_degree(0), this->carry_lut->get_max_degree(0),
         params.glwe_dimension, params.polynomial_size, params.message_modulus,
         params.carry_modulus, carry_lambda, allocate_gpu_memory);
-    auto active_streams_carry_lut = streams.active_gpu_subset(num_aes_inputs);
+    auto active_streams_carry_lut =
+        streams.active_gpu_subset(num_aes_inputs, params.pbs_type);
     this->carry_lut->broadcast_lut(active_streams_carry_lut);
     this->carry_lut->setup_gemm_batch_ks_temp_buffers(size_tracker);
   }

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

@@ -8,7 +8,8 @@
 
 extern std::mutex m;
 extern bool p2p_enabled;
-extern const int THRESHOLD_MULTI_GPU;
+extern const int THRESHOLD_MULTI_GPU_WITH_MULTI_BIT_PARAMS;
+extern const int THRESHOLD_MULTI_GPU_WITH_CLASSICAL_PARAMS;
 
 extern "C" {
 int32_t cuda_setup_multi_gpu(int device_0_id);
@@ -39,7 +40,8 @@ get_variant_element(const std::variant<std::vector<Torus>, Torus> &variant,
   }
 }
 
-uint32_t get_active_gpu_count(uint32_t num_inputs, uint32_t gpu_count);
+uint32_t get_active_gpu_count(uint32_t num_inputs, uint32_t gpu_count,
+                              PBS_TYPE pbs_type);
 
 int get_num_inputs_on_gpu(int total_num_inputs, int gpu_index, int gpu_count);
 
@@ -73,9 +75,10 @@ public:
 
   // Returns a subset of this set as an active subset. An active subset is one
   // that is temporarily used to perform some computation
-  CudaStreams active_gpu_subset(int num_radix_blocks) {
-    return CudaStreams(_streams, _gpu_indexes,
-                       get_active_gpu_count(num_radix_blocks, _gpu_count));
+  CudaStreams active_gpu_subset(int num_radix_blocks, PBS_TYPE pbs_type) {
+    return CudaStreams(
+        _streams, _gpu_indexes,
+        get_active_gpu_count(num_radix_blocks, _gpu_count, pbs_type));
   }
 
   // Returns a CudaStreams struct containing only the ith stream

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

@@ -20,7 +20,8 @@ template <typename Torus> struct boolean_bitop_buffer {
     gpu_memory_allocated = allocate_gpu_memory;
     this->op = op;
     this->params = params;
-    auto active_streams = streams.active_gpu_subset(lwe_ciphertext_count);
+    auto active_streams =
+        streams.active_gpu_subset(lwe_ciphertext_count, params.pbs_type);
     this->unchecked = is_unchecked;
     switch (op) {
     case BITAND:
@@ -119,7 +120,8 @@ template <typename Torus> struct int_bitop_buffer {
     gpu_memory_allocated = allocate_gpu_memory;
     this->op = op;
     this->params = params;
-    auto active_streams = streams.active_gpu_subset(num_radix_blocks);
+    auto active_streams =
+        streams.active_gpu_subset(num_radix_blocks, params.pbs_type);
     switch (op) {
     case BITAND:
     case BITOR:
@@ -216,7 +218,8 @@ template <typename Torus> struct boolean_bitnot_buffer {
           message_extract_lut->get_max_degree(0), params.glwe_dimension,
           params.polynomial_size, params.message_modulus, params.carry_modulus,
           lut_f_message_extract, gpu_memory_allocated);
-      auto active_streams = streams.active_gpu_subset(lwe_ciphertext_count);
+      auto active_streams =
+          streams.active_gpu_subset(lwe_ciphertext_count, params.pbs_type);
       message_extract_lut->broadcast_lut(active_streams);
     }
   }

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

@@ -39,7 +39,8 @@ template <typename Torus> struct int_extend_radix_with_sign_msb_buffer {
           },
           allocate_gpu_memory);
 
-      auto active_streams = streams.active_gpu_subset(num_radix_blocks);
+      auto active_streams =
+          streams.active_gpu_subset(num_radix_blocks, params.pbs_type);
       lut->broadcast_lut(active_streams);
 
       this->last_block = new CudaRadixCiphertextFFI;

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

@@ -14,7 +14,8 @@ template <typename Torus> struct int_zero_out_if_buffer {
                          uint64_t &size_tracker) {
     gpu_memory_allocated = allocate_gpu_memory;
     this->params = params;
-    auto active_streams = streams.active_gpu_subset(num_radix_blocks);
+    auto active_streams =
+        streams.active_gpu_subset(num_radix_blocks, params.pbs_type);
 
     tmp = new CudaRadixCiphertextFFI;
     create_zero_radix_ciphertext_async<Torus>(
@@ -114,9 +115,11 @@ template <typename Torus> struct int_cmux_buffer {
         predicate_lut->get_lut_indexes(0, 0), h_lut_indexes,
         2 * num_radix_blocks * sizeof(Torus), streams.stream(0),
         streams.gpu_index(0), allocate_gpu_memory);
-    auto active_streams_pred = streams.active_gpu_subset(2 * num_radix_blocks);
+    auto active_streams_pred =
+        streams.active_gpu_subset(2 * num_radix_blocks, params.pbs_type);
     predicate_lut->broadcast_lut(active_streams_pred);
-    auto active_streams_msg = streams.active_gpu_subset(num_radix_blocks);
+    auto active_streams_msg =
+        streams.active_gpu_subset(num_radix_blocks, params.pbs_type);
     message_extract_lut->broadcast_lut(active_streams_msg);
   }
 

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

@@ -52,7 +52,8 @@ template <typename Torus> struct int_are_all_block_true_buffer {
         params.glwe_dimension, params.polynomial_size, params.message_modulus,
         params.carry_modulus, is_max_value_f, gpu_memory_allocated);
 
-    auto active_streams = streams.active_gpu_subset(max_chunks);
+    auto active_streams =
+        streams.active_gpu_subset(max_chunks, params.pbs_type);
     is_max_value->broadcast_lut(active_streams);
   }
 
@@ -108,7 +109,8 @@ template <typename Torus> struct int_comparison_eq_buffer {
         params.glwe_dimension, params.polynomial_size, params.message_modulus,
         params.carry_modulus, is_non_zero_lut_f, gpu_memory_allocated);
 
-    auto active_streams = streams.active_gpu_subset(num_radix_blocks);
+    auto active_streams =
+        streams.active_gpu_subset(num_radix_blocks, params.pbs_type);
     is_non_zero_lut->broadcast_lut(active_streams);
 
     // Scalar may have up to num_radix_blocks blocks
@@ -238,7 +240,8 @@ template <typename Torus> struct int_tree_sign_reduction_buffer {
         tree_inner_leaf_lut->get_max_degree(0), params.glwe_dimension,
         params.polynomial_size, params.message_modulus, params.carry_modulus,
         block_selector_f, gpu_memory_allocated);
-    auto active_streams = streams.active_gpu_subset(num_radix_blocks);
+    auto active_streams =
+        streams.active_gpu_subset(num_radix_blocks, params.pbs_type);
     tree_inner_leaf_lut->broadcast_lut(active_streams);
   }
 
@@ -390,7 +393,8 @@ template <typename Torus> struct int_comparison_buffer {
     this->op = op;
     this->is_signed = is_signed;
 
-    auto active_streams = streams.active_gpu_subset(num_radix_blocks);
+    auto active_streams =
+        streams.active_gpu_subset(num_radix_blocks, params.pbs_type);
 
     identity_lut_f = [](Torus x) -> Torus { return x; };
 
@@ -523,7 +527,7 @@ template <typename Torus> struct int_comparison_buffer {
           signed_lut->get_degree(0), signed_lut->get_max_degree(0),
           params.glwe_dimension, params.polynomial_size, params.message_modulus,
           params.carry_modulus, signed_lut_f, gpu_memory_allocated);
-      auto active_streams = streams.active_gpu_subset(1);
+      auto active_streams = streams.active_gpu_subset(1, params.pbs_type);
       signed_lut->broadcast_lut(active_streams);
     }
     preallocated_h_lut = (Torus *)malloc(

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

@@ -65,6 +65,16 @@ void cleanup_cuda_integer_compress_radix_ciphertext_128(CudaStreamsFFI streams,
 
 void cleanup_cuda_integer_decompress_radix_ciphertext_128(
     CudaStreamsFFI streams, int8_t **mem_ptr_void);
+
+void cuda_integer_extract_glwe_128(
+    CudaStreamsFFI streams, void *glwe_array_out,
+    CudaPackedGlweCiphertextListFFI const *glwe_list,
+    uint32_t const glwe_index);
+
+void cuda_integer_extract_glwe_64(
+    CudaStreamsFFI streams, void *glwe_array_out,
+    CudaPackedGlweCiphertextListFFI const *glwe_list,
+    uint32_t const glwe_index);
 }
 
 #endif

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

@@ -116,7 +116,8 @@ template <typename Torus> struct int_decompression {
           effective_compression_carry_modulus,
           encryption_params.message_modulus, encryption_params.carry_modulus,
           decompression_rescale_f, gpu_memory_allocated);
-      auto active_streams = streams.active_gpu_subset(num_blocks_to_decompress);
+      auto active_streams = streams.active_gpu_subset(
+          num_blocks_to_decompress, decompression_rescale_lut->params.pbs_type);
       decompression_rescale_lut->broadcast_lut(active_streams);
     }
   }

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

@@ -356,7 +356,8 @@ template <typename Torus> struct unsigned_int_div_rem_2_2_memory {
           luts[j]->get_degree(0), luts[j]->get_max_degree(0),
           params.glwe_dimension, params.polynomial_size, params.message_modulus,
           params.carry_modulus, lut_f_message_extract, gpu_memory_allocated);
-      auto active_streams = streams.active_gpu_subset(num_blocks);
+      auto active_streams =
+          streams.active_gpu_subset(num_blocks, params.pbs_type);
       luts[j]->broadcast_lut(active_streams);
     }
   }
@@ -1012,7 +1013,7 @@ template <typename Torus> struct unsigned_int_div_rem_memory {
           masking_luts_1[i]->get_max_degree(0), params.glwe_dimension,
           params.polynomial_size, params.message_modulus, params.carry_modulus,
           lut_f_masking, gpu_memory_allocated);
-      auto active_streams_1 = streams.active_gpu_subset(1);
+      auto active_streams_1 = streams.active_gpu_subset(1, params.pbs_type);
       masking_luts_1[i]->broadcast_lut(active_streams_1);
 
       generate_device_accumulator<Torus>(
@@ -1021,7 +1022,8 @@ template <typename Torus> struct unsigned_int_div_rem_memory {
           masking_luts_2[i]->get_max_degree(0), params.glwe_dimension,
           params.polynomial_size, params.message_modulus, params.carry_modulus,
           lut_f_masking, gpu_memory_allocated);
-      auto active_streams_2 = streams.active_gpu_subset(num_blocks);
+      auto active_streams_2 =
+          streams.active_gpu_subset(num_blocks, params.pbs_type);
       masking_luts_2[i]->broadcast_lut(active_streams_2);
     }
 
@@ -1040,7 +1042,8 @@ template <typename Torus> struct unsigned_int_div_rem_memory {
 
     int_radix_lut<Torus> *luts[2] = {message_extract_lut_1,
                                      message_extract_lut_2};
-    auto active_streams = streams.active_gpu_subset(num_blocks);
+    auto active_streams =
+        streams.active_gpu_subset(num_blocks, params.pbs_type);
     for (int j = 0; j < 2; j++) {
       generate_device_accumulator<Torus>(
           streams.stream(0), streams.gpu_index(0), luts[j]->get_lut(0, 0),
@@ -1128,7 +1131,8 @@ template <typename Torus> struct unsigned_int_div_rem_memory {
 
     // merge_overflow_flags_luts
     merge_overflow_flags_luts = new int_radix_lut<Torus> *[num_bits_in_message];
-    auto active_gpu_count_for_bits = streams.active_gpu_subset(1);
+    auto active_gpu_count_for_bits =
+        streams.active_gpu_subset(1, params.pbs_type);
     for (int i = 0; i < num_bits_in_message; i++) {
       auto lut_f_bit = [i](Torus x, Torus y) -> Torus {
         return (x == 0 && y == 0) << i;
@@ -1152,7 +1156,8 @@ template <typename Torus> struct unsigned_int_div_rem_memory {
                               uint32_t num_blocks, bool allocate_gpu_memory,
                               uint64_t &size_tracker) {
     gpu_memory_allocated = allocate_gpu_memory;
-    auto active_streams = streams.active_gpu_subset(2 * num_blocks);
+    auto active_streams =
+        streams.active_gpu_subset(2 * num_blocks, params.pbs_type);
     this->params = params;
 
     if (params.message_modulus == 4 && params.carry_modulus == 4 &&
@@ -1473,7 +1478,8 @@ template <typename Torus> struct int_div_rem_memory {
                      bool allocate_gpu_memory, uint64_t &size_tracker) {
 
     gpu_memory_allocated = allocate_gpu_memory;
-    this->active_streams = streams.active_gpu_subset(num_blocks);
+    this->active_streams =
+        streams.active_gpu_subset(num_blocks, params.pbs_type);
     this->params = params;
     this->is_signed = is_signed;
 
@@ -1559,7 +1565,7 @@ template <typename Torus> struct int_div_rem_memory {
           params.polynomial_size, params.message_modulus, params.carry_modulus,
           f_compare_extracted_signed_bits, gpu_memory_allocated);
       auto active_gpu_count_cmp =
-          streams.active_gpu_subset(1); // only 1 block needed
+          streams.active_gpu_subset(1, params.pbs_type); // only 1 block needed
       compare_signed_bits_lut->broadcast_lut(active_gpu_count_cmp);
     }
   }

backends/tfhe-cuda-backend/cuda/include/integer/ilog2.h

@@ -20,7 +20,8 @@ template <typename Torus> struct int_prepare_count_of_consecutive_bits_buffer {
     this->allocate_gpu_memory = allocate_gpu_memory;
     this->direction = direction;
     this->bit_value = bit_value;
-    auto active_streams = streams.active_gpu_subset(num_radix_blocks);
+    auto active_streams =
+        streams.active_gpu_subset(num_radix_blocks, params.pbs_type);
     this->univ_lut_mem =
         new int_radix_lut<Torus>(streams, params, 1, num_radix_blocks,
                                  allocate_gpu_memory, size_tracker);
@@ -246,7 +247,8 @@ template <typename Torus> struct int_ilog2_buffer {
                                 params.glwe_dimension, params.polynomial_size,
                                 params.message_modulus, params.carry_modulus,
                                 lut_message_lambda, allocate_gpu_memory);
-    auto active_streams = streams.active_gpu_subset(counter_num_blocks);
+    auto active_streams =
+        streams.active_gpu_subset(counter_num_blocks, params.pbs_type);
     lut_message_not->broadcast_lut(active_streams);
 
     this->lut_carry_not =

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

@@ -371,7 +371,8 @@ struct int_radix_lut_custom_input_output {
     this->num_input_blocks = num_input_blocks;
     this->gpu_memory_allocated = allocate_gpu_memory;
 
-    this->active_streams = streams.active_gpu_subset(num_radix_blocks);
+    this->active_streams =
+        streams.active_gpu_subset(num_radix_blocks, params.pbs_type);
   }
 
   void setup_degrees() {
@@ -382,14 +383,18 @@ struct int_radix_lut_custom_input_output {
 
   void allocate_pbs_buffers(int_radix_params params, uint32_t num_radix_blocks,
                             bool allocate_gpu_memory, uint64_t &size_tracker) {
+
+    int threshold = (params.pbs_type == PBS_TYPE::MULTI_BIT)
+                        ? THRESHOLD_MULTI_GPU_WITH_MULTI_BIT_PARAMS
+                        : THRESHOLD_MULTI_GPU_WITH_CLASSICAL_PARAMS;
+
     for (uint i = 0; i < active_streams.count(); i++) {
       cuda_set_device(active_streams.gpu_index(i));
       int8_t *gpu_pbs_buffer;
-      auto num_blocks_on_gpu =
-          std::min((int)num_radix_blocks,
-                   std::max(THRESHOLD_MULTI_GPU,
-                            get_num_inputs_on_gpu(num_radix_blocks, i,
-                                                  active_streams.count())));
+      auto num_blocks_on_gpu = std::min(
+          (int)num_radix_blocks,
+          std::max(threshold, get_num_inputs_on_gpu(num_radix_blocks, i,
+                                                    active_streams.count())));
 
       uint64_t size = 0;
       execute_scratch_pbs<OutputTorus>(
@@ -424,18 +429,22 @@ struct int_radix_lut_custom_input_output {
     /// back to the original indexing
     multi_gpu_alloc_lwe_async(active_streams, lwe_array_in_vec,
                               num_radix_blocks, params.big_lwe_dimension + 1,
-                              size_tracker, allocate_gpu_memory);
+                              size_tracker, params.pbs_type,
+                              allocate_gpu_memory);
     multi_gpu_alloc_lwe_async(active_streams, lwe_after_ks_vec,
                               num_radix_blocks, params.small_lwe_dimension + 1,
-                              size_tracker, allocate_gpu_memory);
+                              size_tracker, params.pbs_type,
+                              allocate_gpu_memory);
     if (num_many_lut > 1) {
       multi_gpu_alloc_lwe_many_lut_output_async(
           active_streams, lwe_after_pbs_vec, num_radix_blocks, num_many_lut,
-          params.big_lwe_dimension + 1, size_tracker, allocate_gpu_memory);
+          params.big_lwe_dimension + 1, size_tracker, params.pbs_type,
+          allocate_gpu_memory);
     } else {
       multi_gpu_alloc_lwe_async(active_streams, lwe_after_pbs_vec,
                                 num_radix_blocks, params.big_lwe_dimension + 1,
-                                size_tracker, allocate_gpu_memory);
+                                size_tracker, params.pbs_type,
+                                allocate_gpu_memory);
     }
     multi_gpu_alloc_array_async(active_streams, lwe_trivial_indexes_vec,
                                 num_radix_blocks, size_tracker,
@@ -451,12 +460,14 @@ struct int_radix_lut_custom_input_output {
   }
 
   void setup_gemm_batch_ks_temp_buffers(uint64_t &size_tracker) {
+    int threshold = (params.pbs_type == PBS_TYPE::MULTI_BIT)
+                        ? THRESHOLD_MULTI_GPU_WITH_MULTI_BIT_PARAMS
+                        : THRESHOLD_MULTI_GPU_WITH_CLASSICAL_PARAMS;
 
-    auto inputs_on_gpu =
-        std::min((int)num_input_blocks,
-                 std::max(THRESHOLD_MULTI_GPU,
-                          get_num_inputs_on_gpu(num_input_blocks, 0,
-                                                active_streams.count())));
+    auto inputs_on_gpu = std::min(
+        (int)num_input_blocks,
+        std::max(threshold, get_num_inputs_on_gpu(num_input_blocks, 0,
+                                                  active_streams.count())));
 
     if (inputs_on_gpu >= get_threshold_ks_gemm()) {
       for (auto i = 0; i < active_streams.count(); ++i) {
@@ -798,16 +809,20 @@ struct int_radix_lut_custom_input_output {
   void allocate_lwe_vector_for_non_trivial_indexes(
       CudaStreams streams, uint64_t max_num_radix_blocks,
       uint64_t &size_tracker, bool allocate_gpu_memory) {
+
+    int threshold = (params.pbs_type == PBS_TYPE::MULTI_BIT)
+                        ? THRESHOLD_MULTI_GPU_WITH_MULTI_BIT_PARAMS
+                        : THRESHOLD_MULTI_GPU_WITH_CLASSICAL_PARAMS;
+
     // We need to create the auxiliary array only in GPU 0
     if (active_streams.count() > 1) {
       lwe_aligned_vec.resize(active_streams.count());
       for (uint i = 0; i < active_streams.count(); i++) {
         uint64_t size_tracker_on_array_i = 0;
-        auto inputs_on_gpu =
-            std::min((int)max_num_radix_blocks,
-                     std::max(THRESHOLD_MULTI_GPU,
-                              get_num_inputs_on_gpu(max_num_radix_blocks, i,
-                                                    active_streams.count())));
+        auto inputs_on_gpu = std::min(
+            (int)max_num_radix_blocks,
+            std::max(threshold, get_num_inputs_on_gpu(max_num_radix_blocks, i,
+                                                      active_streams.count())));
         InputTorus *d_array =
             (InputTorus *)cuda_malloc_with_size_tracking_async(
                 inputs_on_gpu * (params.big_lwe_dimension + 1) *
@@ -998,8 +1013,8 @@ template <typename Torus> struct int_bit_extract_luts_buffer {
         num_radix_blocks * bits_per_block * sizeof(Torus), streams.stream(0),
         streams.gpu_index(0), allocate_gpu_memory);
 
-    auto active_streams =
-        streams.active_gpu_subset(bits_per_block * num_radix_blocks);
+    auto active_streams = streams.active_gpu_subset(
+        bits_per_block * num_radix_blocks, params.pbs_type);
     lut->broadcast_lut(active_streams);
 
     /**
@@ -1266,7 +1281,8 @@ template <typename Torus> struct int_sum_ciphertexts_vec_memory {
           luts_message_carry->get_max_degree(1), params.glwe_dimension,
           params.polynomial_size, message_modulus, params.carry_modulus,
           lut_f_carry, gpu_memory_allocated);
-      auto active_gpu_count_mc = streams.active_gpu_subset(pbs_count);
+      auto active_gpu_count_mc =
+          streams.active_gpu_subset(pbs_count, params.pbs_type);
       luts_message_carry->broadcast_lut(active_gpu_count_mc);
     }
   }
@@ -1436,7 +1452,8 @@ template <typename Torus> struct int_seq_group_prop_memory {
     cuda_memcpy_with_size_tracking_async_to_gpu(
         seq_lut_indexes, h_seq_lut_indexes, num_seq_luts * sizeof(Torus),
         streams.stream(0), streams.gpu_index(0), allocate_gpu_memory);
-    auto active_streams = streams.active_gpu_subset(num_seq_luts);
+    auto active_streams =
+        streams.active_gpu_subset(num_seq_luts, params.pbs_type);
     lut_sequential_algorithm->broadcast_lut(active_streams);
     free(h_seq_lut_indexes);
   };
@@ -1490,7 +1507,8 @@ template <typename Torus> struct int_hs_group_prop_memory {
         lut_hillis_steele->get_max_degree(0), glwe_dimension, polynomial_size,
         message_modulus, carry_modulus, f_lut_hillis_steele,
         gpu_memory_allocated);
-    auto active_streams = streams.active_gpu_subset(num_groups);
+    auto active_streams =
+        streams.active_gpu_subset(num_groups, params.pbs_type);
     lut_hillis_steele->broadcast_lut(active_streams);
   };
   void release(CudaStreams streams) {
@@ -1667,7 +1685,8 @@ template <typename Torus> struct int_shifted_blocks_and_states_memory {
         lut_indexes, h_lut_indexes, lut_indexes_size, streams.stream(0),
         streams.gpu_index(0), allocate_gpu_memory);
     // Do I need to do something else for the multi-gpu?
-    auto active_streams = streams.active_gpu_subset(num_radix_blocks);
+    auto active_streams =
+        streams.active_gpu_subset(num_radix_blocks, params.pbs_type);
     luts_array_first_step->broadcast_lut(active_streams);
   };
   void release(CudaStreams streams) {
@@ -1932,7 +1951,8 @@ template <typename Torus> struct int_prop_simu_group_carries_memory {
         scalar_array_cum_sum, h_scalar_array_cum_sum,
         num_radix_blocks * sizeof(Torus), streams.stream(0),
         streams.gpu_index(0), allocate_gpu_memory);
-    auto active_streams = streams.active_gpu_subset(num_radix_blocks);
+    auto active_streams =
+        streams.active_gpu_subset(num_radix_blocks, params.pbs_type);
     luts_array_second_step->broadcast_lut(active_streams);
 
     if (use_sequential_algorithm_to_resolve_group_carries) {
@@ -1957,7 +1977,8 @@ template <typename Torus> struct int_prop_simu_group_carries_memory {
     cuda_memcpy_with_size_tracking_async_gpu_to_gpu(
         lut_indexes, new_lut_indexes, new_num_blocks * sizeof(Torus),
         streams.stream(0), streams.gpu_index(0), gpu_memory_allocated);
-    auto new_active_streams = streams.active_gpu_subset(new_num_blocks);
+    auto new_active_streams = streams.active_gpu_subset(
+        new_num_blocks, luts_array_second_step->params.pbs_type);
     // We just need to update the lut indexes so we use false here
     luts_array_second_step->broadcast_lut(new_active_streams, false);
 
@@ -2124,7 +2145,7 @@ template <typename Torus> struct int_sc_prop_memory {
           polynomial_size, message_modulus, carry_modulus, f_overflow_fp,
           gpu_memory_allocated);
 
-      auto active_streams = streams.active_gpu_subset(1);
+      auto active_streams = streams.active_gpu_subset(1, params.pbs_type);
       lut_overflow_flag_prep->broadcast_lut(active_streams);
     }
 
@@ -2196,7 +2217,8 @@ template <typename Torus> struct int_sc_prop_memory {
           (num_radix_blocks + 1) * sizeof(Torus), streams.stream(0),
           streams.gpu_index(0), allocate_gpu_memory);
     }
-    auto active_streams = streams.active_gpu_subset(num_radix_blocks + 1);
+    auto active_streams =
+        streams.active_gpu_subset(num_radix_blocks + 1, params.pbs_type);
     lut_message_extract->broadcast_lut(active_streams);
   };
 
@@ -2393,7 +2415,8 @@ template <typename Torus> struct int_shifted_blocks_and_borrow_states_memory {
         lut_indexes, h_lut_indexes, lut_indexes_size, streams.stream(0),
         streams.gpu_index(0), allocate_gpu_memory);
     // Do I need to do something else for the multi-gpu?
-    auto active_streams = streams.active_gpu_subset(num_radix_blocks);
+    auto active_streams =
+        streams.active_gpu_subset(num_radix_blocks, params.pbs_type);
     luts_array_first_step->broadcast_lut(active_streams);
   };
 
@@ -2404,7 +2427,8 @@ template <typename Torus> struct int_shifted_blocks_and_borrow_states_memory {
     cuda_memcpy_with_size_tracking_async_gpu_to_gpu(
         lut_indexes, new_lut_indexes, new_num_blocks * sizeof(Torus),
         streams.stream(0), streams.gpu_index(0), gpu_memory_allocated);
-    auto new_active_streams = streams.active_gpu_subset(new_num_blocks);
+    auto new_active_streams = streams.active_gpu_subset(
+        new_num_blocks, luts_array_first_step->params.pbs_type);
     // We just need to update the lut indexes so we use false here
     luts_array_first_step->broadcast_lut(new_active_streams, false);
   }
@@ -2499,7 +2523,8 @@ template <typename Torus> struct int_borrow_prop_memory {
         lut_message_extract->get_max_degree(0), glwe_dimension, polynomial_size,
         message_modulus, carry_modulus, f_message_extract,
         gpu_memory_allocated);
-    active_streams = streams.active_gpu_subset(num_radix_blocks);
+    active_streams =
+        streams.active_gpu_subset(num_radix_blocks, params.pbs_type);
 
     lut_message_extract->broadcast_lut(active_streams);
 
@@ -2520,7 +2545,8 @@ template <typename Torus> struct int_borrow_prop_memory {
       lut_borrow_flag->broadcast_lut(active_streams);
     }
 
-    active_streams = streams.active_gpu_subset(num_radix_blocks);
+    active_streams =
+        streams.active_gpu_subset(num_radix_blocks, params.pbs_type);
     internal_streams.create_internal_cuda_streams_on_same_gpus(active_streams,
                                                                2);
   };

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

@@ -45,7 +45,8 @@ template <typename Torus> struct int_mul_memory {
           params.polynomial_size, params.message_modulus, params.carry_modulus,
           zero_out_predicate_lut_f, gpu_memory_allocated);
 
-      auto active_streams = streams.active_gpu_subset(num_radix_blocks);
+      auto active_streams =
+          streams.active_gpu_subset(num_radix_blocks, params.pbs_type);
       zero_out_predicate_lut->broadcast_lut(active_streams);
 
       zero_out_mem = new int_zero_out_if_buffer<Torus>(
@@ -122,7 +123,8 @@ template <typename Torus> struct int_mul_memory {
           streams.stream(0), streams.gpu_index(0),
           luts_array->get_lut_indexes(0, lsb_vector_block_count), 1,
           msb_vector_block_count);
-    auto active_streams = streams.active_gpu_subset(total_block_count);
+    auto active_streams =
+        streams.active_gpu_subset(total_block_count, params.pbs_type);
     luts_array->broadcast_lut(active_streams);
     // create memory object for sum ciphertexts
     sum_ciphertexts_mem = new int_sum_ciphertexts_vec_memory<Torus>(

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

@@ -126,7 +126,8 @@ template <typename Torus> struct int_grouped_oprf_memory {
         luts->get_lut_indexes(0, 0), this->h_lut_indexes,
         num_blocks_to_process * sizeof(Torus), streams.stream(0),
         streams.gpu_index(0), allocate_gpu_memory);
-    auto active_streams = streams.active_gpu_subset(num_blocks_to_process);
+    auto active_streams =
+        streams.active_gpu_subset(num_blocks_to_process, params.pbs_type);
     luts->broadcast_lut(active_streams);
 
     cuda_synchronize_stream(streams.stream(0), streams.gpu_index(0));

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

@@ -91,7 +91,8 @@ template <typename Torus> struct int_logical_scalar_shift_buffer {
           cur_lut_bivariate->get_max_degree(0), params.glwe_dimension,
           params.polynomial_size, params.message_modulus, params.carry_modulus,
           shift_lut_f, gpu_memory_allocated);
-      auto active_streams = streams.active_gpu_subset(num_radix_blocks);
+      auto active_streams =
+          streams.active_gpu_subset(num_radix_blocks, params.pbs_type);
       cur_lut_bivariate->broadcast_lut(active_streams);
 
       lut_buffers_bivariate.push_back(cur_lut_bivariate);
@@ -177,7 +178,8 @@ template <typename Torus> struct int_logical_scalar_shift_buffer {
           cur_lut_bivariate->get_max_degree(0), params.glwe_dimension,
           params.polynomial_size, params.message_modulus, params.carry_modulus,
           shift_lut_f, gpu_memory_allocated);
-      auto active_streams = streams.active_gpu_subset(num_radix_blocks);
+      auto active_streams =
+          streams.active_gpu_subset(num_radix_blocks, params.pbs_type);
       cur_lut_bivariate->broadcast_lut(active_streams);
 
       lut_buffers_bivariate.push_back(cur_lut_bivariate);
@@ -220,7 +222,7 @@ template <typename Torus> struct int_arithmetic_scalar_shift_buffer {
                                      uint64_t &size_tracker) {
     gpu_memory_allocated = allocate_gpu_memory;
 
-    auto active_streams = streams.active_gpu_subset(1);
+    auto active_streams = streams.active_gpu_subset(1, params.pbs_type);
     // In the arithmetic shift, a PBS has to be applied to the last rotated
     // block twice: once to shift it, once to compute the padding block to be
     // copied onto all blocks to the left of the last rotated block
@@ -276,7 +278,8 @@ template <typename Torus> struct int_arithmetic_scalar_shift_buffer {
           shift_last_block_lut_univariate->get_max_degree(0),
           params.glwe_dimension, params.polynomial_size, params.message_modulus,
           params.carry_modulus, last_block_lut_f, gpu_memory_allocated);
-      auto active_streams_shift_last = streams.active_gpu_subset(1);
+      auto active_streams_shift_last =
+          streams.active_gpu_subset(1, params.pbs_type);
       shift_last_block_lut_univariate->broadcast_lut(active_streams_shift_last);
 
       lut_buffers_univariate.push_back(shift_last_block_lut_univariate);
@@ -302,7 +305,7 @@ template <typename Torus> struct int_arithmetic_scalar_shift_buffer {
         padding_block_lut_univariate->get_max_degree(0), params.glwe_dimension,
         params.polynomial_size, params.message_modulus, params.carry_modulus,
         padding_block_lut_f, gpu_memory_allocated);
-    // auto active_streams = streams.active_gpu_subset(1);
+    // auto active_streams = streams.active_gpu_subset(1, params.pbs_type);
     padding_block_lut_univariate->broadcast_lut(active_streams);
 
     lut_buffers_univariate.push_back(padding_block_lut_univariate);
@@ -344,7 +347,7 @@ template <typename Torus> struct int_arithmetic_scalar_shift_buffer {
           params.polynomial_size, params.message_modulus, params.carry_modulus,
           blocks_lut_f, gpu_memory_allocated);
       auto active_streams_shift_blocks =
-          streams.active_gpu_subset(num_radix_blocks);
+          streams.active_gpu_subset(num_radix_blocks, params.pbs_type);
       shift_blocks_lut_bivariate->broadcast_lut(active_streams_shift_blocks);
 
       lut_buffers_bivariate.push_back(shift_blocks_lut_bivariate);

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

@@ -119,8 +119,8 @@ template <typename Torus> struct int_shift_and_rotate_buffer {
         mux_lut->get_degree(0), mux_lut->get_max_degree(0),
         params.glwe_dimension, params.polynomial_size, params.message_modulus,
         params.carry_modulus, mux_lut_f, gpu_memory_allocated);
-    auto active_gpu_count_mux =
-        streams.active_gpu_subset(bits_per_block * num_radix_blocks);
+    auto active_gpu_count_mux = streams.active_gpu_subset(
+        bits_per_block * num_radix_blocks, params.pbs_type);
     mux_lut->broadcast_lut(active_gpu_count_mux);
 
     auto cleaning_lut_f = [params](Torus x) -> Torus {
@@ -132,7 +132,7 @@ template <typename Torus> struct int_shift_and_rotate_buffer {
         params.glwe_dimension, params.polynomial_size, params.message_modulus,
         params.carry_modulus, cleaning_lut_f, gpu_memory_allocated);
     auto active_gpu_count_cleaning =
-        streams.active_gpu_subset(num_radix_blocks);
+        streams.active_gpu_subset(num_radix_blocks, params.pbs_type);
     cleaning_lut->broadcast_lut(active_gpu_count_cleaning);
   }
 

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

@@ -108,7 +108,8 @@ template <typename Torus> struct int_overflowing_sub_memory {
         glwe_dimension, polynomial_size, message_modulus, carry_modulus,
         f_message_acc, gpu_memory_allocated);
 
-    auto active_streams = streams.active_gpu_subset(num_radix_blocks);
+    auto active_streams =
+        streams.active_gpu_subset(num_radix_blocks, params.pbs_type);
     luts_array->broadcast_lut(active_streams);
     luts_borrow_propagation_sum->broadcast_lut(active_streams);
     message_acc->broadcast_lut(active_streams);

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

@@ -38,7 +38,8 @@ template <typename Torus> struct int_unchecked_all_eq_slices_buffer {
       num_streams_to_use = 1;
 
     this->num_streams = num_streams_to_use;
-    this->active_streams = streams.active_gpu_subset(num_blocks);
+    this->active_streams =
+        streams.active_gpu_subset(num_blocks, params.pbs_type);
 
     uint32_t num_gpus = active_streams.count();
 

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

@@ -40,7 +40,8 @@ template <typename Torus> struct int_equality_selectors_buffer {
 
     this->num_streams = num_streams_to_use;
 
-    this->active_streams = streams.active_gpu_subset(num_blocks);
+    this->active_streams =
+        streams.active_gpu_subset(num_blocks, params.pbs_type);
 
     this->internal_cuda_streams.create_internal_cuda_streams_on_same_gpus(
         active_streams, num_streams_to_use);
@@ -154,7 +155,8 @@ template <typename Torus> struct int_possible_results_buffer {
 
     this->num_streams = num_streams_to_use;
 
-    this->active_streams = streams.active_gpu_subset(num_blocks);
+    this->active_streams =
+        streams.active_gpu_subset(num_blocks, params.pbs_type);
 
     this->internal_cuda_streams.create_internal_cuda_streams_on_same_gpus(
         active_streams, num_streams_to_use);
@@ -207,7 +209,8 @@ template <typename Torus> struct int_possible_results_buffer {
             params.message_modulus, params.carry_modulus, fns,
             allocate_gpu_memory);
 
-        current_lut->broadcast_lut(streams.active_gpu_subset(1));
+        current_lut->broadcast_lut(
+            streams.active_gpu_subset(1, params.pbs_type));
         stream_luts[lut_count++] = current_lut;
         lut_value_start += luts_in_this_call;
       }
@@ -282,7 +285,8 @@ template <typename Torus> struct int_aggregate_one_hot_buffer {
 
     this->num_streams = num_streams_to_use;
 
-    this->active_streams = streams.active_gpu_subset(num_blocks);
+    this->active_streams =
+        streams.active_gpu_subset(num_blocks, params.pbs_type);
 
     this->internal_cuda_streams.create_internal_cuda_streams_on_same_gpus(
         active_streams, num_streams);
@@ -300,7 +304,8 @@ template <typename Torus> struct int_aggregate_one_hot_buffer {
           params.polynomial_size, params.message_modulus, params.carry_modulus,
           id_fn, allocate_gpu_memory);
 
-      lut->broadcast_lut(streams.active_gpu_subset(num_blocks));
+      lut->broadcast_lut(
+          streams.active_gpu_subset(num_blocks, params.pbs_type));
       this->stream_identity_luts[i] = lut;
     }
 
@@ -321,7 +326,7 @@ template <typename Torus> struct int_aggregate_one_hot_buffer {
         params.polynomial_size, params.message_modulus, params.carry_modulus,
         msg_fn, allocate_gpu_memory);
     this->message_extract_lut->broadcast_lut(
-        streams.active_gpu_subset(num_blocks));
+        streams.active_gpu_subset(num_blocks, params.pbs_type));
 
     this->carry_extract_lut = new int_radix_lut<Torus>(
         streams, params, 1, num_blocks, allocate_gpu_memory, size_tracker);
@@ -333,7 +338,7 @@ template <typename Torus> struct int_aggregate_one_hot_buffer {
         params.polynomial_size, params.message_modulus, params.carry_modulus,
         carry_fn, allocate_gpu_memory);
     this->carry_extract_lut->broadcast_lut(
-        streams.active_gpu_subset(num_blocks));
+        streams.active_gpu_subset(num_blocks, params.pbs_type));
 
     this->partial_aggregated_vectors =
         new CudaRadixCiphertextFFI *[num_streams];
@@ -628,7 +633,8 @@ template <typename Torus> struct int_unchecked_contains_buffer {
       num_streams_to_use = 1;
 
     this->num_streams = num_streams_to_use;
-    this->active_streams = streams.active_gpu_subset(num_blocks);
+    this->active_streams =
+        streams.active_gpu_subset(num_blocks, params.pbs_type);
 
     this->internal_cuda_streams.create_internal_cuda_streams_on_same_gpus(
         active_streams, num_streams_to_use);
@@ -703,7 +709,8 @@ template <typename Torus> struct int_unchecked_contains_clear_buffer {
       num_streams_to_use = 1;
 
     this->num_streams = num_streams_to_use;
-    this->active_streams = streams.active_gpu_subset(num_blocks);
+    this->active_streams =
+        streams.active_gpu_subset(num_blocks, params.pbs_type);
 
     this->internal_cuda_streams.create_internal_cuda_streams_on_same_gpus(
         active_streams, num_streams_to_use);
@@ -1094,7 +1101,8 @@ template <typename Torus> struct int_unchecked_first_index_of_clear_buffer {
       num_streams_to_use = 1;
 
     this->num_streams = num_streams_to_use;
-    this->active_streams = streams.active_gpu_subset(num_blocks);
+    this->active_streams =
+        streams.active_gpu_subset(num_blocks, params.pbs_type);
 
     this->internal_cuda_streams.create_internal_cuda_streams_on_same_gpus(
         active_streams, num_streams_to_use);
@@ -1184,7 +1192,8 @@ template <typename Torus> struct int_unchecked_first_index_of_clear_buffer {
         this->prefix_sum_lut->get_max_degree(0), params.glwe_dimension,
         params.polynomial_size, params.message_modulus, params.carry_modulus,
         prefix_sum_fn, allocate_gpu_memory);
-    this->prefix_sum_lut->broadcast_lut(streams.active_gpu_subset(num_inputs));
+    this->prefix_sum_lut->broadcast_lut(
+        streams.active_gpu_subset(num_inputs, params.pbs_type));
 
     auto cleanup_fn = [ALREADY_SEEN, params](Torus x) -> Torus {
       Torus val = x % params.message_modulus;
@@ -1200,7 +1209,8 @@ template <typename Torus> struct int_unchecked_first_index_of_clear_buffer {
         this->cleanup_lut->get_max_degree(0), params.glwe_dimension,
         params.polynomial_size, params.message_modulus, params.carry_modulus,
         cleanup_fn, allocate_gpu_memory);
-    this->cleanup_lut->broadcast_lut(streams.active_gpu_subset(num_inputs));
+    this->cleanup_lut->broadcast_lut(
+        streams.active_gpu_subset(num_inputs, params.pbs_type));
   }
 
   void release(CudaStreams streams) {
@@ -1292,7 +1302,8 @@ template <typename Torus> struct int_unchecked_first_index_of_buffer {
       num_streams_to_use = 1;
 
     this->num_streams = num_streams_to_use;
-    this->active_streams = streams.active_gpu_subset(num_blocks);
+    this->active_streams =
+        streams.active_gpu_subset(num_blocks, params.pbs_type);
 
     this->internal_cuda_streams.create_internal_cuda_streams_on_same_gpus(
         active_streams, num_streams_to_use);
@@ -1372,7 +1383,8 @@ template <typename Torus> struct int_unchecked_first_index_of_buffer {
         this->prefix_sum_lut->get_max_degree(0), params.glwe_dimension,
         params.polynomial_size, params.message_modulus, params.carry_modulus,
         prefix_sum_fn, allocate_gpu_memory);
-    this->prefix_sum_lut->broadcast_lut(streams.active_gpu_subset(num_inputs));
+    this->prefix_sum_lut->broadcast_lut(
+        streams.active_gpu_subset(num_inputs, params.pbs_type));
 
     auto cleanup_fn = [ALREADY_SEEN, params](Torus x) -> Torus {
       Torus val = x % params.message_modulus;
@@ -1388,7 +1400,8 @@ template <typename Torus> struct int_unchecked_first_index_of_buffer {
         this->cleanup_lut->get_max_degree(0), params.glwe_dimension,
         params.polynomial_size, params.message_modulus, params.carry_modulus,
         cleanup_fn, allocate_gpu_memory);
-    this->cleanup_lut->broadcast_lut(streams.active_gpu_subset(num_inputs));
+    this->cleanup_lut->broadcast_lut(
+        streams.active_gpu_subset(num_inputs, params.pbs_type));
   }
 
   void release(CudaStreams streams) {
@@ -1462,7 +1475,8 @@ template <typename Torus> struct int_unchecked_index_of_buffer {
       num_streams_to_use = 1;
 
     this->num_streams = num_streams_to_use;
-    this->active_streams = streams.active_gpu_subset(num_blocks);
+    this->active_streams =
+        streams.active_gpu_subset(num_blocks, params.pbs_type);
 
     this->internal_cuda_streams.create_internal_cuda_streams_on_same_gpus(
         active_streams, num_streams_to_use);
@@ -1523,7 +1537,8 @@ template <typename Torus> struct int_unchecked_index_of_clear_buffer {
       num_streams_to_use = 1;
 
     this->num_streams = num_streams_to_use;
-    this->active_streams = streams.active_gpu_subset(num_blocks);
+    this->active_streams =
+        streams.active_gpu_subset(num_blocks, params.pbs_type);
 
     this->internal_cuda_streams.create_internal_cuda_streams_on_same_gpus(
         active_streams, num_streams_to_use);

backends/tfhe-cuda-backend/cuda/include/zk/zk_utilities.h

@@ -47,7 +47,7 @@ template <typename Torus> struct compact_lwe_list {
 
 template <typename Torus> struct flattened_compact_lwe_lists {
   Torus *d_ptr;
-  Torus **d_ptr_to_compact_list;
+  Torus **ptr_array_to_d_compact_list;
   const uint32_t *h_num_lwes_per_compact_list;
   uint32_t num_compact_lists;
   uint32_t lwe_dimension;
@@ -59,13 +59,13 @@ template <typename Torus> struct flattened_compact_lwe_lists {
                               uint32_t lwe_dimension)
       : d_ptr(d_ptr), h_num_lwes_per_compact_list(h_num_lwes_per_compact_list),
         num_compact_lists(num_compact_lists), lwe_dimension(lwe_dimension) {
-    d_ptr_to_compact_list =
-        static_cast<Torus **>(malloc(num_compact_lists * sizeof(Torus **)));
+    ptr_array_to_d_compact_list =
+        static_cast<Torus **>(malloc(num_compact_lists * sizeof(Torus *)));
     total_num_lwes = 0;
     auto curr_list = d_ptr;
     for (auto i = 0; i < num_compact_lists; ++i) {
       total_num_lwes += h_num_lwes_per_compact_list[i];
-      d_ptr_to_compact_list[i] = curr_list;
+      ptr_array_to_d_compact_list[i] = curr_list;
       curr_list += lwe_dimension + h_num_lwes_per_compact_list[i];
     }
   }
@@ -75,10 +75,12 @@ template <typename Torus> struct flattened_compact_lwe_lists {
       PANIC("index out of range in flattened_compact_lwe_lists::get");
     }
 
-    return compact_lwe_list(d_ptr_to_compact_list[compact_list_index],
+    return compact_lwe_list(ptr_array_to_d_compact_list[compact_list_index],
                             lwe_dimension,
                             h_num_lwes_per_compact_list[compact_list_index]);
   }
+
+  void release() { free(ptr_array_to_d_compact_list); }
 };
 
 /*
@@ -121,7 +123,6 @@ template <typename Torus> struct zk_expand_mem {
       : computing_params(computing_params), casting_params(casting_params),
         num_compact_lists(num_compact_lists),
         casting_key_type(casting_key_type) {
-
     gpu_memory_allocated = allocate_gpu_memory;
 
     // We copy num_lwes_per_compact_list so we get protection against
@@ -289,7 +290,8 @@ template <typename Torus> struct zk_expand_mem {
         lut_indexes, h_lut_indexes, num_packed_msgs * num_lwes * sizeof(Torus),
         streams.stream(0), streams.gpu_index(0), allocate_gpu_memory);
 
-    auto active_streams = streams.active_gpu_subset(2 * num_lwes);
+    auto active_streams =
+        streams.active_gpu_subset(2 * num_lwes, params.pbs_type);
     message_and_carry_extract_luts->broadcast_lut(active_streams);
 
     message_and_carry_extract_luts->allocate_lwe_vector_for_non_trivial_indexes(
@@ -313,7 +315,6 @@ template <typename Torus> struct zk_expand_mem {
   }
 
   void release(CudaStreams streams) {
-
     message_and_carry_extract_luts->release(streams);
     delete message_and_carry_extract_luts;
 

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

@@ -153,7 +153,8 @@ __host__ void are_all_comparisons_block_true(
         cuda_memcpy_async_to_gpu(is_max_value_lut->get_lut_indexes(0, 0),
                                  h_lut_indexes, num_chunks * sizeof(Torus),
                                  streams.stream(0), streams.gpu_index(0));
-        auto active_streams = streams.active_gpu_subset(num_chunks);
+        auto active_streams =
+            streams.active_gpu_subset(num_chunks, params.pbs_type);
         is_max_value_lut->broadcast_lut(active_streams);
       }
       lut = is_max_value_lut;
@@ -172,8 +173,8 @@ __host__ void are_all_comparisons_block_true(
                                is_max_value_lut->h_lut_indexes,
                                is_max_value_lut->num_blocks * sizeof(Torus),
                                streams.stream(0), streams.gpu_index(0));
-      auto active_gpu_count_is_max =
-          streams.active_gpu_subset(is_max_value_lut->num_blocks);
+      auto active_gpu_count_is_max = streams.active_gpu_subset(
+          is_max_value_lut->num_blocks, params.pbs_type);
       is_max_value_lut->broadcast_lut(active_gpu_count_is_max, false);
 
       reset_radix_ciphertext_blocks(lwe_array_out, 1);
@@ -488,7 +489,7 @@ tree_sign_reduction(CudaStreams streams, CudaRadixCiphertextFFI *lwe_array_out,
       polynomial_size, message_modulus, carry_modulus, f, true,
       tree_buffer->preallocated_h_lut);
 
-  auto active_streams = streams.active_gpu_subset(1);
+  auto active_streams = streams.active_gpu_subset(1, params.pbs_type);
   last_lut->broadcast_lut(active_streams);
 
   // Last leaf

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

@@ -155,3 +155,24 @@ void cleanup_cuda_integer_decompress_radix_ciphertext_128(
   delete mem_ptr;
   *mem_ptr_void = nullptr;
 }
+
+void cuda_integer_extract_glwe_128(
+    CudaStreamsFFI streams, void *glwe_array_out,
+    CudaPackedGlweCiphertextListFFI const *glwe_list,
+    uint32_t const glwe_index) {
+
+  CudaStreams _streams = CudaStreams(streams);
+  host_extract<__uint128_t>(_streams.stream(0), _streams.gpu_index(0),
+                            (__uint128_t *)glwe_array_out, glwe_list,
+                            glwe_index);
+}
+
+void cuda_integer_extract_glwe_64(
+    CudaStreamsFFI streams, void *glwe_array_out,
+    CudaPackedGlweCiphertextListFFI const *glwe_list,
+    uint32_t const glwe_index) {
+
+  CudaStreams _streams = CudaStreams(streams);
+  host_extract<__uint64_t>(_streams.stream(0), _streams.gpu_index(0),
+                           (__uint64_t *)glwe_array_out, glwe_list, glwe_index);
+}

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

@@ -339,7 +339,9 @@ host_integer_decompress(CudaStreams streams,
     /// dimension to a big LWE dimension
     auto encryption_params = h_mem_ptr->encryption_params;
     auto lut = h_mem_ptr->decompression_rescale_lut;
-    auto active_streams = streams.active_gpu_subset(num_blocks_to_decompress);
+    auto active_streams = streams.active_gpu_subset(
+        num_blocks_to_decompress,
+        h_mem_ptr->decompression_rescale_lut->params.pbs_type);
     if (active_streams.count() == 1) {
       execute_pbs_async<Torus, Torus>(
           active_streams, (Torus *)d_lwe_array_out->ptr, lut->lwe_indexes_out,

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

@@ -542,7 +542,8 @@ __host__ void integer_radix_apply_univariate_lookup_table(
   std::vector<Torus *> lwe_after_pbs_vec = lut->lwe_after_pbs_vec;
   std::vector<Torus *> lwe_trivial_indexes_vec = lut->lwe_trivial_indexes_vec;
 
-  auto active_streams = streams.active_gpu_subset(num_radix_blocks);
+  auto active_streams =
+      streams.active_gpu_subset(num_radix_blocks, params.pbs_type);
   if (active_streams.count() == 1) {
     execute_keyswitch_async<Torus>(
         streams.get_ith(0), lwe_after_ks_vec[0], lwe_trivial_indexes_vec[0],
@@ -645,7 +646,8 @@ __host__ void integer_radix_apply_many_univariate_lookup_table(
   std::vector<Torus *> lwe_after_pbs_vec = lut->lwe_after_pbs_vec;
   std::vector<Torus *> lwe_trivial_indexes_vec = lut->lwe_trivial_indexes_vec;
 
-  auto active_streams = streams.active_gpu_subset(num_radix_blocks);
+  auto active_streams =
+      streams.active_gpu_subset(num_radix_blocks, params.pbs_type);
   if (active_streams.count() == 1) {
     execute_keyswitch_async<Torus>(
         streams.get_ith(0), lwe_after_ks_vec[0], lwe_trivial_indexes_vec[0],
@@ -764,7 +766,8 @@ __host__ void integer_radix_apply_bivariate_lookup_table(
   std::vector<Torus *> lwe_after_pbs_vec = lut->lwe_after_pbs_vec;
   std::vector<Torus *> lwe_trivial_indexes_vec = lut->lwe_trivial_indexes_vec;
 
-  auto active_streams = streams.active_gpu_subset(num_radix_blocks);
+  auto active_streams =
+      streams.active_gpu_subset(num_radix_blocks, params.pbs_type);
   if (active_streams.count() == 1) {
     execute_keyswitch_async<Torus>(
         streams.get_ith(0), lwe_after_ks_vec[0], lwe_trivial_indexes_vec[0],
@@ -1812,7 +1815,8 @@ uint64_t scratch_cuda_apply_univariate_lut(
       (params.glwe_dimension + 1) * params.polynomial_size * sizeof(Torus),
       streams.stream(0), streams.gpu_index(0), allocate_gpu_memory);
   *(*mem_ptr)->get_degree(0) = lut_degree;
-  auto active_streams = streams.active_gpu_subset(num_radix_blocks);
+  auto active_streams =
+      streams.active_gpu_subset(num_radix_blocks, params.pbs_type);
   (*mem_ptr)->broadcast_lut(active_streams);
   POP_RANGE()
   return size_tracker;
@@ -1847,7 +1851,8 @@ uint64_t scratch_cuda_apply_many_univariate_lut(
       (params.glwe_dimension + 1) * params.polynomial_size * sizeof(Torus),
       streams.stream(0), streams.gpu_index(0), allocate_gpu_memory);
   *(*mem_ptr)->get_degree(0) = lut_degree;
-  auto active_streams = streams.active_gpu_subset(num_radix_blocks);
+  auto active_streams =
+      streams.active_gpu_subset(num_radix_blocks, params.pbs_type);
   (*mem_ptr)->broadcast_lut(active_streams);
   POP_RANGE()
   return size_tracker;
@@ -1883,7 +1888,8 @@ uint64_t scratch_cuda_apply_bivariate_lut(
       (params.glwe_dimension + 1) * params.polynomial_size * sizeof(Torus),
       streams.stream(0), streams.gpu_index(0), allocate_gpu_memory);
   *(*mem_ptr)->get_degree(0) = lut_degree;
-  auto active_streams = streams.active_gpu_subset(num_radix_blocks);
+  auto active_streams =
+      streams.active_gpu_subset(num_radix_blocks, params.pbs_type);
   (*mem_ptr)->broadcast_lut(active_streams);
   POP_RANGE()
   return size_tracker;
@@ -2336,8 +2342,8 @@ integer_radix_apply_noise_squashing(CudaStreams streams,
 
   // Since the radix ciphertexts are packed, we have to use the num_radix_blocks
   // from the output ct
-  auto active_streams =
-      streams.active_gpu_subset(lwe_array_out->num_radix_blocks);
+  auto active_streams = streams.active_gpu_subset(
+      lwe_array_out->num_radix_blocks, params.pbs_type);
   if (active_streams.count() == 1) {
     execute_keyswitch_async<InputTorus>(
         streams.get_ith(0), lwe_after_ks_vec[0], lwe_trivial_indexes_vec[0],

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

@@ -388,7 +388,8 @@ __host__ void host_integer_partial_sum_ciphertexts_vec(
     current_columns.next_accumulation(total_ciphertexts, total_messages,
                                       needs_processing);
 
-    auto active_streams = streams.active_gpu_subset(total_ciphertexts);
+    auto active_streams =
+        streams.active_gpu_subset(total_ciphertexts, mem_ptr->params.pbs_type);
     GPU_ASSERT(total_ciphertexts <= mem_ptr->luts_message_carry->num_blocks,
                "SUM CT");
 
@@ -442,7 +443,8 @@ __host__ void host_integer_partial_sum_ciphertexts_vec(
         streams.stream(0), streams.gpu_index(0), current_blocks,
         num_radix_blocks, num_radix_blocks + 1);
 
-    auto active_streams = streams.active_gpu_subset(2 * num_radix_blocks);
+    auto active_streams = streams.active_gpu_subset(2 * num_radix_blocks,
+                                                    mem_ptr->params.pbs_type);
 
     if (active_streams.count() == 1) {
       execute_keyswitch_async<Torus>(

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

@@ -29,7 +29,8 @@ void host_integer_grouped_oprf(CudaStreams streams,
                                int_grouped_oprf_memory<Torus> *mem_ptr,
                                void *const *bsks) {
 
-  auto active_streams = streams.active_gpu_subset(num_blocks_to_process);
+  auto active_streams = streams.active_gpu_subset(num_blocks_to_process,
+                                                  mem_ptr->params.pbs_type);
   auto lut = mem_ptr->luts;
 
   if (active_streams.count() == 1) {

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

@@ -34,49 +34,49 @@ void cuda_rerand_64(
 
   switch (rerand_buffer->params.big_lwe_dimension) {
   case 256:
-    rerand_inplace<uint64_t, AmortizedDegree<256>>(
+    host_rerand_inplace<uint64_t, AmortizedDegree<256>>(
         streams, static_cast<uint64_t *>(lwe_array),
         static_cast<const uint64_t *>(
             lwe_flattened_encryptions_of_zero_compact_array_in),
         (uint64_t **)(ksk), rerand_buffer);
     break;
   case 512:
-    rerand_inplace<uint64_t, AmortizedDegree<512>>(
+    host_rerand_inplace<uint64_t, AmortizedDegree<512>>(
         streams, static_cast<uint64_t *>(lwe_array),
         static_cast<const uint64_t *>(
             lwe_flattened_encryptions_of_zero_compact_array_in),
         (uint64_t **)(ksk), rerand_buffer);
     break;
   case 1024:
-    rerand_inplace<uint64_t, AmortizedDegree<1024>>(
+    host_rerand_inplace<uint64_t, AmortizedDegree<1024>>(
         streams, static_cast<uint64_t *>(lwe_array),
         static_cast<const uint64_t *>(
             lwe_flattened_encryptions_of_zero_compact_array_in),
         (uint64_t **)(ksk), rerand_buffer);
     break;
   case 2048:
-    rerand_inplace<uint64_t, AmortizedDegree<2048>>(
+    host_rerand_inplace<uint64_t, AmortizedDegree<2048>>(
         streams, static_cast<uint64_t *>(lwe_array),
         static_cast<const uint64_t *>(
             lwe_flattened_encryptions_of_zero_compact_array_in),
         (uint64_t **)(ksk), rerand_buffer);
     break;
   case 4096:
-    rerand_inplace<uint64_t, AmortizedDegree<4096>>(
+    host_rerand_inplace<uint64_t, AmortizedDegree<4096>>(
         streams, static_cast<uint64_t *>(lwe_array),
         static_cast<const uint64_t *>(
             lwe_flattened_encryptions_of_zero_compact_array_in),
         (uint64_t **)(ksk), rerand_buffer);
     break;
   case 8192:
-    rerand_inplace<uint64_t, AmortizedDegree<8192>>(
+    host_rerand_inplace<uint64_t, AmortizedDegree<8192>>(
         streams, static_cast<uint64_t *>(lwe_array),
         static_cast<const uint64_t *>(
             lwe_flattened_encryptions_of_zero_compact_array_in),
         (uint64_t **)(ksk), rerand_buffer);
     break;
   case 16384:
-    rerand_inplace<uint64_t, AmortizedDegree<16384>>(
+    host_rerand_inplace<uint64_t, AmortizedDegree<16384>>(
         streams, static_cast<uint64_t *>(lwe_array),
         static_cast<const uint64_t *>(
             lwe_flattened_encryptions_of_zero_compact_array_in),
@@ -88,9 +88,6 @@ void cuda_rerand_64(
           " in the interval [256..16384].");
     break;
   }
-
-  cuda_synchronize_stream(static_cast<cudaStream_t>(streams.streams[0]),
-                          streams.gpu_indexes[0]);
 }
 
 void cleanup_cuda_rerand(CudaStreamsFFI streams, int8_t **mem_ptr_void) {

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

@@ -10,7 +10,7 @@
 #include "zk/zk_utilities.h"
 
 template <typename Torus, class params>
-void rerand_inplace(
+void host_rerand_inplace(
     CudaStreams const streams, Torus *lwe_array,
     const Torus *lwe_flattened_encryptions_of_zero_compact_array_in,
     Torus *const *ksk, int_rerand_mem<Torus> *mem_ptr) {
@@ -73,6 +73,7 @@ void rerand_inplace(
                        message_modulus, carry_modulus);
   release_cpu_radix_ciphertext_async(&lwes_ffi);
   release_cpu_radix_ciphertext_async(&ksed_zero_lwes_ffi);
+  compact_lwe_lists.release();
 }
 
 template <typename Torus>

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

@@ -45,7 +45,8 @@ host_scalar_bitop(CudaStreams streams, CudaRadixCiphertextFFI *output,
     cuda_memcpy_async_gpu_to_gpu(lut->get_lut_indexes(0, 0), clear_blocks,
                                  num_clear_blocks * sizeof(Torus),
                                  streams.stream(0), streams.gpu_index(0));
-    auto active_streams = streams.active_gpu_subset(num_clear_blocks);
+    auto active_streams = streams.active_gpu_subset(
+        num_clear_blocks, mem_ptr->lut->params.pbs_type);
     lut->broadcast_lut(active_streams, false);
 
     integer_radix_apply_univariate_lookup_table<Torus>(

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

@@ -146,7 +146,7 @@ __host__ void integer_radix_unsigned_scalar_difference_check(
         lut->get_degree(0), lut->get_max_degree(0), glwe_dimension,
         polynomial_size, message_modulus, carry_modulus, scalar_last_leaf_lut_f,
         true, mem_ptr->diff_buffer->tree_buffer->preallocated_h_lut);
-    auto active_streams = streams.active_gpu_subset(1);
+    auto active_streams = streams.active_gpu_subset(1, params.pbs_type);
     lut->broadcast_lut(active_streams);
 
     integer_radix_apply_univariate_lookup_table<Torus>(
@@ -240,7 +240,7 @@ __host__ void integer_radix_unsigned_scalar_difference_check(
         polynomial_size, message_modulus, carry_modulus,
         scalar_bivariate_last_leaf_lut_f, true,
         mem_ptr->diff_buffer->tree_buffer->preallocated_h_lut);
-    auto active_streams = streams.active_gpu_subset(1);
+    auto active_streams = streams.active_gpu_subset(1, params.pbs_type);
     lut->broadcast_lut(active_streams);
 
     integer_radix_apply_bivariate_lookup_table<Torus>(
@@ -274,7 +274,7 @@ __host__ void integer_radix_unsigned_scalar_difference_check(
           params.glwe_dimension, params.polynomial_size, params.message_modulus,
           params.carry_modulus, one_block_lut_f, true,
           mem_ptr->preallocated_h_lut);
-      auto active_streams = streams.active_gpu_subset(1);
+      auto active_streams = streams.active_gpu_subset(1, params.pbs_type);
       one_block_lut->broadcast_lut(active_streams);
 
       integer_radix_apply_univariate_lookup_table<Torus>(
@@ -419,7 +419,7 @@ __host__ void integer_radix_signed_scalar_difference_check(
         polynomial_size, message_modulus, carry_modulus,
         scalar_bivariate_last_leaf_lut_f, true,
         mem_ptr->diff_buffer->tree_buffer->preallocated_h_lut);
-    auto active_streams = streams.active_gpu_subset(1);
+    auto active_streams = streams.active_gpu_subset(1, params.pbs_type);
     lut->broadcast_lut(active_streams);
 
     integer_radix_apply_bivariate_lookup_table<Torus>(
@@ -521,7 +521,7 @@ __host__ void integer_radix_signed_scalar_difference_check(
         signed_msb_lut->get_max_degree(0), params.glwe_dimension,
         params.polynomial_size, params.message_modulus, params.carry_modulus,
         lut_f, true, mem_ptr->preallocated_h_lut);
-    auto active_streams = streams.active_gpu_subset(1);
+    auto active_streams = streams.active_gpu_subset(1, params.pbs_type);
     signed_msb_lut->broadcast_lut(active_streams);
 
     CudaRadixCiphertextFFI sign_block;
@@ -567,7 +567,7 @@ __host__ void integer_radix_signed_scalar_difference_check(
           params.glwe_dimension, params.polynomial_size, params.message_modulus,
           params.carry_modulus, one_block_lut_f, true,
           mem_ptr->preallocated_h_lut);
-      auto active_streams = streams.active_gpu_subset(1);
+      auto active_streams = streams.active_gpu_subset(1, params.pbs_type);
       one_block_lut->broadcast_lut(active_streams);
 
       integer_radix_apply_univariate_lookup_table<Torus>(
@@ -785,8 +785,8 @@ __host__ void host_scalar_equality_check(
           num_halved_scalar_blocks * sizeof(Torus), lsb_streams.stream(0),
           lsb_streams.gpu_index(0));
     }
-    auto active_streams =
-        lsb_streams.active_gpu_subset(num_halved_scalar_blocks);
+    auto active_streams = lsb_streams.active_gpu_subset(
+        num_halved_scalar_blocks, params.pbs_type);
     // We use false cause we only will broadcast the indexes
     scalar_comparison_luts->broadcast_lut(active_streams, false);
 

backends/tfhe-cuda-backend/cuda/src/pbs/programmable_bootstrap_cg_multibit.cuh

@@ -30,7 +30,7 @@ __global__ void __launch_bounds__(params::degree / params::opt)
         Torus *global_accumulator, uint32_t lwe_dimension,
         uint32_t glwe_dimension, uint32_t polynomial_size, uint32_t base_log,
         uint32_t level_count, uint32_t grouping_factor, uint32_t lwe_offset,
-        uint64_t lwe_chunk_size, uint64_t keybundle_size_per_input,
+        uint32_t lwe_chunk_size, uint64_t keybundle_size_per_input,
         int8_t *device_mem, uint64_t device_memory_size_per_block,
         uint32_t num_many_lut, uint32_t lut_stride) {
 
@@ -321,8 +321,9 @@ __host__ void execute_cg_external_product_loop(
       lwe_chunk_size * level_count * (glwe_dimension + 1) *
       (glwe_dimension + 1) * (polynomial_size / 2);
 
-  uint64_t chunk_size = std::min(
-      lwe_chunk_size, (uint64_t)(lwe_dimension / grouping_factor) - lwe_offset);
+  uint32_t chunk_size = (uint32_t)(std::min(
+      lwe_chunk_size,
+      (uint64_t)(lwe_dimension / grouping_factor) - lwe_offset));
 
   auto d_mem = buffer->d_mem_acc_cg;
   auto keybundle_fft = buffer->keybundle_fft;