test(core_crypto): improve coverage for various files

Those statements, if they contain a custom string to display,
are not covered in the test even when implementing tests that
trigger the assert macro.

.github/workflows/aws_tfhe_fast_tests.yml

@@ -51,7 +51,7 @@ jobs:
           echo "Fork git sha: ${{ inputs.fork_git_sha }}"
 
       - name: Checkout tfhe-rs
-        uses: actions/checkout@f43a0e5ff2bd294095638e18286ca9a3d1956744
+        uses: actions/checkout@b4ffde65f46336ab88eb53be808477a3936bae11
         with:
           repository: ${{ inputs.fork_repo }}
           ref: ${{ inputs.fork_git_sha }}
@@ -66,6 +66,10 @@ jobs:
           toolchain: stable
           default: true
 
+      - name: Run concrete-csprng tests
+        run: |
+          make test_concrete_csprng
+
       - name: Run core tests
         run: |
           AVX512_SUPPORT=ON make test_core_crypto
@@ -106,6 +110,10 @@ jobs:
         run: |
           make test_high_level_api
 
+      - name: Run safe deserialization tests
+        run: |
+          make test_safe_deserialization
+
       - name: Slack Notification
         if: ${{ always() }}
         continue-on-error: true

.github/workflows/aws_tfhe_integer_tests.yml

@@ -1,4 +1,4 @@
-name: AWS Integer Tests on CPU
+name: AWS Unsigned Integer Tests on CPU
 
 env:
   CARGO_TERM_COLOR: always
@@ -23,13 +23,13 @@ on:
         description: "Action runner name"
         type: string
       request_id:
-        description: 'Slab request ID'
+        description: "Slab request ID"
         type: string
       fork_repo:
-        description: 'Name of forked repo as user/repo'
+        description: "Name of forked repo as user/repo"
         type: string
       fork_git_sha:
-        description: 'Git SHA to checkout from fork'
+        description: "Git SHA to checkout from fork"
         type: string
 
 jobs:
@@ -50,7 +50,7 @@ jobs:
           echo "Fork git sha: ${{ inputs.fork_git_sha }}"
 
       - name: Checkout tfhe-rs
-        uses: actions/checkout@f43a0e5ff2bd294095638e18286ca9a3d1956744
+        uses: actions/checkout@b4ffde65f46336ab88eb53be808477a3936bae11
         with:
           repository: ${{ inputs.fork_repo }}
           ref: ${{ inputs.fork_git_sha }}
@@ -65,13 +65,21 @@ jobs:
           toolchain: stable
           default: true
 
+      - name: Gen Keys if required
+        run: |
+          make GEN_KEY_CACHE_MULTI_BIT_ONLY=TRUE gen_key_cache
+
+      - name: Run unsigned integer multi-bit tests
+        run: |
+          AVX512_SUPPORT=ON make test_unsigned_integer_multi_bit_ci
+
       - name: Gen Keys if required
         run: |
           make gen_key_cache
 
-      - name: Run integer tests
+      - name: Run unsigned integer tests
         run: |
-          BIG_TESTS_INSTANCE=TRUE make test_integer_ci
+          AVX512_SUPPORT=ON BIG_TESTS_INSTANCE=TRUE make test_unsigned_integer_ci
 
       - name: Slack Notification
         if: ${{ always() }}

.github/workflows/aws_tfhe_signed_integer_tests.yml

@@ -1,4 +1,4 @@
-name: AWS Multi Bit Tests on CPU
+name: AWS Signed Integer Tests on CPU
 
 env:
   CARGO_TERM_COLOR: always
@@ -23,13 +23,13 @@ on:
         description: "Action runner name"
         type: string
       request_id:
-        description: 'Slab request ID'
+        description: "Slab request ID"
         type: string
       fork_repo:
-        description: 'Name of forked repo as user/repo'
+        description: "Name of forked repo as user/repo"
         type: string
       fork_git_sha:
-        description: 'Git SHA to checkout from fork'
+        description: "Git SHA to checkout from fork"
         type: string
 
 jobs:
@@ -50,7 +50,7 @@ jobs:
           echo "Fork git sha: ${{ inputs.fork_git_sha }}"
 
       - name: Checkout tfhe-rs
-        uses: actions/checkout@f43a0e5ff2bd294095638e18286ca9a3d1956744
+        uses: actions/checkout@b4ffde65f46336ab88eb53be808477a3936bae11
         with:
           repository: ${{ inputs.fork_repo }}
           ref: ${{ inputs.fork_git_sha }}
@@ -73,9 +73,17 @@ jobs:
         run: |
           make test_shortint_multi_bit_ci
 
-      - name: Run integer multi-bit tests
+      - name: Run signed integer multi-bit tests
         run: |
-          make test_integer_multi_bit_ci
+          AVX512_SUPPORT=ON make test_signed_integer_multi_bit_ci
+
+      - name: Gen Keys if required
+        run: |
+          make gen_key_cache
+
+      - name: Run signed integer tests
+        run: |
+          AVX512_SUPPORT=ON BIG_TESTS_INSTANCE=TRUE make test_signed_integer_ci
 
       - name: Slack Notification
         if: ${{ always() }}

.github/workflows/aws_tfhe_tests.yml

@@ -50,7 +50,7 @@ jobs:
           echo "Fork git sha: ${{ inputs.fork_git_sha }}"
 
       - name: Checkout tfhe-rs
-        uses: actions/checkout@f43a0e5ff2bd294095638e18286ca9a3d1956744
+        uses: actions/checkout@b4ffde65f46336ab88eb53be808477a3936bae11
         with:
           repository: ${{ inputs.fork_repo }}
           ref: ${{ inputs.fork_git_sha }}
@@ -65,6 +65,10 @@ jobs:
           toolchain: stable
           default: true
 
+      - name: Run concrete-csprng tests
+        run: |
+          make test_concrete_csprng
+
       - name: Run core tests
         run: |
           AVX512_SUPPORT=ON make test_core_crypto
@@ -96,6 +100,12 @@ jobs:
       - name: Run example tests
         run: |
           make test_examples
+          make dark_market
+
+      - name: Run apps tests
+        run: |
+          make test_trivium
+          make test_kreyvium
 
       - name: Slack Notification
         if: ${{ always() }}

.github/workflows/aws_tfhe_wasm_tests.yml

@@ -50,7 +50,7 @@ jobs:
           echo "Fork git sha: ${{ inputs.fork_git_sha }}"
 
       - name: Checkout tfhe-rs
-        uses: actions/checkout@f43a0e5ff2bd294095638e18286ca9a3d1956744
+        uses: actions/checkout@b4ffde65f46336ab88eb53be808477a3936bae11
         with:
           repository: ${{ inputs.fork_repo }}
           ref: ${{ inputs.fork_git_sha }}

.github/workflows/boolean_benchmark.yml

@@ -19,6 +19,14 @@ on:
       request_id:
         description: "Slab request ID"
         type: string
+      # This input is not used in this workflow but still mandatory since a calling workflow could
+      # use it. If a triggering command include a user_inputs field, then the triggered workflow
+      # must include this very input, otherwise the workflow won't be called.
+      # See start_full_benchmarks.yml as example.
+      user_inputs:
+        description: "Type of benchmarks to run"
+        type: string
+        default: "weekly_benchmarks"
 
 env:
   CARGO_TERM_COLOR: always
@@ -43,7 +51,7 @@ jobs:
           echo "BENCH_DATE=$(date --iso-8601=seconds)" >> "${GITHUB_ENV}"
 
       - name: Checkout tfhe-rs repo with tags
-        uses: actions/checkout@f43a0e5ff2bd294095638e18286ca9a3d1956744
+        uses: actions/checkout@b4ffde65f46336ab88eb53be808477a3936bae11
         with:
           fetch-depth: 0
 
@@ -88,13 +96,13 @@ jobs:
           --append-results
 
       - name: Upload parsed results artifact
-        uses: actions/upload-artifact@0b7f8abb1508181956e8e162db84b466c27e18ce
+        uses: actions/upload-artifact@a8a3f3ad30e3422c9c7b888a15615d19a852ae32
         with:
           name: ${{ github.sha }}_boolean
           path: ${{ env.RESULTS_FILENAME }}
 
       - name: Checkout Slab repo
-        uses: actions/checkout@f43a0e5ff2bd294095638e18286ca9a3d1956744
+        uses: actions/checkout@b4ffde65f46336ab88eb53be808477a3936bae11
         with:
           repository: zama-ai/slab
           path: slab

.github/workflows/cargo_build.yml

@@ -21,12 +21,26 @@ jobs:
       fail-fast: false
 
     steps:
-      - uses: actions/checkout@f43a0e5ff2bd294095638e18286ca9a3d1956744
+      - uses: actions/checkout@b4ffde65f46336ab88eb53be808477a3936bae11
+
+      - name: Install and run newline linter checks
+        if: matrix.os == 'ubuntu-latest'
+        run: |
+          wget https://github.com/fernandrone/linelint/releases/download/0.0.6/linelint-linux-amd64
+          echo "16b70fb7b471d6f95cbdc0b4e5dc2b0ac9e84ba9ecdc488f7bdf13df823aca4b linelint-linux-amd64" > checksum
+          sha256sum -c checksum || exit 1
+          chmod +x linelint-linux-amd64
+          mv linelint-linux-amd64 /usr/local/bin/linelint
+          make check_newline
 
       - name: Run pcc checks
         run: |
           make pcc
 
+      - name: Build concrete-csprng
+        run: |
+          make build_concrete_csprng
+
       - name: Build Release core
         run: |
           make build_core AVX512_SUPPORT=ON

.github/workflows/code_coverage.yml

@@ -0,0 +1,119 @@
+name: Code Coverage
+
+env:
+  CARGO_TERM_COLOR: always
+  ACTION_RUN_URL: ${{ github.server_url }}/${{ github.repository }}/actions/runs/${{ github.run_id }}
+  RUSTFLAGS: "-C target-cpu=native"
+
+on:
+  # Allows you to run this workflow manually from the Actions tab as an alternative.
+  workflow_dispatch:
+    # All the inputs are provided by Slab
+    inputs:
+      instance_id:
+        description: "AWS instance ID"
+        type: string
+      instance_image_id:
+        description: "AWS instance AMI ID"
+        type: string
+      instance_type:
+        description: "AWS instance product type"
+        type: string
+      runner_name:
+        description: "Action runner name"
+        type: string
+      request_id:
+        description: 'Slab request ID'
+        type: string
+      fork_repo:
+        description: 'Name of forked repo as user/repo'
+        type: string
+      fork_git_sha:
+        description: 'Git SHA to checkout from fork'
+        type: string
+
+jobs:
+  code-coverage:
+    concurrency:
+      group: ${{ github.workflow }}_${{ github.ref }}_${{ inputs.instance_image_id }}_${{ inputs.instance_type }}
+      cancel-in-progress: true
+    runs-on: ${{ inputs.runner_name }}
+    timeout-minutes: 1080
+    steps:
+      # Step used for log purpose.
+      - name: Instance configuration used
+        run: |
+          echo "ID: ${{ inputs.instance_id }}"
+          echo "AMI: ${{ inputs.instance_image_id }}"
+          echo "Type: ${{ inputs.instance_type }}"
+          echo "Request ID: ${{ inputs.request_id }}"
+          echo "Fork repo: ${{ inputs.fork_repo }}"
+          echo "Fork git sha: ${{ inputs.fork_git_sha }}"
+
+      - name: Checkout tfhe-rs
+        uses: actions/checkout@b4ffde65f46336ab88eb53be808477a3936bae11
+        with:
+          repository: ${{ inputs.fork_repo }}
+          ref: ${{ inputs.fork_git_sha }}
+
+      - name: Set up home
+        run: |
+          echo "HOME=/home/ubuntu" >> "${GITHUB_ENV}"
+
+      - name: Install latest stable
+        uses: actions-rs/toolchain@16499b5e05bf2e26879000db0c1d13f7e13fa3af
+        with:
+          toolchain: stable
+          default: true
+
+      - name: Check for file changes
+        id: changed-files
+        uses: tj-actions/changed-files@1c938490c880156b746568a518594309cfb3f66b
+        with:
+          files_yaml: |
+            tfhe:
+              - tfhe/src/**
+            concrete_csprng:
+              - concrete-csprng/src/**
+
+      - name: Generate Keys
+        if: steps.changed-files.outputs.tfhe_any_changed == 'true'
+        run: |
+          make GEN_KEY_CACHE_COVERAGE_ONLY=TRUE gen_key_cache
+          make gen_key_cache_core_crypto
+
+      - name: Run coverage for core_crypto
+        if: steps.changed-files.outputs.tfhe_any_changed == 'true'
+        run: |
+          make test_core_crypto_cov AVX512_SUPPORT=ON
+
+      - name: Run coverage for boolean
+        if: steps.changed-files.outputs.tfhe_any_changed == 'true'
+        run: |
+          make test_boolean_cov
+
+      - name: Run coverage for shortint
+        if: steps.changed-files.outputs.tfhe_any_changed == 'true'
+        run: |
+          make test_shortint_cov
+
+      - name: Upload tfhe coverage to Codecov
+        uses: codecov/codecov-action@eaaf4bedf32dbdc6b720b63067d99c4d77d6047d
+        if: steps.changed-files.outputs.tfhe_any_changed == 'true'
+        with:
+          token: ${{ secrets.CODECOV_TOKEN }}
+          directory: ./coverage/
+          fail_ci_if_error: true
+          files: shortint/cobertura.xml,boolean/cobertura.xml,core_crypto/cobertura.xml,core_crypto_avx512/cobertura.xml
+
+      - name: Slack Notification
+        if: ${{ failure() }}
+        continue-on-error: true
+        uses: rtCamp/action-slack-notify@b24d75fe0e728a4bf9fc42ee217caa686d141ee8
+        env:
+          SLACK_COLOR: ${{ job.status }}
+          SLACK_CHANNEL: ${{ secrets.SLACK_CHANNEL }}
+          SLACK_ICON: https://pbs.twimg.com/profile_images/1274014582265298945/OjBKP9kn_400x400.png
+          SLACK_MESSAGE: "Code coverage finished with status: ${{ job.status }}. (${{ env.ACTION_RUN_URL }})"
+          SLACK_USERNAME: ${{ secrets.BOT_USERNAME }}
+          SLACK_WEBHOOK: ${{ secrets.SLACK_WEBHOOK }}

.github/workflows/csprng_randomness_testing.yml

@@ -0,0 +1,74 @@
+name: CSPRNG randomness testing Workflow
+
+env:
+  CARGO_TERM_COLOR: always
+  ACTION_RUN_URL: ${{ github.server_url }}/${{ github.repository }}/actions/runs/${{ github.run_id }}
+  RUSTFLAGS: "-C target-cpu=native"
+
+on:
+  # Allows you to run this workflow manually from the Actions tab as an alternative.
+  workflow_dispatch:
+    # All the inputs are provided by Slab
+    inputs:
+      instance_id:
+        description: "AWS instance ID"
+        type: string
+      instance_image_id:
+        description: "AWS instance AMI ID"
+        type: string
+      instance_type:
+        description: "AWS instance product type"
+        type: string
+      runner_name:
+        description: "Action runner name"
+        type: string
+      request_id:
+        description: 'Slab request ID'
+        type: string
+      fork_repo:
+        description: 'Name of forked repo as user/repo'
+        type: string
+      fork_git_sha:
+        description: 'Git SHA to checkout from fork'
+        type: string
+
+jobs:
+  csprng-randomness-teting:
+    name: CSPRNG randomness testing
+    concurrency:
+      group: ${{ github.workflow }}_${{ github.ref }}_${{ inputs.instance_image_id }}_${{ inputs.instance_type }}
+      cancel-in-progress: true
+    runs-on: ${{ inputs.runner_name }}
+
+    steps:
+      - name: Checkout tfhe-rs
+        uses: actions/checkout@b4ffde65f46336ab88eb53be808477a3936bae11
+        with:
+          repository: ${{ inputs.fork_repo }}
+          ref: ${{ inputs.fork_git_sha }}
+
+      - name: Set up home
+        run: |
+          echo "HOME=/home/ubuntu" >> "${GITHUB_ENV}"
+
+      - name: Install latest stable
+        uses: actions-rs/toolchain@16499b5e05bf2e26879000db0c1d13f7e13fa3af
+        with:
+          toolchain: stable
+          default: true
+
+      - name: Dieharder randomness test suite
+        run: |
+          make dieharder_csprng
+
+      - name: Slack Notification
+        if: ${{ failure() }}
+        continue-on-error: true
+        uses: rtCamp/action-slack-notify@b24d75fe0e728a4bf9fc42ee217caa686d141ee8
+        env:
+          SLACK_COLOR: ${{ job.status }}
+          SLACK_CHANNEL: ${{ secrets.SLACK_CHANNEL }}
+          SLACK_ICON: https://pbs.twimg.com/profile_images/1274014582265298945/OjBKP9kn_400x400.png
+          SLACK_MESSAGE: "concrete-csprng randomness check finished with status: ${{ job.status }}. (${{ env.ACTION_RUN_URL }})"
+          SLACK_USERNAME: ${{ secrets.BOT_USERNAME }}
+          SLACK_WEBHOOK: ${{ secrets.SLACK_WEBHOOK }}

.github/workflows/integer_benchmark.yml

@@ -44,7 +44,7 @@ jobs:
           echo "BENCH_DATE=$(date --iso-8601=seconds)" >> "${GITHUB_ENV}"
 
       - name: Checkout tfhe-rs repo with tags
-        uses: actions/checkout@f43a0e5ff2bd294095638e18286ca9a3d1956744
+        uses: actions/checkout@b4ffde65f46336ab88eb53be808477a3936bae11
         with:
           fetch-depth: 0
 
@@ -69,7 +69,7 @@ jobs:
             parse_integer_benches
 
       - name: Upload csv results artifact
-        uses: actions/upload-artifact@0b7f8abb1508181956e8e162db84b466c27e18ce
+        uses: actions/upload-artifact@a8a3f3ad30e3422c9c7b888a15615d19a852ae32
         with:
           name: ${{ github.sha }}_csv_integer
           path: ${{ env.PARSE_INTEGER_BENCH_CSV_FILE }}
@@ -90,13 +90,13 @@ jobs:
           --throughput
 
       - name: Upload parsed results artifact
-        uses: actions/upload-artifact@0b7f8abb1508181956e8e162db84b466c27e18ce
+        uses: actions/upload-artifact@a8a3f3ad30e3422c9c7b888a15615d19a852ae32
         with:
           name: ${{ github.sha }}_integer
           path: ${{ env.RESULTS_FILENAME }}
 
       - name: Checkout Slab repo
-        uses: actions/checkout@f43a0e5ff2bd294095638e18286ca9a3d1956744
+        uses: actions/checkout@b4ffde65f46336ab88eb53be808477a3936bae11
         with:
           repository: zama-ai/slab
           path: slab

.github/workflows/integer_full_benchmark.yml

@@ -19,6 +19,10 @@ on:
       request_id:
         description: "Slab request ID"
         type: string
+      user_inputs:
+        description: "Type of benchmarks to run"
+        type: string
+        default: "weekly_benchmarks"
 
 env:
   CARGO_TERM_COLOR: always
@@ -26,8 +30,33 @@ env:
   ACTION_RUN_URL: ${{ github.server_url }}/${{ github.repository }}/actions/runs/${{ github.run_id }}
 
 jobs:
+  prepare-matrix:
+    name: Prepare operations matrix
+    runs-on: ubuntu-latest
+    outputs:
+      op_flavor: ${{ steps.set_op_flavor.outputs.op_flavor }}
+    steps:
+      - name: Weekly benchmarks
+        if: ${{ github.event.inputs.user_inputs == 'weekly_benchmarks' }}
+        run: |
+          echo "OP_FLAVOR=[\"default\", \"default_comp\", \"default_scalar\", \"default_scalar_comp\"]" >> ${GITHUB_ENV}
+
+      - name: Quarterly benchmarks
+        if: ${{ github.event.inputs.user_inputs == 'quarterly_benchmarks' }}
+        run: |
+          echo "OP_FLAVOR=[\"default\", \"default_comp\", \"default_scalar\", \"default_scalar_comp\", \
+          \"smart\", \"smart_comp\", \"smart_scalar\", \"smart_parallelized\", \"smart_parallelized_comp\", \"smart_scalar_parallelized\", \"smart_scalar_parallelized_comp\", \
+          \"unchecked\", \"unchecked_comp\", \"unchecked_scalar\", \"unchecked_scalar_comp\", \
+          \"misc\"]" >> ${GITHUB_ENV}
+
+      -  name: Set operation flavor output
+         id: set_op_flavor
+         run: |
+          echo "op_flavor=${{ toJSON(env.OP_FLAVOR) }}" >> ${GITHUB_OUTPUT}
+
   integer-benchmarks:
     name: Execute integer benchmarks for all operations flavor
+    needs: prepare-matrix
     runs-on: ${{ github.event.inputs.runner_name }}
     if: ${{ !cancelled() }}
     continue-on-error: true
@@ -35,7 +64,7 @@ jobs:
       max-parallel: 1
       matrix:
         command: [ integer, integer_multi_bit]
-        op_flavor: [ default, default_comp, default_scalar, default_scalar_comp, smart, smart_comp, smart_scalar, smart_parallelized, smart_parallelized_comp, smart_scalar_parallelized, unchecked, unchecked_comp, unchecked_scalar, unchecked_scalar_comp, misc ]
+        op_flavor: ${{ fromJson(needs.prepare-matrix.outputs.op_flavor) }}
     steps:
       - name: Instance configuration used
         run: |
@@ -45,7 +74,7 @@ jobs:
           echo "Request ID: ${{ inputs.request_id }}"
 
       - name: Checkout tfhe-rs repo with tags
-        uses: actions/checkout@f43a0e5ff2bd294095638e18286ca9a3d1956744
+        uses: actions/checkout@b4ffde65f46336ab88eb53be808477a3936bae11
         with:
           fetch-depth: 0
 
@@ -67,7 +96,7 @@ jobs:
           override: true
 
       - name: Checkout Slab repo
-        uses: actions/checkout@f43a0e5ff2bd294095638e18286ca9a3d1956744
+        uses: actions/checkout@b4ffde65f46336ab88eb53be808477a3936bae11
         with:
           repository: zama-ai/slab
           path: slab
@@ -91,7 +120,7 @@ jobs:
           --throughput
 
       - name: Upload parsed results artifact
-        uses: actions/upload-artifact@0b7f8abb1508181956e8e162db84b466c27e18ce
+        uses: actions/upload-artifact@a8a3f3ad30e3422c9c7b888a15615d19a852ae32
         with:
           name: ${{ github.sha }}_${{ matrix.command }}_${{ matrix.op_flavor }}
           path: ${{ env.RESULTS_FILENAME }}

.github/workflows/integer_multi_bit_benchmark.yml

@@ -44,7 +44,7 @@ jobs:
           echo "BENCH_DATE=$(date --iso-8601=seconds)" >> "${GITHUB_ENV}"
 
       - name: Checkout tfhe-rs repo with tags
-        uses: actions/checkout@f43a0e5ff2bd294095638e18286ca9a3d1956744
+        uses: actions/checkout@b4ffde65f46336ab88eb53be808477a3936bae11
         with:
           fetch-depth: 0
 
@@ -69,7 +69,7 @@ jobs:
             parse_integer_benches
 
       - name: Upload csv results artifact
-        uses: actions/upload-artifact@0b7f8abb1508181956e8e162db84b466c27e18ce
+        uses: actions/upload-artifact@a8a3f3ad30e3422c9c7b888a15615d19a852ae32
         with:
           name: ${{ github.sha }}_csv_integer
           path: ${{ env.PARSE_INTEGER_BENCH_CSV_FILE }}
@@ -90,13 +90,13 @@ jobs:
           --throughput
 
       - name: Upload parsed results artifact
-        uses: actions/upload-artifact@0b7f8abb1508181956e8e162db84b466c27e18ce
+        uses: actions/upload-artifact@a8a3f3ad30e3422c9c7b888a15615d19a852ae32
         with:
           name: ${{ github.sha }}_integer
           path: ${{ env.RESULTS_FILENAME }}
 
       - name: Checkout Slab repo
-        uses: actions/checkout@f43a0e5ff2bd294095638e18286ca9a3d1956744
+        uses: actions/checkout@b4ffde65f46336ab88eb53be808477a3936bae11
         with:
           repository: zama-ai/slab
           path: slab

.github/workflows/m1_tests.yml

@@ -15,7 +15,6 @@ env:
   CARGO_TERM_COLOR: always
   RUSTFLAGS: "-C target-cpu=native"
   ACTION_RUN_URL: ${{ github.server_url }}/${{ github.repository }}/actions/runs/${{ github.run_id }}
-  CARGO_PROFILE: release_lto_off
   FAST_TESTS: "TRUE"
 
 concurrency:
@@ -28,7 +27,7 @@ jobs:
     runs-on: ["self-hosted", "m1mac"]
 
     steps:
-      - uses: actions/checkout@f43a0e5ff2bd294095638e18286ca9a3d1956744
+      - uses: actions/checkout@b4ffde65f46336ab88eb53be808477a3936bae11
 
       - name: Install latest stable
         uses: actions-rs/toolchain@16499b5e05bf2e26879000db0c1d13f7e13fa3af
@@ -40,6 +39,10 @@ jobs:
         run: |
           make pcc
 
+      - name: Build concrete-csprng
+        run: |
+          make build_concrete_csprng
+
       - name: Build Release core
         run: |
           make build_core
@@ -64,6 +67,10 @@ jobs:
         run: |
           make build_c_api
 
+      - name: Run concrete-csprng tests
+        run: |
+          make test_concrete_csprng
+
       - name: Run core tests
         run: |
           make test_core_crypto
@@ -103,10 +110,9 @@ jobs:
         run: |
           make test_shortint_multi_bit_ci
 
-      # # These multi bit integer tests are too slow on M1 with low core count and low RAM
-      # - name: Run integer multi bit tests
-      #   run: |
-      #     make test_integer_multi_bit_ci
+      - name: Run integer multi bit tests
+        run: |
+          make test_integer_multi_bit_ci
 
   remove_label:
     name: Remove m1_test label

.github/workflows/make_release.yml

@@ -30,7 +30,7 @@ jobs:
     runs-on: ubuntu-latest
     steps:
       - name: Checkout
-        uses: actions/checkout@f43a0e5ff2bd294095638e18286ca9a3d1956744
+        uses: actions/checkout@b4ffde65f46336ab88eb53be808477a3936bae11
         with:
           fetch-depth: 0
 
@@ -49,7 +49,7 @@ jobs:
 
       - name: Publish web package
         if: ${{ inputs.push_web_package }}
-        uses: JS-DevTools/npm-publish@5a85faf05d2ade2d5b6682bfe5359915d5159c6c
+        uses: JS-DevTools/npm-publish@4b07b26a2f6e0a51846e1870223e545bae91c552
         with:
           token: ${{ secrets.NPM_TOKEN }}
           package: tfhe/pkg/package.json
@@ -65,7 +65,7 @@ jobs:
 
       - name: Publish Node package
         if: ${{ inputs.push_node_package }}
-        uses: JS-DevTools/npm-publish@5a85faf05d2ade2d5b6682bfe5359915d5159c6c
+        uses: JS-DevTools/npm-publish@4b07b26a2f6e0a51846e1870223e545bae91c552
         with:
           token: ${{ secrets.NPM_TOKEN }}
           package: tfhe/pkg/package.json
@@ -79,6 +79,6 @@ jobs:
           SLACK_COLOR: ${{ job.status }}
           SLACK_CHANNEL: ${{ secrets.SLACK_CHANNEL }}
           SLACK_ICON: https://pbs.twimg.com/profile_images/1274014582265298945/OjBKP9kn_400x400.png
-          SLACK_MESSAGE: "Integer benchmarks failed. (${{ env.ACTION_RUN_URL }})"
+          SLACK_MESSAGE: "tfhe release failed: (${{ env.ACTION_RUN_URL }})"
           SLACK_USERNAME: ${{ secrets.BOT_USERNAME }}
           SLACK_WEBHOOK: ${{ secrets.SLACK_WEBHOOK }}

.github/workflows/make_release_concrete_csprng.yml

@@ -0,0 +1,42 @@
+# Publish new release of tfhe-rs on various platform.
+name: Publish concrete-csprng release
+
+on:
+  workflow_dispatch:
+    inputs:
+      dry_run:
+        description: "Dry-run"
+        type: boolean
+        default: true
+
+env:
+  ACTION_RUN_URL: ${{ github.server_url }}/${{ github.repository }}/actions/runs/${{ github.run_id }}
+
+jobs:
+  publish_release:
+    name: Publish concrete-csprng Release
+    runs-on: ubuntu-latest
+    steps:
+      - name: Checkout
+        uses: actions/checkout@b4ffde65f46336ab88eb53be808477a3936bae11
+        with:
+          fetch-depth: 0
+
+      - name: Publish crate.io package
+        env:
+          CRATES_TOKEN: ${{ secrets.CARGO_REGISTRY_TOKEN }}
+          DRY_RUN: ${{ inputs.dry_run && '--dry-run' || '' }}
+        run: |
+          cargo publish -p concrete-csprng --token ${{ env.CRATES_TOKEN }} ${{ env.DRY_RUN }}
+
+      - name: Slack Notification
+        if: ${{ failure() }}
+        continue-on-error: true
+        uses: rtCamp/action-slack-notify@b24d75fe0e728a4bf9fc42ee217caa686d141ee8
+        env:
+          SLACK_COLOR: ${{ job.status }}
+          SLACK_CHANNEL: ${{ secrets.SLACK_CHANNEL }}
+          SLACK_ICON: https://pbs.twimg.com/profile_images/1274014582265298945/OjBKP9kn_400x400.png
+          SLACK_MESSAGE: "concrete-csprng release failed: (${{ env.ACTION_RUN_URL }})"
+          SLACK_USERNAME: ${{ secrets.BOT_USERNAME }}
+          SLACK_WEBHOOK: ${{ secrets.SLACK_WEBHOOK }}

.github/workflows/parameters_check.yml

@@ -17,10 +17,10 @@ jobs:
     runs-on: ubuntu-latest
     steps:
       - name: Checkout tfhe-rs
-        uses: actions/checkout@f43a0e5ff2bd294095638e18286ca9a3d1956744
+        uses: actions/checkout@b4ffde65f46336ab88eb53be808477a3936bae11
 
       - name: Checkout lattice-estimator
-        uses: actions/checkout@f43a0e5ff2bd294095638e18286ca9a3d1956744
+        uses: actions/checkout@b4ffde65f46336ab88eb53be808477a3936bae11
         with:
           repository: malb/lattice-estimator
           path: lattice_estimator
@@ -32,7 +32,7 @@ jobs:
 
       - name: Collect parameters
         run: |
-          make write_params_to_file
+          CARGO_PROFILE=devo make write_params_to_file
 
       - name: Perform security check
         run: |

.github/workflows/pbs_benchmark.yml

@@ -19,6 +19,14 @@ on:
       request_id:
         description: "Slab request ID"
         type: string
+      # This input is not used in this workflow but still mandatory since a calling workflow could
+      # use it. If a triggering command include a user_inputs field, then the triggered workflow
+      # must include this very input, otherwise the workflow won't be called.
+      # See start_full_benchmarks.yml as example.
+      user_inputs:
+        description: "Type of benchmarks to run"
+        type: string
+        default: "weekly_benchmarks"
 
 env:
   CARGO_TERM_COLOR: always
@@ -43,7 +51,7 @@ jobs:
           echo "BENCH_DATE=$(date --iso-8601=seconds)" >> "${GITHUB_ENV}"
 
       - name: Checkout tfhe-rs repo with tags
-        uses: actions/checkout@f43a0e5ff2bd294095638e18286ca9a3d1956744
+        uses: actions/checkout@b4ffde65f46336ab88eb53be808477a3936bae11
         with:
           fetch-depth: 0
 
@@ -78,13 +86,13 @@ jobs:
           --throughput
 
       - name: Upload parsed results artifact
-        uses: actions/upload-artifact@0b7f8abb1508181956e8e162db84b466c27e18ce
+        uses: actions/upload-artifact@a8a3f3ad30e3422c9c7b888a15615d19a852ae32
         with:
           name: ${{ github.sha }}_pbs
           path: ${{ env.RESULTS_FILENAME }}
 
       - name: Checkout Slab repo
-        uses: actions/checkout@f43a0e5ff2bd294095638e18286ca9a3d1956744
+        uses: actions/checkout@b4ffde65f46336ab88eb53be808477a3936bae11
         with:
           repository: zama-ai/slab
           path: slab

.github/workflows/shortint_benchmark.yml

@@ -43,7 +43,7 @@ jobs:
           echo "BENCH_DATE=$(date --iso-8601=seconds)" >> "${GITHUB_ENV}"
 
       - name: Checkout tfhe-rs repo with tags
-        uses: actions/checkout@f43a0e5ff2bd294095638e18286ca9a3d1956744
+        uses: actions/checkout@b4ffde65f46336ab88eb53be808477a3936bae11
         with:
           fetch-depth: 0
 
@@ -88,13 +88,13 @@ jobs:
           --append-results
 
       - name: Upload parsed results artifact
-        uses: actions/upload-artifact@0b7f8abb1508181956e8e162db84b466c27e18ce
+        uses: actions/upload-artifact@a8a3f3ad30e3422c9c7b888a15615d19a852ae32
         with:
           name: ${{ github.sha }}_shortint
           path: ${{ env.RESULTS_FILENAME }}
 
       - name: Checkout Slab repo
-        uses: actions/checkout@f43a0e5ff2bd294095638e18286ca9a3d1956744
+        uses: actions/checkout@b4ffde65f46336ab88eb53be808477a3936bae11
         with:
           repository: zama-ai/slab
           path: slab

.github/workflows/shortint_full_benchmark.yml

@@ -19,6 +19,14 @@ on:
       request_id:
         description: "Slab request ID"
         type: string
+      # This input is not used in this workflow but still mandatory since a calling workflow could
+      # use it. If a triggering command include a user_inputs field, then the triggered workflow
+      # must include this very input, otherwise the workflow won't be called.
+      # See start_full_benchmarks.yml as example.
+      user_inputs:
+        description: "Type of benchmarks to run"
+        type: string
+        default: "weekly_benchmarks"
 
 env:
   CARGO_TERM_COLOR: always
@@ -43,7 +51,7 @@ jobs:
           echo "Request ID: ${{ inputs.request_id }}"
 
       - name: Checkout tfhe-rs repo with tags
-        uses: actions/checkout@f43a0e5ff2bd294095638e18286ca9a3d1956744
+        uses: actions/checkout@b4ffde65f46336ab88eb53be808477a3936bae11
         with:
           fetch-depth: 0
 
@@ -65,7 +73,7 @@ jobs:
           override: true
 
       - name: Checkout Slab repo
-        uses: actions/checkout@f43a0e5ff2bd294095638e18286ca9a3d1956744
+        uses: actions/checkout@b4ffde65f46336ab88eb53be808477a3936bae11
         with:
           repository: zama-ai/slab
           path: slab
@@ -104,7 +112,7 @@ jobs:
           --append-results
 
       - name: Upload parsed results artifact
-        uses: actions/upload-artifact@0b7f8abb1508181956e8e162db84b466c27e18ce
+        uses: actions/upload-artifact@a8a3f3ad30e3422c9c7b888a15615d19a852ae32
         with:
           name: ${{ github.sha }}_shortint_${{ matrix.op_flavor }}
           path: ${{ env.RESULTS_FILENAME }}

.github/workflows/signed_integer_benchmark.yml

@@ -0,0 +1,129 @@
+# Run signed integer benchmarks on an AWS instance and return parsed results to Slab CI bot.
+name: Signed Integer benchmarks
+
+on:
+  workflow_dispatch:
+    inputs:
+      instance_id:
+        description: "Instance ID"
+        type: string
+      instance_image_id:
+        description: "Instance AMI ID"
+        type: string
+      instance_type:
+        description: "Instance product type"
+        type: string
+      runner_name:
+        description: "Action runner name"
+        type: string
+      request_id:
+        description: "Slab request ID"
+        type: string
+
+env:
+  CARGO_TERM_COLOR: always
+  RESULTS_FILENAME: parsed_benchmark_results_${{ github.sha }}.json
+  PARSE_INTEGER_BENCH_CSV_FILE: tfhe_rs_integer_benches_${{ github.sha }}.csv
+  ACTION_RUN_URL: ${{ github.server_url }}/${{ github.repository }}/actions/runs/${{ github.run_id }}
+
+jobs:
+  run-integer-benchmarks:
+    name: Execute signed integer benchmarks in EC2
+    runs-on: ${{ github.event.inputs.runner_name }}
+    if: ${{ !cancelled() }}
+    steps:
+      - name: Instance configuration used
+        run: |
+          echo "IDs: ${{ inputs.instance_id }}"
+          echo "AMI: ${{ inputs.instance_image_id }}"
+          echo "Type: ${{ inputs.instance_type }}"
+          echo "Request ID: ${{ inputs.request_id }}"
+
+      - name: Get benchmark date
+        run: |
+          echo "BENCH_DATE=$(date --iso-8601=seconds)" >> "${GITHUB_ENV}"
+
+      - name: Checkout tfhe-rs repo with tags
+        uses: actions/checkout@b4ffde65f46336ab88eb53be808477a3936bae11
+        with:
+          fetch-depth: 0
+
+      - name: Set up home
+        # "Install rust" step require root user to have a HOME directory which is not set.
+        run: |
+          echo "HOME=/home/ubuntu" >> "${GITHUB_ENV}"
+
+      - name: Install rust
+        uses: actions-rs/toolchain@16499b5e05bf2e26879000db0c1d13f7e13fa3af
+        with:
+          toolchain: nightly
+          override: true
+
+      - name: Run benchmarks with AVX512
+        run: |
+          make AVX512_SUPPORT=ON FAST_BENCH=TRUE bench_signed_integer
+
+      - name: Parse benchmarks to csv
+        run: |
+          make PARSE_INTEGER_BENCH_CSV_FILE=${{ env.PARSE_INTEGER_BENCH_CSV_FILE }} \
+            parse_integer_benches
+
+      - name: Upload csv results artifact
+        uses: actions/upload-artifact@a8a3f3ad30e3422c9c7b888a15615d19a852ae32
+        with:
+          name: ${{ github.sha }}_csv_integer
+          path: ${{ env.PARSE_INTEGER_BENCH_CSV_FILE }}
+
+      - name: Parse results
+        run: |
+          COMMIT_DATE="$(git --no-pager show -s --format=%cd --date=iso8601-strict ${{ github.sha }})"
+          COMMIT_HASH="$(git describe --tags --dirty)"
+          python3 ./ci/benchmark_parser.py target/criterion ${{ env.RESULTS_FILENAME }} \
+          --database tfhe_rs \
+          --hardware ${{ inputs.instance_type }} \
+          --project-version "${COMMIT_HASH}" \
+          --branch ${{ github.ref_name }} \
+          --commit-date "${COMMIT_DATE}" \
+          --bench-date "${{ env.BENCH_DATE }}" \
+          --walk-subdirs \
+          --name-suffix avx512 \
+          --throughput
+
+      - name: Upload parsed results artifact
+        uses: actions/upload-artifact@a8a3f3ad30e3422c9c7b888a15615d19a852ae32
+        with:
+          name: ${{ github.sha }}_integer
+          path: ${{ env.RESULTS_FILENAME }}
+
+      - name: Checkout Slab repo
+        uses: actions/checkout@b4ffde65f46336ab88eb53be808477a3936bae11
+        with:
+          repository: zama-ai/slab
+          path: slab
+          token: ${{ secrets.CONCRETE_ACTIONS_TOKEN }}
+
+      - name: Send data to Slab
+        shell: bash
+        run: |
+          echo "Computing HMac on results file"
+          SIGNATURE="$(slab/scripts/hmac_calculator.sh ${{ env.RESULTS_FILENAME }} '${{ secrets.JOB_SECRET }}')"
+          echo "Sending results to Slab..."
+          curl -v -k \
+          -H "Content-Type: application/json" \
+          -H "X-Slab-Repository: ${{ github.repository }}" \
+          -H "X-Slab-Command: store_data_v2" \
+          -H "X-Hub-Signature-256: sha256=${SIGNATURE}" \
+          -d @${{ env.RESULTS_FILENAME }} \
+          ${{ secrets.SLAB_URL }}
+
+      - name: Slack Notification
+        if: ${{ failure() }}
+        continue-on-error: true
+        uses: rtCamp/action-slack-notify@b24d75fe0e728a4bf9fc42ee217caa686d141ee8
+        env:
+          SLACK_COLOR: ${{ job.status }}
+          SLACK_CHANNEL: ${{ secrets.SLACK_CHANNEL }}
+          SLACK_ICON: https://pbs.twimg.com/profile_images/1274014582265298945/OjBKP9kn_400x400.png
+          SLACK_MESSAGE: "Signed integer benchmarks failed. (${{ env.ACTION_RUN_URL }})"
+          SLACK_USERNAME: ${{ secrets.BOT_USERNAME }}
+          SLACK_WEBHOOK: ${{ secrets.SLACK_WEBHOOK }}

.github/workflows/signed_integer_full_benchmark.yml

@@ -0,0 +1,133 @@
+# Run all signed integer benchmarks on an AWS instance and return parsed results to Slab CI bot.
+name: Signed Integer full benchmarks
+
+on:
+  workflow_dispatch:
+    inputs:
+      instance_id:
+        description: "Instance ID"
+        type: string
+      instance_image_id:
+        description: "Instance AMI ID"
+        type: string
+      instance_type:
+        description: "Instance product type"
+        type: string
+      runner_name:
+        description: "Action runner name"
+        type: string
+      request_id:
+        description: "Slab request ID"
+        type: string
+      user_inputs:
+        description: "Type of benchmarks to run"
+        type: string
+        default: "weekly_benchmarks"
+
+env:
+  CARGO_TERM_COLOR: always
+  RESULTS_FILENAME: parsed_benchmark_results_${{ github.sha }}.json
+  ACTION_RUN_URL: ${{ github.server_url }}/${{ github.repository }}/actions/runs/${{ github.run_id }}
+
+jobs:
+  integer-benchmarks:
+    name: Execute signed integer benchmarks for all operations flavor
+    runs-on: ${{ github.event.inputs.runner_name }}
+    if: ${{ !cancelled() }}
+    continue-on-error: true
+    strategy:
+      max-parallel: 1
+      matrix:
+        command: [ integer, integer_multi_bit ]
+        op_flavor: [ default, default_comp, default_scalar, default_scalar_comp,
+                     unchecked, unchecked_comp, unchecked_scalar, unchecked_scalar_comp ]
+    steps:
+      - name: Instance configuration used
+        run: |
+          echo "IDs: ${{ inputs.instance_id }}"
+          echo "AMI: ${{ inputs.instance_image_id }}"
+          echo "Type: ${{ inputs.instance_type }}"
+          echo "Request ID: ${{ inputs.request_id }}"
+
+      - name: Checkout tfhe-rs repo with tags
+        uses: actions/checkout@b4ffde65f46336ab88eb53be808477a3936bae11
+        with:
+          fetch-depth: 0
+
+      - name: Get benchmark details
+        run: |
+          echo "BENCH_DATE=$(date --iso-8601=seconds)" >> "${GITHUB_ENV}"
+          echo "COMMIT_DATE=$(git --no-pager show -s --format=%cd --date=iso8601-strict ${{ github.sha }})" >> "${GITHUB_ENV}"
+          echo "COMMIT_HASH=$(git describe --tags --dirty)" >> "${GITHUB_ENV}"
+
+      - name: Set up home
+        # "Install rust" step require root user to have a HOME directory which is not set.
+        run: |
+          echo "HOME=/home/ubuntu" >> "${GITHUB_ENV}"
+
+      - name: Install rust
+        uses: actions-rs/toolchain@16499b5e05bf2e26879000db0c1d13f7e13fa3af
+        with:
+          toolchain: nightly
+          override: true
+
+      - name: Checkout Slab repo
+        uses: actions/checkout@b4ffde65f46336ab88eb53be808477a3936bae11
+        with:
+          repository: zama-ai/slab
+          path: slab
+          token: ${{ secrets.CONCRETE_ACTIONS_TOKEN }}
+
+      - name: Run benchmarks with AVX512
+        run: |
+          make AVX512_SUPPORT=ON BENCH_OP_FLAVOR=${{ matrix.op_flavor }} bench_signed_${{ matrix.command }}
+
+      - name: Parse results
+        run: |
+          python3 ./ci/benchmark_parser.py target/criterion ${{ env.RESULTS_FILENAME }} \
+          --database tfhe_rs \
+          --hardware ${{ inputs.instance_type }} \
+          --project-version "${{ env.COMMIT_HASH }}" \
+          --branch ${{ github.ref_name }} \
+          --commit-date "${{ env.COMMIT_DATE }}" \
+          --bench-date "${{ env.BENCH_DATE }}" \
+          --walk-subdirs \
+          --name-suffix avx512 \
+          --throughput
+
+      - name: Upload parsed results artifact
+        uses: actions/upload-artifact@a8a3f3ad30e3422c9c7b888a15615d19a852ae32
+        with:
+          name: ${{ github.sha }}_${{ matrix.command }}_${{ matrix.op_flavor }}
+          path: ${{ env.RESULTS_FILENAME }}
+
+      - name: Send data to Slab
+        shell: bash
+        run: |
+          echo "Computing HMac on results file"
+          SIGNATURE="$(slab/scripts/hmac_calculator.sh ${{ env.RESULTS_FILENAME }} '${{ secrets.JOB_SECRET }}')"
+          echo "Sending results to Slab..."
+          curl -v -k \
+          -H "Content-Type: application/json" \
+          -H "X-Slab-Repository: ${{ github.repository }}" \
+          -H "X-Slab-Command: store_data_v2" \
+          -H "X-Hub-Signature-256: sha256=${SIGNATURE}" \
+          -d @${{ env.RESULTS_FILENAME }} \
+          ${{ secrets.SLAB_URL }}
+
+  slack-notification:
+    name: Slack Notification
+    runs-on: ${{ github.event.inputs.runner_name }}
+    if: ${{ failure() }}
+    needs: integer-benchmarks
+    steps:
+      - name: Notify
+        continue-on-error: true
+        uses: rtCamp/action-slack-notify@b24d75fe0e728a4bf9fc42ee217caa686d141ee8
+        env:
+          SLACK_COLOR: ${{ job.status }}
+          SLACK_CHANNEL: ${{ secrets.SLACK_CHANNEL }}
+          SLACK_ICON: https://pbs.twimg.com/profile_images/1274014582265298945/OjBKP9kn_400x400.png
+          SLACK_MESSAGE: "Signed integer full benchmarks failed. (${{ env.ACTION_RUN_URL }})"
+          SLACK_USERNAME: ${{ secrets.BOT_USERNAME }}
+          SLACK_WEBHOOK: ${{ secrets.SLACK_WEBHOOK }}

.github/workflows/signed_integer_multi_bit_benchmark.yml

@@ -0,0 +1,129 @@
+# Run signed integer benchmarks with multi-bit cryptographic parameters on an AWS instance and return parsed results to Slab CI bot.
+name: Signed Integer Multi-bit benchmarks
+
+on:
+  workflow_dispatch:
+    inputs:
+      instance_id:
+        description: "Instance ID"
+        type: string
+      instance_image_id:
+        description: "Instance AMI ID"
+        type: string
+      instance_type:
+        description: "Instance product type"
+        type: string
+      runner_name:
+        description: "Action runner name"
+        type: string
+      request_id:
+        description: "Slab request ID"
+        type: string
+
+env:
+  CARGO_TERM_COLOR: always
+  RESULTS_FILENAME: parsed_benchmark_results_${{ github.sha }}.json
+  PARSE_INTEGER_BENCH_CSV_FILE: tfhe_rs_integer_benches_${{ github.sha }}.csv
+  ACTION_RUN_URL: ${{ github.server_url }}/${{ github.repository }}/actions/runs/${{ github.run_id }}
+
+jobs:
+  run-integer-benchmarks:
+    name: Execute signed integer multi-bit benchmarks in EC2
+    runs-on: ${{ github.event.inputs.runner_name }}
+    if: ${{ !cancelled() }}
+    steps:
+      - name: Instance configuration used
+        run: |
+          echo "IDs: ${{ inputs.instance_id }}"
+          echo "AMI: ${{ inputs.instance_image_id }}"
+          echo "Type: ${{ inputs.instance_type }}"
+          echo "Request ID: ${{ inputs.request_id }}"
+
+      - name: Get benchmark date
+        run: |
+          echo "BENCH_DATE=$(date --iso-8601=seconds)" >> "${GITHUB_ENV}"
+
+      - name: Checkout tfhe-rs repo with tags
+        uses: actions/checkout@b4ffde65f46336ab88eb53be808477a3936bae11
+        with:
+          fetch-depth: 0
+
+      - name: Set up home
+        # "Install rust" step require root user to have a HOME directory which is not set.
+        run: |
+          echo "HOME=/home/ubuntu" >> "${GITHUB_ENV}"
+
+      - name: Install rust
+        uses: actions-rs/toolchain@16499b5e05bf2e26879000db0c1d13f7e13fa3af
+        with:
+          toolchain: nightly
+          override: true
+
+      - name: Run multi-bit benchmarks with AVX512
+        run: |
+          make AVX512_SUPPORT=ON FAST_BENCH=TRUE bench_signed_integer_multi_bit
+
+      - name: Parse benchmarks to csv
+        run: |
+          make PARSE_INTEGER_BENCH_CSV_FILE=${{ env.PARSE_INTEGER_BENCH_CSV_FILE }} \
+            parse_integer_benches
+
+      - name: Upload csv results artifact
+        uses: actions/upload-artifact@a8a3f3ad30e3422c9c7b888a15615d19a852ae32
+        with:
+          name: ${{ github.sha }}_csv_integer
+          path: ${{ env.PARSE_INTEGER_BENCH_CSV_FILE }}
+
+      - name: Parse results
+        run: |
+          COMMIT_DATE="$(git --no-pager show -s --format=%cd --date=iso8601-strict ${{ github.sha }})"
+          COMMIT_HASH="$(git describe --tags --dirty)"
+          python3 ./ci/benchmark_parser.py target/criterion ${{ env.RESULTS_FILENAME }} \
+          --database tfhe_rs \
+          --hardware ${{ inputs.instance_type }} \
+          --project-version "${COMMIT_HASH}" \
+          --branch ${{ github.ref_name }} \
+          --commit-date "${COMMIT_DATE}" \
+          --bench-date "${{ env.BENCH_DATE }}" \
+          --walk-subdirs \
+          --name-suffix avx512 \
+          --throughput
+
+      - name: Upload parsed results artifact
+        uses: actions/upload-artifact@a8a3f3ad30e3422c9c7b888a15615d19a852ae32
+        with:
+          name: ${{ github.sha }}_integer
+          path: ${{ env.RESULTS_FILENAME }}
+
+      - name: Checkout Slab repo
+        uses: actions/checkout@b4ffde65f46336ab88eb53be808477a3936bae11
+        with:
+          repository: zama-ai/slab
+          path: slab
+          token: ${{ secrets.CONCRETE_ACTIONS_TOKEN }}
+
+      - name: Send data to Slab
+        shell: bash
+        run: |
+          echo "Computing HMac on results file"
+          SIGNATURE="$(slab/scripts/hmac_calculator.sh ${{ env.RESULTS_FILENAME }} '${{ secrets.JOB_SECRET }}')"
+          echo "Sending results to Slab..."
+          curl -v -k \
+          -H "Content-Type: application/json" \
+          -H "X-Slab-Repository: ${{ github.repository }}" \
+          -H "X-Slab-Command: store_data_v2" \
+          -H "X-Hub-Signature-256: sha256=${SIGNATURE}" \
+          -d @${{ env.RESULTS_FILENAME }} \
+          ${{ secrets.SLAB_URL }}
+
+      - name: Slack Notification
+        if: ${{ failure() }}
+        continue-on-error: true
+        uses: rtCamp/action-slack-notify@b24d75fe0e728a4bf9fc42ee217caa686d141ee8
+        env:
+          SLACK_COLOR: ${{ job.status }}
+          SLACK_CHANNEL: ${{ secrets.SLACK_CHANNEL }}
+          SLACK_ICON: https://pbs.twimg.com/profile_images/1274014582265298945/OjBKP9kn_400x400.png
+          SLACK_MESSAGE: "Signed integer benchmarks failed. (${{ env.ACTION_RUN_URL }})"
+          SLACK_USERNAME: ${{ secrets.BOT_USERNAME }}
+          SLACK_WEBHOOK: ${{ secrets.SLACK_WEBHOOK }}

.github/workflows/start_benchmarks.yml

@@ -20,10 +20,18 @@ on:
         description: "Run integer benches"
         type: boolean
         default: true
+      signed_integer_bench:
+        description: "Run signed integer benches"
+        type: boolean
+        default: true
       integer_multi_bit_bench:
         description: "Run integer multi bit benches"
         type: boolean
         default: true
+      signed_integer_multi_bit_bench:
+        description: "Run signed integer multi bit benches"
+        type: boolean
+        default: true
       pbs_bench:
         description: "Run PBS benches"
         type: boolean
@@ -38,17 +46,20 @@ jobs:
     if: ${{ (github.event_name == 'push' && github.repository == 'zama-ai/tfhe-rs') || github.event_name == 'workflow_dispatch' }}
     strategy:
       matrix:
-        command: [boolean_bench, shortint_bench, integer_bench, integer_multi_bit_bench, pbs_bench, wasm_client_bench]
+        command: [ boolean_bench, shortint_bench,
+                   integer_bench, integer_multi_bit_bench,
+                   signed_integer_bench, signed_integer_multi_bit_bench,
+                   pbs_bench, wasm_client_bench ]
     runs-on: ubuntu-latest
     steps:
       - name: Checkout tfhe-rs
-        uses: actions/checkout@f43a0e5ff2bd294095638e18286ca9a3d1956744
+        uses: actions/checkout@b4ffde65f46336ab88eb53be808477a3936bae11
         with:
           fetch-depth: 0
 
       - name: Check for file changes
         id: changed-files
-        uses: tj-actions/changed-files@2f7246cb26e8bb6709b6cbfc1fec7febfe82e96a
+        uses: tj-actions/changed-files@1c938490c880156b746568a518594309cfb3f66b
         with:
           files_yaml: |
             common_benches:
@@ -69,13 +80,23 @@ jobs:
             integer_bench:
               - tfhe/src/shortint/**
               - tfhe/src/integer/**
-              - tfhe/benches/integer/**
+              - tfhe/benches/integer/bench.rs
               - .github/workflows/integer_benchmark.yml
             integer_multi_bit_bench:
               - tfhe/src/shortint/**
               - tfhe/src/integer/**
-              - tfhe/benches/integer/**
-              - .github/workflows/integer_benchmark.yml
+              - tfhe/benches/integer/bench.rs
+              - .github/workflows/integer_multi_bit_benchmark.yml
+            signed_integer_bench:
+              - tfhe/src/shortint/**
+              - tfhe/src/integer/**
+              - tfhe/benches/integer/signed_bench.rs
+              - .github/workflows/signed_integer_benchmark.yml
+            signed_integer_multi_bit_bench:
+              - tfhe/src/shortint/**
+              - tfhe/src/integer/**
+              - tfhe/benches/integer/signed_bench.rs
+              - .github/workflows/signed_integer_multi_bit_benchmark.yml
             pbs_bench:
               - tfhe/src/core_crypto/**
               - tfhe/benches/core_crypto/**
@@ -85,7 +106,7 @@ jobs:
               - .github/workflows/wasm_client_benchmark.yml
 
       - name: Checkout Slab repo
-        uses: actions/checkout@f43a0e5ff2bd294095638e18286ca9a3d1956744
+        uses: actions/checkout@b4ffde65f46336ab88eb53be808477a3936bae11
         with:
           repository: zama-ai/slab
           path: slab

.github/workflows/start_full_benchmarks.yml

@@ -3,34 +3,57 @@ name: Start full suite benchmarks
 
 on:
   schedule:
-    # Job will be triggered each Saturday at 1a.m.
+    # Weekly benchmarks will be triggered each Saturday at 1a.m.
     - cron: '0 1 * * 6'
+    # Quarterly benchmarks will be triggered right before end of quarter, the 25th of the current month at 4a.m.
+    # These benchmarks are far longer to execute hence the reason to run them only four time a year.
+    - cron: '0 4 25 MAR,JUN,SEP,DEC *'
   workflow_dispatch:
+    inputs:
+      benchmark_type:
+        description: 'Benchmark type'
+        required: true
+        default: 'weekly'
+        type: choice
+        options:
+          - weekly
+          - quarterly
 
 jobs:
   start-benchmarks:
     if: ${{ (github.event_name == 'schedule' && github.repository == 'zama-ai/tfhe-rs') || github.event_name == 'workflow_dispatch' }}
     strategy:
       matrix:
-        command: [ boolean_bench, shortint_full_bench, integer_full_bench, pbs_bench, wasm_client_bench ]
+        command: [ boolean_bench, shortint_full_bench, integer_full_bench,
+                   signed_integer_full_bench, pbs_bench, wasm_client_bench ]
     runs-on: ubuntu-latest
     steps:
       - name: Checkout tfhe-rs
-        uses: actions/checkout@f43a0e5ff2bd294095638e18286ca9a3d1956744
+        uses: actions/checkout@b4ffde65f46336ab88eb53be808477a3936bae11
         with:
           fetch-depth: 0
 
       - name: Checkout Slab repo
-        uses: actions/checkout@f43a0e5ff2bd294095638e18286ca9a3d1956744
+        uses: actions/checkout@b4ffde65f46336ab88eb53be808477a3936bae11
         with:
           repository: zama-ai/slab
           path: slab
           token: ${{ secrets.CONCRETE_ACTIONS_TOKEN }}
 
+      - name: Set benchmarks type as weekly
+        if: (github.event_name == 'workflow_dispatch' && inputs.benchmark_type == 'weekly') || github.event.schedule == '0 1 * * 6'
+        run: |
+          echo "BENCH_TYPE=weekly_benchmarks" >> "${GITHUB_ENV}"
+
+      - name: Set benchmarks type as quarterly
+        if: (github.event_name == 'workflow_dispatch' && inputs.benchmark_type == 'quarterly') || github.event.schedule == '0 4 25 MAR,JUN,SEP,DEC *'
+        run: |
+          echo "BENCH_TYPE=quarterly_benchmarks" >> "${GITHUB_ENV}"
+
       - name: Start AWS job in Slab
         shell: bash
         run: |
-          echo -n '{"command": "${{ matrix.command }}", "git_ref": "${{ github.ref }}", "sha": "${{ github.sha }}"}' > command.json
+          echo -n '{"command": "${{ matrix.command }}", "git_ref": "${{ github.ref }}", "sha": "${{ github.sha }}", "user_inputs": "${{ env.BENCH_TYPE }}"}' > command.json
           SIGNATURE="$(slab/scripts/hmac_calculator.sh command.json '${{ secrets.JOB_SECRET }}')"
           curl -v -k \
           --fail-with-body \

.github/workflows/sync_on_push.yml

@@ -13,11 +13,11 @@ jobs:
     runs-on: ubuntu-latest
     steps:
       - name: Checkout repo
-        uses: actions/checkout@f43a0e5ff2bd294095638e18286ca9a3d1956744
+        uses: actions/checkout@b4ffde65f46336ab88eb53be808477a3936bae11
         with:
           fetch-depth: 0
       - name: Save repo
-        uses: actions/upload-artifact@0b7f8abb1508181956e8e162db84b466c27e18ce
+        uses: actions/upload-artifact@a8a3f3ad30e3422c9c7b888a15615d19a852ae32
         with:
           name: repo-archive
           path: '.'

.github/workflows/trigger_aws_tests_on_pr.yml

@@ -12,6 +12,16 @@ jobs:
     permissions:
       pull-requests: write
     steps:
+      - name: Get current labels
+        uses: snnaplab/get-labels-action@f426df40304808ace3b5282d4f036515f7609576
+
+      - name: Remove approved label
+        if: ${{ github.event_name == 'pull_request' && contains(fromJSON(env.LABELS), 'approved') }}
+        uses: actions-ecosystem/action-remove-labels@2ce5d41b4b6aa8503e285553f75ed56e0a40bae0
+        with:
+          github_token: ${{ secrets.GITHUB_TOKEN }}
+          labels: approved
+
       - name: Launch fast tests
         if: ${{ github.event_name == 'pull_request' }}
         uses: mshick/add-pr-comment@a65df5f64fc741e91c59b8359a4bc56e57aaf5b1
@@ -20,8 +30,17 @@ jobs:
           message: |
             @slab-ci cpu_fast_test
 
+      - name: Add approved label
+        uses: actions-ecosystem/action-add-labels@18f1af5e3544586314bbe15c0273249c770b2daf
+        if: ${{ github.event_name == 'pull_request_review' && github.event.review.state == 'approved' && !contains(fromJSON(env.LABELS), 'approved') }}
+        with:
+          github_token: ${{ secrets.GITHUB_TOKEN }}
+          labels: approved
+
+      # PR label 'approved' presence is checked to avoid running the full test suite several times
+      # in case of multiple approvals without new commits in between.
       - name: Launch full tests suite
-        if: ${{ github.event_name == 'pull_request_review' && github.event.review.state == 'approved' }}
+        if: ${{ github.event_name == 'pull_request_review' && github.event.review.state == 'approved' && !contains(fromJSON(env.LABELS), 'approved') }}
         uses: mshick/add-pr-comment@a65df5f64fc741e91c59b8359a4bc56e57aaf5b1
         with:
           allow-repeats: true
@@ -29,6 +48,7 @@ jobs:
             Pull Request has been approved :tada:
             Launching full test suite...
             @slab-ci cpu_test
-            @slab-ci cpu_integer_test
-            @slab-ci cpu_multi_bit_test
+            @slab-ci cpu_unsigned_integer_test
+            @slab-ci cpu_signed_integer_test
             @slab-ci cpu_wasm_test
+            @slab-ci csprng_randomness_testing

.github/workflows/wasm_client_benchmark.yml

@@ -19,6 +19,14 @@ on:
       request_id:
         description: "Slab request ID"
         type: string
+      # This input is not used in this workflow but still mandatory since a calling workflow could
+      # use it. If a triggering command include a user_inputs field, then the triggered workflow
+      # must include this very input, otherwise the workflow won't be called.
+      # See start_full_benchmarks.yml as example.
+      user_inputs:
+        description: "Type of benchmarks to run"
+        type: string
+        default: "weekly_benchmarks"
 
 env:
   CARGO_TERM_COLOR: always
@@ -43,7 +51,7 @@ jobs:
           echo "BENCH_DATE=$(date --iso-8601=seconds)" >> "${GITHUB_ENV}"
 
       - name: Checkout tfhe-rs repo with tags
-        uses: actions/checkout@f43a0e5ff2bd294095638e18286ca9a3d1956744
+        uses: actions/checkout@b4ffde65f46336ab88eb53be808477a3936bae11
         with:
           fetch-depth: 0
 
@@ -89,13 +97,13 @@ jobs:
           --append-results
 
       - name: Upload parsed results artifact
-        uses: actions/upload-artifact@0b7f8abb1508181956e8e162db84b466c27e18ce
+        uses: actions/upload-artifact@a8a3f3ad30e3422c9c7b888a15615d19a852ae32
         with:
           name: ${{ github.sha }}_wasm
           path: ${{ env.RESULTS_FILENAME }}
 
       - name: Checkout Slab repo
-        uses: actions/checkout@f43a0e5ff2bd294095638e18286ca9a3d1956744
+        uses: actions/checkout@b4ffde65f46336ab88eb53be808477a3936bae11
         with:
           repository: zama-ai/slab
           path: slab

.gitignore

@@ -3,9 +3,9 @@ target/
 .vscode/
 
 # Path we use for internal-keycache during tests
-./keys/
+/keys/
 # In case of symlinked keys
-./keys
+/keys
 
 **/Cargo.lock
 **/*.bin
@@ -13,3 +13,9 @@ target/
 # Some of our bench outputs
 /tfhe/benchmarks_parameters
 **/*.csv
+
+# dieharder run log
+dieharder_run.log
+
+# Coverage reports
+/coverage/

.linelint.yml

@@ -0,0 +1,14 @@
+ignore:
+  - .git
+  - target
+  - tfhe/benchmarks_parameters
+  - tfhe/web_wasm_parallel_tests/node_modules
+  - tfhe/web_wasm_parallel_tests/dist
+  - keys
+  - coverage
+
+rules:
+  # checks if file ends in a newline character
+  end-of-file:
+    enable: true
+    single-new-line: true

Cargo.toml

@@ -1,6 +1,6 @@
 [workspace]
 resolver = "2"
-members = ["tfhe", "tasks", "apps/trivium"]
+members = ["tfhe", "tasks", "apps/trivium", "concrete-csprng"]
 
 [profile.bench]
 lto = "fat"

Makefile

@@ -3,19 +3,30 @@ OS:=$(shell uname)
 RS_CHECK_TOOLCHAIN:=$(shell cat toolchain.txt | tr -d '\n')
 CARGO_RS_CHECK_TOOLCHAIN:=+$(RS_CHECK_TOOLCHAIN)
 TARGET_ARCH_FEATURE:=$(shell ./scripts/get_arch_feature.sh)
-RS_BUILD_TOOLCHAIN:=$(shell \
-	( (echo $(TARGET_ARCH_FEATURE) | grep -q x86) && echo stable) || echo $(RS_CHECK_TOOLCHAIN))
+RS_BUILD_TOOLCHAIN:=stable
 CARGO_RS_BUILD_TOOLCHAIN:=+$(RS_BUILD_TOOLCHAIN)
 CARGO_PROFILE?=release
-MIN_RUST_VERSION:=$(shell grep rust-version tfhe/Cargo.toml | cut -d '=' -f 2 | xargs)
+MIN_RUST_VERSION:=$(shell grep '^rust-version[[:space:]]*=' tfhe/Cargo.toml | cut -d '=' -f 2 | xargs)
 AVX512_SUPPORT?=OFF
 WASM_RUSTFLAGS:=
 BIG_TESTS_INSTANCE?=FALSE
 GEN_KEY_CACHE_MULTI_BIT_ONLY?=FALSE
+GEN_KEY_CACHE_COVERAGE_ONLY?=FALSE
 PARSE_INTEGER_BENCH_CSV_FILE?=tfhe_rs_integer_benches.csv
 FAST_TESTS?=FALSE
 FAST_BENCH?=FALSE
 BENCH_OP_FLAVOR?=DEFAULT
+NODE_VERSION=20
+# sed: -n, do not print input stream, -e means a script/expression
+# 1,/version/ indicates from the first line, to the line matching version at the start of the line
+# p indicates to print, so we keep only the start of the Cargo.toml until we hit the first version
+# entry which should be the version of tfhe
+TFHE_CURRENT_VERSION:=\
+$(shell sed -n -e '1,/^version/p' tfhe/Cargo.toml | \
+grep '^version[[:space:]]*=' | cut -d '=' -f 2 | xargs)
+# Cargo has a hard time distinguishing between our package from the workspace and a package that
+# could be a dependency, so we build an unambiguous spec here
+TFHE_SPEC:=tfhe@$(TFHE_CURRENT_VERSION)
 # This is done to avoid forgetting it, we still precise the RUSTFLAGS in the commands to be able to
 # copy paste the command in the terminal and change them if required without forgetting the flags
 export RUSTFLAGS?=-C target-cpu=native
@@ -32,10 +43,32 @@ else
 		MULTI_BIT_ONLY=
 endif
 
+ifeq ($(GEN_KEY_CACHE_COVERAGE_ONLY),TRUE)
+		COVERAGE_ONLY=--coverage-only
+else
+		COVERAGE_ONLY=
+endif
+
 # Variables used only for regex_engine example
 REGEX_STRING?=''
 REGEX_PATTERN?=''
 
+# Exclude these files from coverage reports
+define COVERAGE_EXCLUDED_FILES
+--exclude-files apps/trivium/src/trivium/* \
+--exclude-files apps/trivium/src/kreyvium/* \
+--exclude-files apps/trivium/src/static_deque/* \
+--exclude-files apps/trivium/src/trans_ciphering/* \
+--exclude-files tasks/src/* \
+--exclude-files tfhe/benches/boolean/* \
+--exclude-files tfhe/benches/core_crypto/* \
+--exclude-files tfhe/benches/shortint/* \
+--exclude-files tfhe/benches/integer/* \
+--exclude-files tfhe/benches/* \
+--exclude-files tfhe/examples/regex_engine/* \
+--exclude-files tfhe/examples/utilities/*
+endef
+
 .PHONY: rs_check_toolchain # Echo the rust toolchain used for checks
 rs_check_toolchain:
 	@echo $(RS_CHECK_TOOLCHAIN)
@@ -77,136 +110,169 @@ install_wasm_pack: install_rs_build_toolchain
 install_node:
 	curl -o- https://raw.githubusercontent.com/nvm-sh/nvm/v0.39.3/install.sh | $(SHELL)
 	source ~/.bashrc
-	$(SHELL) -i -c 'nvm install node' || \
+	$(SHELL) -i -c 'nvm install $(NODE_VERSION)' || \
 	( echo "Unable to install node, unknown error." && exit 1 )
 
+.PHONY: install_dieharder # Install dieharder for apt distributions or macOS
+install_dieharder:
+	@dieharder -h > /dev/null 2>&1 || \
+	if [[ "$(OS)" == "Linux" ]]; then \
+		sudo apt update && sudo apt install -y dieharder; \
+	elif [[ "$(OS)" == "Darwin" ]]; then\
+		brew install dieharder; \
+	fi || ( echo "Unable to install dieharder, unknown error." && exit 1 )
+
+.PHONY: install_tarpaulin # Install tarpaulin to perform code coverage
+install_tarpaulin: install_rs_build_toolchain
+	@cargo tarpaulin --version > /dev/null 2>&1 || \
+	cargo $(CARGO_RS_BUILD_TOOLCHAIN) install cargo-tarpaulin --locked || \
+	( echo "Unable to install cargo tarpaulin, unknown error." && exit 1 )
+
+.PHONY: check_linelint_installed # Check if linelint newline linter is installed
+check_linelint_installed:
+	@printf "\n" | linelint - > /dev/null 2>&1 || \
+	( echo "Unable to locate linelint. Try installing it: https://github.com/fernandrone/linelint/releases" && exit 1 )
+
 .PHONY: fmt # Format rust code
 fmt: install_rs_check_toolchain
 	cargo "$(CARGO_RS_CHECK_TOOLCHAIN)" fmt
 
-.PHONT: check_fmt # Check rust code format
+.PHONY: check_fmt # Check rust code format
 check_fmt: install_rs_check_toolchain
 	cargo "$(CARGO_RS_CHECK_TOOLCHAIN)" fmt --check
 
+.PHONY: fix_newline # Fix newline at end of file issues to be UNIX compliant
+fix_newline: check_linelint_installed
+	linelint -a .
+
+.PHONY: check_newline # Check for newline at end of file to be UNIX compliant
+check_newline: check_linelint_installed
+	linelint .
+
 .PHONY: clippy_core # Run clippy lints on core_crypto with and without experimental features
 clippy_core: install_rs_check_toolchain
 	RUSTFLAGS="$(RUSTFLAGS)" cargo "$(CARGO_RS_CHECK_TOOLCHAIN)" clippy \
 		--features=$(TARGET_ARCH_FEATURE) \
-		-p tfhe -- --no-deps -D warnings
+		-p $(TFHE_SPEC) -- --no-deps -D warnings
 	RUSTFLAGS="$(RUSTFLAGS)" cargo "$(CARGO_RS_CHECK_TOOLCHAIN)" clippy \
 		--features=$(TARGET_ARCH_FEATURE),experimental \
-		-p tfhe -- --no-deps -D warnings
+		-p $(TFHE_SPEC) -- --no-deps -D warnings
 
 .PHONY: clippy_boolean # Run clippy lints enabling the boolean features
 clippy_boolean: install_rs_check_toolchain
 	RUSTFLAGS="$(RUSTFLAGS)" cargo "$(CARGO_RS_CHECK_TOOLCHAIN)" clippy \
 		--features=$(TARGET_ARCH_FEATURE),boolean \
-		-p tfhe -- --no-deps -D warnings
+		-p $(TFHE_SPEC) -- --no-deps -D warnings
 
 .PHONY: clippy_shortint # Run clippy lints enabling the shortint features
 clippy_shortint: install_rs_check_toolchain
 	RUSTFLAGS="$(RUSTFLAGS)" cargo "$(CARGO_RS_CHECK_TOOLCHAIN)" clippy \
 		--features=$(TARGET_ARCH_FEATURE),shortint \
-		-p tfhe -- --no-deps -D warnings
+		-p $(TFHE_SPEC) -- --no-deps -D warnings
 
 .PHONY: clippy_integer # Run clippy lints enabling the integer features
 clippy_integer: install_rs_check_toolchain
 	RUSTFLAGS="$(RUSTFLAGS)" cargo "$(CARGO_RS_CHECK_TOOLCHAIN)" clippy \
 		--features=$(TARGET_ARCH_FEATURE),integer \
-		-p tfhe -- --no-deps -D warnings
+		-p $(TFHE_SPEC) -- --no-deps -D warnings
 
 .PHONY: clippy # Run clippy lints enabling the boolean, shortint, integer
 clippy: install_rs_check_toolchain
 	RUSTFLAGS="$(RUSTFLAGS)" cargo "$(CARGO_RS_CHECK_TOOLCHAIN)" clippy --all-targets \
 		--features=$(TARGET_ARCH_FEATURE),boolean,shortint,integer \
-		-p tfhe -- --no-deps -D warnings
+		-p $(TFHE_SPEC) -- --no-deps -D warnings
 
 .PHONY: clippy_c_api # Run clippy lints enabling the boolean, shortint and the C API
 clippy_c_api: install_rs_check_toolchain
 	RUSTFLAGS="$(RUSTFLAGS)" cargo "$(CARGO_RS_CHECK_TOOLCHAIN)" clippy \
 		--features=$(TARGET_ARCH_FEATURE),boolean-c-api,shortint-c-api \
-		-p tfhe -- --no-deps -D warnings
+		-p $(TFHE_SPEC) -- --no-deps -D warnings
 
 .PHONY: clippy_js_wasm_api # Run clippy lints enabling the boolean, shortint, integer and the js wasm API
 clippy_js_wasm_api: install_rs_check_toolchain
 	RUSTFLAGS="$(RUSTFLAGS)" cargo "$(CARGO_RS_CHECK_TOOLCHAIN)" clippy \
 		--features=boolean-client-js-wasm-api,shortint-client-js-wasm-api,integer-client-js-wasm-api \
-		-p tfhe -- --no-deps -D warnings
+		-p $(TFHE_SPEC) -- --no-deps -D warnings
 
 .PHONY: clippy_tasks # Run clippy lints on helper tasks crate.
 clippy_tasks:
 	RUSTFLAGS="$(RUSTFLAGS)" cargo "$(CARGO_RS_CHECK_TOOLCHAIN)" clippy \
 		-p tasks -- --no-deps -D warnings
 
+.PHONY: clippy_trivium # Run clippy lints on Trivium app
+clippy_trivium: install_rs_check_toolchain
+	RUSTFLAGS="$(RUSTFLAGS)" cargo "$(CARGO_RS_CHECK_TOOLCHAIN)" clippy \
+		-p tfhe-trivium -- --no-deps -D warnings
+
 .PHONY: clippy_all_targets # Run clippy lints on all targets (benches, examples, etc.)
 clippy_all_targets:
 	RUSTFLAGS="$(RUSTFLAGS)" cargo "$(CARGO_RS_CHECK_TOOLCHAIN)" clippy --all-targets \
-		--features=$(TARGET_ARCH_FEATURE),boolean,shortint,integer \
-		-p tfhe -- --no-deps -D warnings
+		--features=$(TARGET_ARCH_FEATURE),boolean,shortint,integer,internal-keycache,safe-deserialization \
+		-p $(TFHE_SPEC) -- --no-deps -D warnings
+
+.PHONY: clippy_concrete_csprng # Run clippy lints on concrete-csprng
+clippy_concrete_csprng:
+	RUSTFLAGS="$(RUSTFLAGS)" cargo "$(CARGO_RS_CHECK_TOOLCHAIN)" clippy --all-targets \
+		--features=$(TARGET_ARCH_FEATURE) \
+		-p concrete-csprng -- --no-deps -D warnings
 
 .PHONY: clippy_all # Run all clippy targets
 clippy_all: clippy clippy_boolean clippy_shortint clippy_integer clippy_all_targets clippy_c_api \
-clippy_js_wasm_api clippy_tasks clippy_core
+clippy_js_wasm_api clippy_tasks clippy_core clippy_concrete_csprng clippy_trivium
 
 .PHONY: clippy_fast # Run main clippy targets
-clippy_fast: clippy clippy_all_targets clippy_c_api clippy_js_wasm_api clippy_tasks clippy_core
-
-.PHONY: gen_key_cache # Run the script to generate keys and cache them for shortint tests
-gen_key_cache: install_rs_build_toolchain
-	RUSTFLAGS="$(RUSTFLAGS)" cargo $(CARGO_RS_BUILD_TOOLCHAIN) run --profile $(CARGO_PROFILE) \
-		--example generates_test_keys \
-		--features=$(TARGET_ARCH_FEATURE),shortint,internal-keycache -p tfhe -- \
-		$(MULTI_BIT_ONLY)
+clippy_fast: clippy clippy_all_targets clippy_c_api clippy_js_wasm_api clippy_tasks clippy_core \
+clippy_concrete_csprng
 
 .PHONY: build_core # Build core_crypto without experimental features
 build_core: install_rs_build_toolchain install_rs_check_toolchain
 	RUSTFLAGS="$(RUSTFLAGS)" cargo $(CARGO_RS_BUILD_TOOLCHAIN) build --profile $(CARGO_PROFILE) \
-		--features=$(TARGET_ARCH_FEATURE) -p tfhe
+		--features=$(TARGET_ARCH_FEATURE) -p $(TFHE_SPEC)
 	@if [[ "$(AVX512_SUPPORT)" == "ON" ]]; then \
 		RUSTFLAGS="$(RUSTFLAGS)" cargo $(CARGO_RS_CHECK_TOOLCHAIN) build --profile $(CARGO_PROFILE) \
-			--features=$(TARGET_ARCH_FEATURE),$(AVX512_FEATURE) -p tfhe; \
+			--features=$(TARGET_ARCH_FEATURE),$(AVX512_FEATURE) -p $(TFHE_SPEC); \
 	fi
 
 .PHONY: build_core_experimental # Build core_crypto with experimental features
 build_core_experimental: install_rs_build_toolchain install_rs_check_toolchain
 	RUSTFLAGS="$(RUSTFLAGS)" cargo $(CARGO_RS_BUILD_TOOLCHAIN) build --profile $(CARGO_PROFILE) \
-		--features=$(TARGET_ARCH_FEATURE),experimental -p tfhe
+		--features=$(TARGET_ARCH_FEATURE),experimental -p $(TFHE_SPEC)
 	@if [[ "$(AVX512_SUPPORT)" == "ON" ]]; then \
 		RUSTFLAGS="$(RUSTFLAGS)" cargo $(CARGO_RS_CHECK_TOOLCHAIN) build --profile $(CARGO_PROFILE) \
-			--features=$(TARGET_ARCH_FEATURE),experimental,$(AVX512_FEATURE) -p tfhe; \
+			--features=$(TARGET_ARCH_FEATURE),experimental,$(AVX512_FEATURE) -p $(TFHE_SPEC); \
 	fi
 
 .PHONY: build_boolean # Build with boolean enabled
 build_boolean: install_rs_build_toolchain
 	RUSTFLAGS="$(RUSTFLAGS)" cargo $(CARGO_RS_BUILD_TOOLCHAIN) build --profile $(CARGO_PROFILE) \
-		--features=$(TARGET_ARCH_FEATURE),boolean -p tfhe --all-targets
+		--features=$(TARGET_ARCH_FEATURE),boolean -p $(TFHE_SPEC) --all-targets
 
 .PHONY: build_shortint # Build with shortint enabled
 build_shortint: install_rs_build_toolchain
 	RUSTFLAGS="$(RUSTFLAGS)" cargo $(CARGO_RS_BUILD_TOOLCHAIN) build --profile $(CARGO_PROFILE) \
-		--features=$(TARGET_ARCH_FEATURE),shortint -p tfhe --all-targets
+		--features=$(TARGET_ARCH_FEATURE),shortint -p $(TFHE_SPEC) --all-targets
 
 .PHONY: build_integer # Build with integer enabled
 build_integer: install_rs_build_toolchain
 	RUSTFLAGS="$(RUSTFLAGS)" cargo $(CARGO_RS_BUILD_TOOLCHAIN) build --profile $(CARGO_PROFILE) \
-		--features=$(TARGET_ARCH_FEATURE),integer -p tfhe --all-targets
+		--features=$(TARGET_ARCH_FEATURE),integer -p $(TFHE_SPEC) --all-targets
 
 .PHONY: build_tfhe_full # Build with boolean, shortint and integer enabled
 build_tfhe_full: install_rs_build_toolchain
 	RUSTFLAGS="$(RUSTFLAGS)" cargo $(CARGO_RS_BUILD_TOOLCHAIN) build --profile $(CARGO_PROFILE) \
-		--features=$(TARGET_ARCH_FEATURE),boolean,shortint,integer -p tfhe --all-targets
+		--features=$(TARGET_ARCH_FEATURE),boolean,shortint,integer -p $(TFHE_SPEC) --all-targets
 
 .PHONY: build_c_api # Build the C API for boolean, shortint and integer
 build_c_api: install_rs_check_toolchain
 	RUSTFLAGS="$(RUSTFLAGS)" cargo $(CARGO_RS_CHECK_TOOLCHAIN) build --profile $(CARGO_PROFILE) \
-		--features=$(TARGET_ARCH_FEATURE),boolean-c-api,shortint-c-api,high-level-c-api, \
-		-p tfhe
+		--features=$(TARGET_ARCH_FEATURE),boolean-c-api,shortint-c-api,high-level-c-api,safe-deserialization \
+		-p $(TFHE_SPEC)
 
 .PHONY: build_c_api_experimental_deterministic_fft # Build the C API for boolean, shortint and integer with experimental deterministic FFT
 build_c_api_experimental_deterministic_fft: install_rs_check_toolchain
 	RUSTFLAGS="$(RUSTFLAGS)" cargo $(CARGO_RS_CHECK_TOOLCHAIN) build --profile $(CARGO_PROFILE) \
-		--features=$(TARGET_ARCH_FEATURE),boolean-c-api,shortint-c-api,high-level-c-api,experimental-force_fft_algo_dif4 \
-		-p tfhe
+		--features=$(TARGET_ARCH_FEATURE),boolean-c-api,shortint-c-api,high-level-c-api,safe-deserialization,experimental-force_fft_algo_dif4 \
+		-p $(TFHE_SPEC)
 
 .PHONY: build_web_js_api # Build the js API targeting the web browser
 build_web_js_api: install_rs_build_toolchain install_wasm_pack
@@ -231,25 +297,53 @@ build_node_js_api: install_rs_build_toolchain install_wasm_pack
 		wasm-pack build --release --target=nodejs \
 		-- --features=boolean-client-js-wasm-api,shortint-client-js-wasm-api,integer-client-js-wasm-api
 
+.PHONY: build_concrete_csprng # Build concrete_csprng
+build_concrete_csprng: install_rs_build_toolchain
+	RUSTFLAGS="$(RUSTFLAGS)" cargo $(CARGO_RS_BUILD_TOOLCHAIN) build --profile $(CARGO_PROFILE) \
+		--features=$(TARGET_ARCH_FEATURE) -p concrete-csprng --all-targets
+
 .PHONY: test_core_crypto # Run the tests of the core_crypto module including experimental ones
 test_core_crypto: install_rs_build_toolchain install_rs_check_toolchain
 	RUSTFLAGS="$(RUSTFLAGS)" cargo $(CARGO_RS_BUILD_TOOLCHAIN) test --profile $(CARGO_PROFILE) \
-		--features=$(TARGET_ARCH_FEATURE),experimental -p tfhe -- core_crypto::
+		--features=$(TARGET_ARCH_FEATURE),experimental -p $(TFHE_SPEC) -- core_crypto::
 	@if [[ "$(AVX512_SUPPORT)" == "ON" ]]; then \
 		RUSTFLAGS="$(RUSTFLAGS)" cargo $(CARGO_RS_CHECK_TOOLCHAIN) test --profile $(CARGO_PROFILE) \
-			--features=$(TARGET_ARCH_FEATURE),experimental,$(AVX512_FEATURE) -p tfhe -- core_crypto::; \
+			--features=$(TARGET_ARCH_FEATURE),experimental,$(AVX512_FEATURE) -p $(TFHE_SPEC) -- core_crypto::; \
+	fi
+
+.PHONY: test_core_crypto_cov # Run the tests of the core_crypto module with code coverage
+test_core_crypto_cov: install_rs_build_toolchain install_rs_check_toolchain install_tarpaulin
+	RUSTFLAGS="$(RUSTFLAGS)" cargo $(CARGO_RS_BUILD_TOOLCHAIN) tarpaulin --profile $(CARGO_PROFILE) \
+		--out xml --ignore-panics --output-dir coverage/core_crypto --line --engine llvm --timeout 500 \
+		--implicit-test-threads $(COVERAGE_EXCLUDED_FILES) \
+		--features=$(TARGET_ARCH_FEATURE),experimental,internal-keycache,__coverage \
+		-p $(TFHE_SPEC) -- core_crypto::
+	@if [[ "$(AVX512_SUPPORT)" == "ON" ]]; then \
+		RUSTFLAGS="$(RUSTFLAGS)" cargo $(CARGO_RS_CHECK_TOOLCHAIN) tarpaulin --profile $(CARGO_PROFILE) \
+			--out xml --ignore-panics --output-dir coverage/core_crypto_avx512 --line --engine llvm --timeout 500 \
+			--implicit-test-threads $(COVERAGE_EXCLUDED_FILES) \
+			--features=$(TARGET_ARCH_FEATURE),experimental,internal-keycache,__coverage,$(AVX512_FEATURE) \
+			-p $(TFHE_SPEC) -- core_crypto::; \
 	fi
 
 .PHONY: test_boolean # Run the tests of the boolean module
 test_boolean: install_rs_build_toolchain
 	RUSTFLAGS="$(RUSTFLAGS)" cargo $(CARGO_RS_BUILD_TOOLCHAIN) test --profile $(CARGO_PROFILE) \
-		--features=$(TARGET_ARCH_FEATURE),boolean -p tfhe -- boolean::
+		--features=$(TARGET_ARCH_FEATURE),boolean -p $(TFHE_SPEC) -- boolean::
+
+.PHONY: test_boolean_cov # Run the tests of the boolean module with code coverage
+test_boolean_cov: install_rs_check_toolchain install_tarpaulin
+	RUSTFLAGS="$(RUSTFLAGS)" cargo $(CARGO_RS_CHECK_TOOLCHAIN) tarpaulin --profile $(CARGO_PROFILE) \
+		--out xml --ignore-panics --output-dir coverage/boolean --line --engine llvm --timeout 500 \
+		$(COVERAGE_EXCLUDED_FILES) \
+		--features=$(TARGET_ARCH_FEATURE),boolean,internal-keycache,__coverage \
+		-p $(TFHE_SPEC) -- boolean::
 
 .PHONY: test_c_api_rs # Run the rust tests for the C API
 test_c_api_rs: install_rs_check_toolchain
 	RUSTFLAGS="$(RUSTFLAGS)" cargo $(CARGO_RS_CHECK_TOOLCHAIN) test --profile $(CARGO_PROFILE) \
-		--features=$(TARGET_ARCH_FEATURE),boolean-c-api,shortint-c-api,high-level-c-api \
-		-p tfhe \
+		--features=$(TARGET_ARCH_FEATURE),boolean-c-api,shortint-c-api,high-level-c-api,safe-deserialization \
+		-p $(TFHE_SPEC) \
 		c_api
 
 .PHONY: test_c_api_c # Run the C tests for the C API
@@ -276,37 +370,82 @@ test_shortint_multi_bit_ci: install_rs_build_toolchain install_cargo_nextest
 .PHONY: test_shortint # Run all the tests for shortint
 test_shortint: install_rs_build_toolchain
 	RUSTFLAGS="$(RUSTFLAGS)" cargo $(CARGO_RS_BUILD_TOOLCHAIN) test --profile $(CARGO_PROFILE) \
-		--features=$(TARGET_ARCH_FEATURE),shortint,internal-keycache -p tfhe -- shortint::
+		--features=$(TARGET_ARCH_FEATURE),shortint,internal-keycache -p $(TFHE_SPEC) -- shortint::
+
+.PHONY: test_shortint_cov # Run the tests of the shortint module with code coverage
+test_shortint_cov: install_rs_check_toolchain install_tarpaulin
+	RUSTFLAGS="$(RUSTFLAGS)" cargo $(CARGO_RS_CHECK_TOOLCHAIN) tarpaulin --profile $(CARGO_PROFILE) \
+		--out xml --ignore-panics --output-dir coverage/shortint --line --engine llvm --timeout 500 \
+		$(COVERAGE_EXCLUDED_FILES) \
+		--features=$(TARGET_ARCH_FEATURE),shortint,internal-keycache,__coverage \
+		-p $(TFHE_SPEC) -- shortint::
 
 .PHONY: test_integer_ci # Run the tests for integer ci
-test_integer_ci: install_rs_build_toolchain install_cargo_nextest
+test_integer_ci: install_rs_check_toolchain install_cargo_nextest
 	BIG_TESTS_INSTANCE="$(BIG_TESTS_INSTANCE)" \
 	FAST_TESTS="$(FAST_TESTS)" \
-		./scripts/integer-tests.sh --rust-toolchain $(CARGO_RS_BUILD_TOOLCHAIN) \
-		--cargo-profile "$(CARGO_PROFILE)"
+		./scripts/integer-tests.sh --rust-toolchain $(CARGO_RS_CHECK_TOOLCHAIN) \
+		--cargo-profile "$(CARGO_PROFILE)" --avx512-support "$(AVX512_SUPPORT)"
+
+.PHONY: test_unsigned_integer_ci # Run the tests for unsigned integer ci
+test_unsigned_integer_ci: install_rs_check_toolchain install_cargo_nextest
+	BIG_TESTS_INSTANCE="$(BIG_TESTS_INSTANCE)" \
+	FAST_TESTS="$(FAST_TESTS)" \
+		./scripts/integer-tests.sh --rust-toolchain $(CARGO_RS_CHECK_TOOLCHAIN) \
+		--cargo-profile "$(CARGO_PROFILE)" --avx512-support "$(AVX512_SUPPORT)" \
+		--unsigned-only
+
+.PHONY: test_signed_integer_ci # Run the tests for signed integer ci
+test_signed_integer_ci: install_rs_check_toolchain install_cargo_nextest
+	BIG_TESTS_INSTANCE="$(BIG_TESTS_INSTANCE)" \
+	FAST_TESTS="$(FAST_TESTS)" \
+		./scripts/integer-tests.sh --rust-toolchain $(CARGO_RS_CHECK_TOOLCHAIN) \
+		--cargo-profile "$(CARGO_PROFILE)" --avx512-support "$(AVX512_SUPPORT)" \
+		--signed-only
 
 .PHONY: test_integer_multi_bit_ci # Run the tests for integer ci running only multibit tests
-test_integer_multi_bit_ci: install_rs_build_toolchain install_cargo_nextest
+test_integer_multi_bit_ci: install_rs_check_toolchain install_cargo_nextest
 	BIG_TESTS_INSTANCE="$(BIG_TESTS_INSTANCE)" \
 	FAST_TESTS="$(FAST_TESTS)" \
-		./scripts/integer-tests.sh --rust-toolchain $(CARGO_RS_BUILD_TOOLCHAIN) \
-		--cargo-profile "$(CARGO_PROFILE)" --multi-bit
+		./scripts/integer-tests.sh --rust-toolchain $(CARGO_RS_CHECK_TOOLCHAIN) \
+		--cargo-profile "$(CARGO_PROFILE)" --multi-bit --avx512-support "$(AVX512_SUPPORT)"
+
+.PHONY: test_unsigned_integer_multi_bit_ci # Run the tests for nsigned integer ci running only multibit tests
+test_unsigned_integer_multi_bit_ci: install_rs_check_toolchain install_cargo_nextest
+	BIG_TESTS_INSTANCE="$(BIG_TESTS_INSTANCE)" \
+	FAST_TESTS="$(FAST_TESTS)" \
+		./scripts/integer-tests.sh --rust-toolchain $(CARGO_RS_CHECK_TOOLCHAIN) \
+		--cargo-profile "$(CARGO_PROFILE)" --multi-bit --avx512-support "$(AVX512_SUPPORT)" \
+		--unsigned-only
+
+.PHONY: test_signed_integer_multi_bit_ci # Run the tests for nsigned integer ci running only multibit tests
+test_signed_integer_multi_bit_ci: install_rs_check_toolchain install_cargo_nextest
+	BIG_TESTS_INSTANCE="$(BIG_TESTS_INSTANCE)" \
+	FAST_TESTS="$(FAST_TESTS)" \
+		./scripts/integer-tests.sh --rust-toolchain $(CARGO_RS_CHECK_TOOLCHAIN) \
+		--cargo-profile "$(CARGO_PROFILE)" --multi-bit --avx512-support "$(AVX512_SUPPORT)" \
+		--signed-only
+
+.PHONY: test_safe_deserialization # Run the tests for safe deserialization
+test_safe_deserialization: install_rs_build_toolchain install_cargo_nextest
+	RUSTFLAGS="$(RUSTFLAGS)" cargo $(CARGO_RS_BUILD_TOOLCHAIN) test --profile $(CARGO_PROFILE) \
+		--features=$(TARGET_ARCH_FEATURE),boolean,shortint,integer,internal-keycache,safe-deserialization -p $(TFHE_SPEC) -- safe_deserialization::
 
 .PHONY: test_integer # Run all the tests for integer
 test_integer: install_rs_build_toolchain
 	RUSTFLAGS="$(RUSTFLAGS)" cargo $(CARGO_RS_BUILD_TOOLCHAIN) test --profile $(CARGO_PROFILE) \
-		--features=$(TARGET_ARCH_FEATURE),integer,internal-keycache -p tfhe -- integer::
+		--features=$(TARGET_ARCH_FEATURE),integer,internal-keycache -p $(TFHE_SPEC) -- integer::
 
 .PHONY: test_high_level_api # Run all the tests for high_level_api
 test_high_level_api: install_rs_build_toolchain
 	RUSTFLAGS="$(RUSTFLAGS)" cargo $(CARGO_RS_BUILD_TOOLCHAIN) test --profile $(CARGO_PROFILE) \
-		--features=$(TARGET_ARCH_FEATURE),boolean,shortint,integer,internal-keycache -p tfhe \
+		--features=$(TARGET_ARCH_FEATURE),boolean,shortint,integer,internal-keycache -p $(TFHE_SPEC) \
 		-- high_level_api::
 
 .PHONY: test_user_doc # Run tests from the .md documentation
 test_user_doc: install_rs_build_toolchain
 	RUSTFLAGS="$(RUSTFLAGS)" cargo $(CARGO_RS_BUILD_TOOLCHAIN) test --profile $(CARGO_PROFILE) --doc \
-		--features=$(TARGET_ARCH_FEATURE),boolean,shortint,integer,internal-keycache -p tfhe \
+		--features=$(TARGET_ARCH_FEATURE),boolean,shortint,integer,internal-keycache -p $(TFHE_SPEC) \
 		-- test_user_docs::
 
 .PHONY: test_regex_engine # Run tests for regex_engine example
@@ -327,14 +466,17 @@ test_examples: test_sha256_bool test_regex_engine
 .PHONY: test_trivium # Run tests for trivium
 test_trivium: install_rs_build_toolchain
 	RUSTFLAGS="$(RUSTFLAGS)" cargo $(CARGO_RS_BUILD_TOOLCHAIN) test --profile $(CARGO_PROFILE) \
-		trivium --features=$(TARGET_ARCH_FEATURE),boolean,shortint,integer \
-		-- --test-threads=1
+		-p tfhe-trivium -- --test-threads=1 trivium::
 
 .PHONY: test_kreyvium # Run tests for kreyvium
 test_kreyvium: install_rs_build_toolchain
 	RUSTFLAGS="$(RUSTFLAGS)" cargo $(CARGO_RS_BUILD_TOOLCHAIN) test --profile $(CARGO_PROFILE) \
-		kreyvium --features=$(TARGET_ARCH_FEATURE),boolean,shortint,integer \
-		-- --test-threads=1
+		-p tfhe-trivium -- --test-threads=1 kreyvium::
+
+.PHONY: test_concrete_csprng # Run concrete-csprng tests
+test_concrete_csprng:
+	RUSTFLAGS="$(RUSTFLAGS)" cargo $(CARGO_RS_BUILD_TOOLCHAIN) test --profile $(CARGO_PROFILE) \
+		--features=$(TARGET_ARCH_FEATURE) -p concrete-csprng
 
 .PHONY: doc # Build rust doc
 doc: install_rs_check_toolchain
@@ -366,18 +508,18 @@ format_doc_latex:
 .PHONY: check_compile_tests # Build tests in debug without running them
 check_compile_tests:
 	RUSTFLAGS="$(RUSTFLAGS)" cargo $(CARGO_RS_BUILD_TOOLCHAIN) test --no-run \
-		--features=$(TARGET_ARCH_FEATURE),experimental,boolean,shortint,integer,internal-keycache \
-		-p tfhe
+		--features=$(TARGET_ARCH_FEATURE),experimental,boolean,shortint,integer,internal-keycache,safe-deserialization \
+		-p $(TFHE_SPEC)
 
 	@if [[ "$(OS)" == "Linux" || "$(OS)" == "Darwin" ]]; then \
-		"$(MAKE)" build_c_api; \
+		"$(MAKE)" build_c_api && \
 		./scripts/c_api_tests.sh --build-only; \
 	fi
 
 .PHONY: build_nodejs_test_docker # Build a docker image with tools to run nodejs tests for wasm API
 build_nodejs_test_docker:
 	DOCKER_BUILDKIT=1 docker build --build-arg RUST_TOOLCHAIN="$(RS_BUILD_TOOLCHAIN)" \
-		-f docker/Dockerfile.wasm_tests -t tfhe-wasm-tests .
+		-f docker/Dockerfile.wasm_tests --build-arg NODE_VERSION=$(NODE_VERSION) -t tfhe-wasm-tests .
 
 .PHONY: test_nodejs_wasm_api_in_docker # Run tests for the nodejs on wasm API in a docker container
 test_nodejs_wasm_api_in_docker: build_nodejs_test_docker
@@ -401,7 +543,8 @@ test_web_js_api_parallel: build_web_js_api_parallel
 .PHONY: ci_test_web_js_api_parallel # Run tests for the web wasm api
 ci_test_web_js_api_parallel: build_web_js_api_parallel
 	source ~/.nvm/nvm.sh && \
-	nvm use node && \
+	nvm install $(NODE_VERSION) && \
+	nvm use $(NODE_VERSION) && \
 	$(MAKE) -C tfhe/web_wasm_parallel_tests test-ci
 
 .PHONY: no_tfhe_typo # Check we did not invert the h and f in tfhe
@@ -412,31 +555,50 @@ no_tfhe_typo:
 no_dbg_log:
 	@./scripts/no_dbg_calls.sh
 
+.PHONY: dieharder_csprng # Run the dieharder test suite on our CSPRNG implementation
+dieharder_csprng: install_dieharder build_concrete_csprng
+	./scripts/dieharder_test.sh
+
 #
 # Benchmarks
 #
 
-.PHONY: bench_integer # Run benchmarks for integer
+.PHONY: bench_integer # Run benchmarks for unsigned integer
 bench_integer: install_rs_check_toolchain
 	RUSTFLAGS="$(RUSTFLAGS)" __TFHE_RS_BENCH_OP_FLAVOR=$(BENCH_OP_FLAVOR) __TFHE_RS_FAST_BENCH=$(FAST_BENCH) \
 	cargo $(CARGO_RS_CHECK_TOOLCHAIN) bench \
 	--bench integer-bench \
-	--features=$(TARGET_ARCH_FEATURE),integer,internal-keycache,$(AVX512_FEATURE) -p tfhe --
+	--features=$(TARGET_ARCH_FEATURE),integer,internal-keycache,$(AVX512_FEATURE) -p $(TFHE_SPEC) --
 
-.PHONY: bench_integer_multi_bit # Run benchmarks for integer using multi-bit parameters
+.PHONY: bench_integer_multi_bit # Run benchmarks for unsigned integer using multi-bit parameters
 bench_integer_multi_bit: install_rs_check_toolchain
 	RUSTFLAGS="$(RUSTFLAGS)" __TFHE_RS_BENCH_TYPE=MULTI_BIT \
 	__TFHE_RS_BENCH_OP_FLAVOR=$(BENCH_OP_FLAVOR) __TFHE_RS_FAST_BENCH=$(FAST_BENCH) \
 	cargo $(CARGO_RS_CHECK_TOOLCHAIN) bench \
 	--bench integer-bench \
-	--features=$(TARGET_ARCH_FEATURE),integer,internal-keycache,$(AVX512_FEATURE) -p tfhe --
+	--features=$(TARGET_ARCH_FEATURE),integer,internal-keycache,$(AVX512_FEATURE) -p $(TFHE_SPEC) --
+
+.PHONY: bench_signed_integer # Run benchmarks for signed integer
+bench_signed_integer: install_rs_check_toolchain
+	RUSTFLAGS="$(RUSTFLAGS)" __TFHE_RS_BENCH_OP_FLAVOR=$(BENCH_OP_FLAVOR) __TFHE_RS_FAST_BENCH=$(FAST_BENCH) \
+	cargo $(CARGO_RS_CHECK_TOOLCHAIN) bench \
+	--bench integer-signed-bench \
+	--features=$(TARGET_ARCH_FEATURE),integer,internal-keycache,$(AVX512_FEATURE) -p $(TFHE_SPEC) --
+
+.PHONY: bench_signed_integer_multi_bit # Run benchmarks for signed integer using multi-bit parameters
+bench_signed_integer_multi_bit: install_rs_check_toolchain
+	RUSTFLAGS="$(RUSTFLAGS)" __TFHE_RS_BENCH_TYPE=MULTI_BIT \
+	__TFHE_RS_BENCH_OP_FLAVOR=$(BENCH_OP_FLAVOR) __TFHE_RS_FAST_BENCH=$(FAST_BENCH) \
+	cargo $(CARGO_RS_CHECK_TOOLCHAIN) bench \
+	--bench integer-signed-bench \
+	--features=$(TARGET_ARCH_FEATURE),integer,internal-keycache,$(AVX512_FEATURE) -p $(TFHE_SPEC) --
 
 .PHONY: bench_shortint # Run benchmarks for shortint
 bench_shortint: install_rs_check_toolchain
 	RUSTFLAGS="$(RUSTFLAGS)" __TFHE_RS_BENCH_OP_FLAVOR=$(BENCH_OP_FLAVOR) \
 	cargo $(CARGO_RS_CHECK_TOOLCHAIN) bench \
 	--bench shortint-bench \
-	--features=$(TARGET_ARCH_FEATURE),shortint,internal-keycache,$(AVX512_FEATURE) -p tfhe
+	--features=$(TARGET_ARCH_FEATURE),shortint,internal-keycache,$(AVX512_FEATURE) -p $(TFHE_SPEC)
 
 .PHONY: bench_shortint_multi_bit # Run benchmarks for shortint using multi-bit parameters
 bench_shortint_multi_bit: install_rs_check_toolchain
@@ -444,20 +606,20 @@ bench_shortint_multi_bit: install_rs_check_toolchain
 	__TFHE_RS_BENCH_OP_FLAVOR=$(BENCH_OP_FLAVOR) \
 	cargo $(CARGO_RS_CHECK_TOOLCHAIN) bench \
 	--bench shortint-bench \
-	--features=$(TARGET_ARCH_FEATURE),shortint,internal-keycache,$(AVX512_FEATURE) -p tfhe --
+	--features=$(TARGET_ARCH_FEATURE),shortint,internal-keycache,$(AVX512_FEATURE) -p $(TFHE_SPEC) --
 
 
 .PHONY: bench_boolean # Run benchmarks for boolean
 bench_boolean: install_rs_check_toolchain
 	RUSTFLAGS="$(RUSTFLAGS)" cargo $(CARGO_RS_CHECK_TOOLCHAIN) bench \
 	--bench boolean-bench \
-	--features=$(TARGET_ARCH_FEATURE),boolean,internal-keycache,$(AVX512_FEATURE) -p tfhe
+	--features=$(TARGET_ARCH_FEATURE),boolean,internal-keycache,$(AVX512_FEATURE) -p $(TFHE_SPEC)
 
 .PHONY: bench_pbs # Run benchmarks for PBS
 bench_pbs: install_rs_check_toolchain
 	RUSTFLAGS="$(RUSTFLAGS)" cargo $(CARGO_RS_CHECK_TOOLCHAIN) bench \
 	--bench pbs-bench \
-	--features=$(TARGET_ARCH_FEATURE),boolean,shortint,internal-keycache,$(AVX512_FEATURE) -p tfhe
+	--features=$(TARGET_ARCH_FEATURE),boolean,shortint,internal-keycache,$(AVX512_FEATURE) -p $(TFHE_SPEC)
 
 .PHONY: bench_web_js_api_parallel # Run benchmarks for the web wasm api
 bench_web_js_api_parallel: build_web_js_api_parallel
@@ -472,6 +634,18 @@ ci_bench_web_js_api_parallel: build_web_js_api_parallel
 #
 # Utility tools
 #
+.PHONY: gen_key_cache # Run the script to generate keys and cache them for shortint tests
+gen_key_cache: install_rs_build_toolchain
+	RUSTFLAGS="$(RUSTFLAGS)" cargo $(CARGO_RS_BUILD_TOOLCHAIN) run --profile $(CARGO_PROFILE) \
+		--example generates_test_keys \
+		--features=$(TARGET_ARCH_FEATURE),boolean,shortint,internal-keycache -- \
+		$(MULTI_BIT_ONLY) $(COVERAGE_ONLY)
+
+.PHONY: gen_key_cache_core_crypto # Run function to generate keys and cache them for core_crypto tests
+gen_key_cache_core_crypto: install_rs_build_toolchain
+	RUSTFLAGS="$(RUSTFLAGS)" cargo $(CARGO_RS_BUILD_TOOLCHAIN) test --tests --profile $(CARGO_PROFILE) \
+		--features=$(TARGET_ARCH_FEATURE),experimental,internal-keycache -p $(TFHE_SPEC) -- --nocapture \
+		core_crypto::keycache::generate_keys
 
 .PHONY: measure_hlapi_compact_pk_ct_sizes # Measure sizes of public keys and ciphertext for high-level API
 measure_hlapi_compact_pk_ct_sizes: install_rs_check_toolchain
@@ -541,7 +715,7 @@ pcc: no_tfhe_typo no_dbg_log check_fmt lint_doc clippy_all check_compile_tests
 fpcc: no_tfhe_typo no_dbg_log check_fmt lint_doc clippy_fast check_compile_tests
 
 .PHONY: conformance # Automatically fix problems that can be fixed
-conformance: fmt
+conformance: fix_newline fmt
 
 .PHONY: help # Generate list of targets with descriptions
 help:

README.md

@@ -47,7 +47,7 @@ tfhe = { version = "*", features = ["boolean", "shortint", "integer", "x86_64-un
 ```toml
 tfhe = { version = "*", features = ["boolean", "shortint", "integer", "aarch64-unix"] }
 ```
-Note: users with ARM devices must use `TFHE-rs` by compiling using the `nightly` toolchain.
+Note: users with ARM devices must compile `TFHE-rs` using a stable toolchain with version >= 1.72.
 
 
 + For x86_64-based machines with the [`rdseed instruction`](https://en.wikipedia.org/wiki/RDRAND) 
@@ -70,9 +70,7 @@ use tfhe::{generate_keys, set_server_key, ConfigBuilder, FheUint32, FheUint8};
 
 fn main() -> Result<(), Box<dyn std::error::Error>> {
     // Basic configuration to use homomorphic integers
-    let config = ConfigBuilder::all_disabled()
-        .enable_default_integers()
-        .build();
+    let config = ConfigBuilder::default().build();
 
     // Key generation
     let (client_key, server_keys) = generate_keys(config);
@@ -92,10 +90,10 @@ fn main() -> Result<(), Box<dyn std::error::Error>> {
     // On the server side:
     set_server_key(server_keys);
 
-    // Clear equivalent computations: 1344 * 8 = 10752
+    // Clear equivalent computations: 1344 * 5 = 6720
     let encrypted_res_mul = &encrypted_a * &encrypted_b;
 
-    // Clear equivalent computations: 1344 >> 8 = 42
+    // Clear equivalent computations: 1344 >> 5 = 42
     encrypted_a = &encrypted_res_mul >> &encrypted_b;
 
     // Clear equivalent computations: let casted_a = a as u8;

apps/trivium/Cargo.toml

@@ -17,7 +17,7 @@ path = "../../tfhe"
 features = [ "boolean", "shortint", "integer", "aarch64-unix" ]
 
 [dev-dependencies]
-criterion = { version = "0.4", features = [ "html_reports" ]}
+criterion = { version = "0.5.1", features = [ "html_reports" ]}
 
 [[bench]]
 name = "trivium"

apps/trivium/README.md

@@ -120,7 +120,7 @@ fn main() {
 
 # FHE byte Trivium implementation
 
-The same objects have also been implemented to stream bytes insead of booleans. They can be constructed and used in the same way via the functions `TriviumStreamByte::<u8>::new` and 
+The same objects have also been implemented to stream bytes instead of booleans. They can be constructed and used in the same way via the functions `TriviumStreamByte::<u8>::new` and 
 `TriviumStreamByte::<FheUint8>::new` with the same arguments as before. The `FheUint8` version is significantly slower than the `FheBool` version, because not running 
 with the same cryptographic parameters. Its interest lie in its trans-ciphering capabilities: `TriviumStreamByte<FheUint8>` implements the trait `TransCiphering`, 
 meaning it implements the functions `trans_encrypt_64`. This function takes as input a `FheUint64` and outputs a `FheUint64`, the output being

apps/trivium/benches/kreyvium_bool.rs

@@ -6,7 +6,7 @@ use tfhe_trivium::KreyviumStream;
 use criterion::Criterion;
 
 pub fn kreyvium_bool_gen(c: &mut Criterion) {
-    let config = ConfigBuilder::all_disabled().enable_default_bool().build();
+    let config = ConfigBuilder::default().build();
     let (client_key, server_key) = generate_keys(config);
 
     let key_string = "0053A6F94C9FF24598EB000000000000".to_string();
@@ -41,7 +41,7 @@ pub fn kreyvium_bool_gen(c: &mut Criterion) {
 }
 
 pub fn kreyvium_bool_warmup(c: &mut Criterion) {
-    let config = ConfigBuilder::all_disabled().enable_default_bool().build();
+    let config = ConfigBuilder::default().build();
     let (client_key, server_key) = generate_keys(config);
 
     let key_string = "0053A6F94C9FF24598EB000000000000".to_string();

apps/trivium/benches/kreyvium_byte.rs

@@ -6,9 +6,8 @@ use tfhe_trivium::{KreyviumStreamByte, TransCiphering};
 use criterion::Criterion;
 
 pub fn kreyvium_byte_gen(c: &mut Criterion) {
-    let config = ConfigBuilder::all_disabled()
-        .enable_default_integers()
-        .enable_function_evaluation_integers()
+    let config = ConfigBuilder::default()
+        .enable_function_evaluation()
         .build();
     let (client_key, server_key) = generate_keys(config);
 
@@ -36,9 +35,8 @@ pub fn kreyvium_byte_gen(c: &mut Criterion) {
 }
 
 pub fn kreyvium_byte_trans(c: &mut Criterion) {
-    let config = ConfigBuilder::all_disabled()
-        .enable_default_integers()
-        .enable_function_evaluation_integers()
+    let config = ConfigBuilder::default()
+        .enable_function_evaluation()
         .build();
     let (client_key, server_key) = generate_keys(config);
 
@@ -67,9 +65,8 @@ pub fn kreyvium_byte_trans(c: &mut Criterion) {
 }
 
 pub fn kreyvium_byte_warmup(c: &mut Criterion) {
-    let config = ConfigBuilder::all_disabled()
-        .enable_default_integers()
-        .enable_function_evaluation_integers()
+    let config = ConfigBuilder::default()
+        .enable_function_evaluation()
         .build();
     let (client_key, server_key) = generate_keys(config);
 

apps/trivium/benches/kreyvium_shortint.rs

@@ -8,9 +8,7 @@ use tfhe_trivium::{KreyviumStreamShortint, TransCiphering};
 use criterion::Criterion;
 
 pub fn kreyvium_shortint_warmup(c: &mut Criterion) {
-    let config = ConfigBuilder::all_disabled()
-        .enable_default_integers()
-        .build();
+    let config = ConfigBuilder::default().build();
     let (hl_client_key, hl_server_key) = generate_keys(config);
     let underlying_ck: tfhe::shortint::ClientKey = (*hl_client_key.as_ref()).clone().into();
     let underlying_sk: tfhe::shortint::ServerKey = (*hl_server_key.as_ref()).clone().into();
@@ -60,9 +58,7 @@ pub fn kreyvium_shortint_warmup(c: &mut Criterion) {
 }
 
 pub fn kreyvium_shortint_gen(c: &mut Criterion) {
-    let config = ConfigBuilder::all_disabled()
-        .enable_default_integers()
-        .build();
+    let config = ConfigBuilder::default().build();
     let (hl_client_key, hl_server_key) = generate_keys(config);
     let underlying_ck: tfhe::shortint::ClientKey = (*hl_client_key.as_ref()).clone().into();
     let underlying_sk: tfhe::shortint::ServerKey = (*hl_server_key.as_ref()).clone().into();
@@ -107,9 +103,7 @@ pub fn kreyvium_shortint_gen(c: &mut Criterion) {
 }
 
 pub fn kreyvium_shortint_trans(c: &mut Criterion) {
-    let config = ConfigBuilder::all_disabled()
-        .enable_default_integers()
-        .build();
+    let config = ConfigBuilder::default().build();
     let (hl_client_key, hl_server_key) = generate_keys(config);
     let underlying_ck: tfhe::shortint::ClientKey = (*hl_client_key.as_ref()).clone().into();
     let underlying_sk: tfhe::shortint::ServerKey = (*hl_server_key.as_ref()).clone().into();

apps/trivium/benches/trivium_bool.rs

@@ -6,7 +6,7 @@ use tfhe_trivium::TriviumStream;
 use criterion::Criterion;
 
 pub fn trivium_bool_gen(c: &mut Criterion) {
-    let config = ConfigBuilder::all_disabled().enable_default_bool().build();
+    let config = ConfigBuilder::default().build();
     let (client_key, server_key) = generate_keys(config);
 
     let key_string = "0053A6F94C9FF24598EB".to_string();
@@ -41,7 +41,7 @@ pub fn trivium_bool_gen(c: &mut Criterion) {
 }
 
 pub fn trivium_bool_warmup(c: &mut Criterion) {
-    let config = ConfigBuilder::all_disabled().enable_default_bool().build();
+    let config = ConfigBuilder::default().build();
     let (client_key, server_key) = generate_keys(config);
 
     let key_string = "0053A6F94C9FF24598EB".to_string();

apps/trivium/benches/trivium_byte.rs

@@ -6,9 +6,7 @@ use tfhe_trivium::{TransCiphering, TriviumStreamByte};
 use criterion::Criterion;
 
 pub fn trivium_byte_gen(c: &mut Criterion) {
-    let config = ConfigBuilder::all_disabled()
-        .enable_default_integers()
-        .build();
+    let config = ConfigBuilder::default().build();
     let (client_key, server_key) = generate_keys(config);
 
     let key_string = "0053A6F94C9FF24598EB".to_string();
@@ -35,9 +33,7 @@ pub fn trivium_byte_gen(c: &mut Criterion) {
 }
 
 pub fn trivium_byte_trans(c: &mut Criterion) {
-    let config = ConfigBuilder::all_disabled()
-        .enable_default_integers()
-        .build();
+    let config = ConfigBuilder::default().build();
     let (client_key, server_key) = generate_keys(config);
 
     let key_string = "0053A6F94C9FF24598EB".to_string();
@@ -65,9 +61,7 @@ pub fn trivium_byte_trans(c: &mut Criterion) {
 }
 
 pub fn trivium_byte_warmup(c: &mut Criterion) {
-    let config = ConfigBuilder::all_disabled()
-        .enable_default_integers()
-        .build();
+    let config = ConfigBuilder::default().build();
     let (client_key, server_key) = generate_keys(config);
 
     let key_string = "0053A6F94C9FF24598EB".to_string();

apps/trivium/benches/trivium_shortint.rs

@@ -8,9 +8,7 @@ use tfhe_trivium::{TransCiphering, TriviumStreamShortint};
 use criterion::Criterion;
 
 pub fn trivium_shortint_warmup(c: &mut Criterion) {
-    let config = ConfigBuilder::all_disabled()
-        .enable_default_integers()
-        .build();
+    let config = ConfigBuilder::default().build();
     let (hl_client_key, hl_server_key) = generate_keys(config);
     let underlying_ck: tfhe::shortint::ClientKey = (*hl_client_key.as_ref()).clone().into();
     let underlying_sk: tfhe::shortint::ServerKey = (*hl_server_key.as_ref()).clone().into();
@@ -60,9 +58,7 @@ pub fn trivium_shortint_warmup(c: &mut Criterion) {
 }
 
 pub fn trivium_shortint_gen(c: &mut Criterion) {
-    let config = ConfigBuilder::all_disabled()
-        .enable_default_integers()
-        .build();
+    let config = ConfigBuilder::default().build();
     let (hl_client_key, hl_server_key) = generate_keys(config);
     let underlying_ck: tfhe::shortint::ClientKey = (*hl_client_key.as_ref()).clone().into();
     let underlying_sk: tfhe::shortint::ServerKey = (*hl_server_key.as_ref()).clone().into();
@@ -107,9 +103,7 @@ pub fn trivium_shortint_gen(c: &mut Criterion) {
 }
 
 pub fn trivium_shortint_trans(c: &mut Criterion) {
-    let config = ConfigBuilder::all_disabled()
-        .enable_default_integers()
-        .build();
+    let config = ConfigBuilder::default().build();
     let (hl_client_key, hl_server_key) = generate_keys(config);
     let underlying_ck: tfhe::shortint::ClientKey = (*hl_client_key.as_ref()).clone().into();
     let underlying_sk: tfhe::shortint::ServerKey = (*hl_server_key.as_ref()).clone().into();

apps/trivium/src/kreyvium/kreyvium.rs

@@ -1,5 +1,5 @@
 //! This module implements the Kreyvium stream cipher, using booleans or FheBool
-//! for the representaion of the inner bits.
+//! for the representation of the inner bits.
 
 use crate::static_deque::StaticDeque;
 
@@ -35,7 +35,7 @@ pub struct KreyviumStream<T> {
 }
 
 impl KreyviumStream<bool> {
-    /// Contructor for `KreyviumStream<bool>`: arguments are the secret key and the input vector.
+    /// Constructor for `KreyviumStream<bool>`: arguments are the secret key and the input vector.
     /// Outputs a KreyviumStream object already initialized (1152 steps have been run before
     /// returning)
     pub fn new(mut key: [bool; 128], mut iv: [bool; 128]) -> KreyviumStream<bool> {
@@ -118,7 +118,7 @@ where
     T: KreyviumBoolInput<T> + std::marker::Send + std::marker::Sync,
     for<'a> &'a T: KreyviumBoolInput<T>,
 {
-    /// Internal generic contructor: arguments are already prepared registers, and an optional FHE
+    /// Internal generic constructor: arguments are already prepared registers, and an optional FHE
     /// server key
     fn new_from_registers(
         a_register: [T; 93],

apps/trivium/src/kreyvium/kreyvium_byte.rs

@@ -1,5 +1,5 @@
 //! This module implements the Kreyvium stream cipher, using u8 or FheUint8
-//! for the representaion of the inner bits.
+//! for the representation of the inner bits.
 
 use crate::static_deque::{StaticByteDeque, StaticByteDequeInput};
 
@@ -31,7 +31,7 @@ impl KreyviumByteInput<FheUint8> for &FheUint8 {}
 /// representation of bits (u8 or FheUint8). To be able to compute FHE operations, it also owns
 /// an Option for a ServerKey.
 /// Since the original Kreyvium registers' sizes are not a multiple of 8, these registers (which
-/// store byte-like objects) have a size that is the eigth of the closest multiple of 8 above the
+/// store byte-like objects) have a size that is the eighth of the closest multiple of 8 above the
 /// originals' sizes.
 pub struct KreyviumStreamByte<T> {
     a_byte: StaticByteDeque<12, T>,
@@ -43,7 +43,7 @@ pub struct KreyviumStreamByte<T> {
 }
 
 impl KreyviumStreamByte<u8> {
-    /// Contructor for `KreyviumStreamByte<u8>`: arguments are the secret key and the input vector.
+    /// Constructor for `KreyviumStreamByte<u8>`: arguments are the secret key and the input vector.
     /// Outputs a KreyviumStream object already initialized (1152 steps have been run before
     /// returning)
     pub fn new(key_bytes: [u8; 16], iv_bytes: [u8; 16]) -> KreyviumStreamByte<u8> {
@@ -146,7 +146,7 @@ where
     T: KreyviumByteInput<T> + Send,
     for<'a> &'a T: KreyviumByteInput<T>,
 {
-    /// Internal generic contructor: arguments are already prepared registers, and an optional FHE
+    /// Internal generic constructor: arguments are already prepared registers, and an optional FHE
     /// server key
     fn new_from_registers(
         a_register: [T; 12],

apps/trivium/src/kreyvium/kreyvium_shortint.rs

@@ -19,7 +19,7 @@ pub struct KreyviumStreamShortint {
 }
 
 impl KreyviumStreamShortint {
-    /// Contructor for KreyviumStreamShortint: arguments are the secret key and the input vector,
+    /// Constructor for KreyviumStreamShortint: arguments are the secret key and the input vector,
     /// and a ServerKey reference. Outputs a KreyviumStream object already initialized (1152
     /// steps have been run before returning)
     pub fn new(
@@ -149,7 +149,7 @@ impl KreyviumStreamShortint {
                             .unchecked_add_assign(&mut new_c, c5);
                         self.internal_server_key
                             .unchecked_add_assign(&mut new_c, &temp_b);
-                        self.internal_server_key.clear_carry_assign(&mut new_c);
+                        self.internal_server_key.message_extract_assign(&mut new_c);
                         new_c
                     },
                     || {

apps/trivium/src/kreyvium/test.rs

@@ -170,7 +170,7 @@ fn kreyvium_test_4() {
 
 #[test]
 fn kreyvium_test_fhe_long() {
-    let config = ConfigBuilder::all_disabled().enable_default_bool().build();
+    let config = ConfigBuilder::default().build();
     let (client_key, server_key) = generate_keys(config);
 
     let key_string = "0053A6F94C9FF24598EB000000000000".to_string();
@@ -217,9 +217,7 @@ use tfhe::shortint::prelude::*;
 
 #[test]
 fn kreyvium_test_shortint_long() {
-    let config = ConfigBuilder::all_disabled()
-        .enable_default_integers()
-        .build();
+    let config = ConfigBuilder::default().build();
     let (hl_client_key, hl_server_key) = generate_keys(config);
     let underlying_ck: tfhe::shortint::ClientKey = (*hl_client_key.as_ref()).clone().into();
     let underlying_sk: tfhe::shortint::ServerKey = (*hl_server_key.as_ref()).clone().into();
@@ -302,9 +300,8 @@ fn kreyvium_test_clear_byte() {
 
 #[test]
 fn kreyvium_test_byte_long() {
-    let config = ConfigBuilder::all_disabled()
-        .enable_default_integers()
-        .enable_function_evaluation_integers()
+    let config = ConfigBuilder::default()
+        .enable_function_evaluation()
         .build();
     let (client_key, server_key) = generate_keys(config);
 
@@ -342,9 +339,8 @@ fn kreyvium_test_byte_long() {
 
 #[test]
 fn kreyvium_test_fhe_byte_transciphering_long() {
-    let config = ConfigBuilder::all_disabled()
-        .enable_default_integers()
-        .enable_function_evaluation_integers()
+    let config = ConfigBuilder::default()
+        .enable_function_evaluation()
         .build();
     let (client_key, server_key) = generate_keys(config);
 

apps/trivium/src/static_deque/static_byte_deque.rs

@@ -1,6 +1,6 @@
 //! This module implements the StaticByteDeque struct: a deque of bytes. The idea
 //! is that this is a wrapper around StaticDeque, but StaticByteDeque has an additional
-//! functionnality: it can construct the "intermediate" bytes, made of parts of other bytes.
+//! functionality: it can construct the "intermediate" bytes, made of parts of other bytes.
 //! This is pretending to store bits, and allows accessing bits in chunks of 8 consecutive.
 
 use crate::static_deque::StaticDeque;

apps/trivium/src/static_deque/static_deque.rs

@@ -5,7 +5,7 @@
 use core::ops::{Index, IndexMut};
 
 /// StaticDeque: a struct implementing a deque whose size is known at compile time.
-/// It has 2 members: the static array conatining the data (never empty), and a cursor
+/// It has 2 members: the static array containing the data (never empty), and a cursor
 /// equal to the index of the oldest element (and the next one to be overwritten).
 #[derive(Clone)]
 pub struct StaticDeque<const N: usize, T> {

apps/trivium/src/trans_ciphering/mod.rs

@@ -4,6 +4,7 @@
 use crate::{KreyviumStreamByte, KreyviumStreamShortint, TriviumStreamByte, TriviumStreamShortint};
 use tfhe::shortint::Ciphertext;
 
+use tfhe::prelude::*;
 use tfhe::{set_server_key, unset_server_key, FheUint64, FheUint8, ServerKey};
 
 use rayon::prelude::*;

apps/trivium/src/trivium/mod.rs

@@ -1,6 +1,5 @@
-#[allow(clippy::module_inception)]
-mod trivium;
-pub use trivium::TriviumStream;
+mod trivium_bool;
+pub use trivium_bool::TriviumStream;
 
 mod trivium_byte;
 pub use trivium_byte::TriviumStreamByte;

apps/trivium/src/trivium/test.rs

@@ -232,7 +232,7 @@ fn trivium_test_clear_byte() {
 
 #[test]
 fn trivium_test_fhe_long() {
-    let config = ConfigBuilder::all_disabled().enable_default_bool().build();
+    let config = ConfigBuilder::default().build();
     let (client_key, server_key) = generate_keys(config);
 
     let key_string = "0053A6F94C9FF24598EB".to_string();
@@ -277,9 +277,7 @@ fn trivium_test_fhe_long() {
 
 #[test]
 fn trivium_test_fhe_byte_long() {
-    let config = ConfigBuilder::all_disabled()
-        .enable_default_integers()
-        .build();
+    let config = ConfigBuilder::default().build();
     let (client_key, server_key) = generate_keys(config);
 
     let key_string = "0053A6F94C9FF24598EB".to_string();
@@ -316,9 +314,7 @@ fn trivium_test_fhe_byte_long() {
 
 #[test]
 fn trivium_test_fhe_byte_transciphering_long() {
-    let config = ConfigBuilder::all_disabled()
-        .enable_default_integers()
-        .build();
+    let config = ConfigBuilder::default().build();
     let (client_key, server_key) = generate_keys(config);
 
     let key_string = "0053A6F94C9FF24598EB".to_string();
@@ -357,9 +353,7 @@ use tfhe::shortint::prelude::*;
 
 #[test]
 fn trivium_test_shortint_long() {
-    let config = ConfigBuilder::all_disabled()
-        .enable_default_integers()
-        .build();
+    let config = ConfigBuilder::default().build();
     let (hl_client_key, hl_server_key) = generate_keys(config);
     let underlying_ck: tfhe::shortint::ClientKey = (*hl_client_key.as_ref()).clone().into();
     let underlying_sk: tfhe::shortint::ServerKey = (*hl_server_key.as_ref()).clone().into();

apps/trivium/src/trivium/trivium_bool.rs

@@ -1,5 +1,5 @@
 //! This module implements the Trivium stream cipher, using booleans or FheBool
-//! for the representaion of the inner bits.
+//! for the representation of the inner bits.
 
 use crate::static_deque::StaticDeque;
 
@@ -33,7 +33,7 @@ pub struct TriviumStream<T> {
 }
 
 impl TriviumStream<bool> {
-    /// Contructor for `TriviumStream<bool>`: arguments are the secret key and the input vector.
+    /// Constructor for `TriviumStream<bool>`: arguments are the secret key and the input vector.
     /// Outputs a TriviumStream object already initialized (1152 steps have been run before
     /// returning)
     pub fn new(key: [bool; 80], iv: [bool; 80]) -> TriviumStream<bool> {
@@ -94,7 +94,7 @@ where
     T: TriviumBoolInput<T> + std::marker::Send + std::marker::Sync,
     for<'a> &'a T: TriviumBoolInput<T>,
 {
-    /// Internal generic contructor: arguments are already prepared registers, and an optional FHE
+    /// Internal generic constructor: arguments are already prepared registers, and an optional FHE
     /// server key
     fn new_from_registers(
         a_register: [T; 93],

apps/trivium/src/trivium/trivium_byte.rs

@@ -1,5 +1,5 @@
 //! This module implements the Trivium stream cipher, using u8 or FheUint8
-//! for the representaion of the inner bits.
+//! for the representation of the inner bits.
 
 use crate::static_deque::{StaticByteDeque, StaticByteDequeInput};
 
@@ -31,7 +31,7 @@ impl TriviumByteInput<FheUint8> for &FheUint8 {}
 /// representation of bits (u8 or FheUint8). To be able to compute FHE operations, it also owns
 /// an Option for a ServerKey.
 /// Since the original Trivium registers' sizes are not a multiple of 8, these registers (which
-/// store byte-like objects) have a size that is the eigth of the closest multiple of 8 above the
+/// store byte-like objects) have a size that is the eighth of the closest multiple of 8 above the
 /// originals' sizes.
 pub struct TriviumStreamByte<T> {
     a_byte: StaticByteDeque<12, T>,
@@ -41,7 +41,7 @@ pub struct TriviumStreamByte<T> {
 }
 
 impl TriviumStreamByte<u8> {
-    /// Contructor for `TriviumStreamByte<u8>`: arguments are the secret key and the input vector.
+    /// Constructor for `TriviumStreamByte<u8>`: arguments are the secret key and the input vector.
     /// Outputs a TriviumStream object already initialized (1152 steps have been run before
     /// returning)
     pub fn new(key: [u8; 10], iv: [u8; 10]) -> TriviumStreamByte<u8> {
@@ -111,7 +111,7 @@ where
     T: TriviumByteInput<T> + Send,
     for<'a> &'a T: TriviumByteInput<T>,
 {
-    /// Internal generic contructor: arguments are already prepared registers, and an optional FHE
+    /// Internal generic constructor: arguments are already prepared registers, and an optional FHE
     /// server key
     fn new_from_registers(
         a_register: [T; 12],

apps/trivium/src/trivium/trivium_shortint.rs

@@ -17,9 +17,9 @@ pub struct TriviumStreamShortint {
 }
 
 impl TriviumStreamShortint {
-    /// Contructor for TriviumStreamShortint: arguments are the secret key and the input vector, and
-    /// a ServerKey reference. Outputs a TriviumStream object already initialized (1152 steps
-    /// have been run before returning)
+    /// Constructor for TriviumStreamShortint: arguments are the secret key and the input vector,
+    /// and a ServerKey reference. Outputs a TriviumStream object already initialized (1152
+    /// steps have been run before returning)
     pub fn new(
         key: [Ciphertext; 80],
         iv: [u64; 80],
@@ -113,7 +113,7 @@ impl TriviumStreamShortint {
                             .unchecked_add_assign(&mut new_a, a5);
                         self.internal_server_key
                             .unchecked_add_assign(&mut new_a, &temp_c);
-                        self.internal_server_key.clear_carry_assign(&mut new_a);
+                        self.internal_server_key.message_extract_assign(&mut new_a);
                         new_a
                     },
                     || {
@@ -122,7 +122,7 @@ impl TriviumStreamShortint {
                             .unchecked_add_assign(&mut new_b, b5);
                         self.internal_server_key
                             .unchecked_add_assign(&mut new_b, &temp_a);
-                        self.internal_server_key.clear_carry_assign(&mut new_b);
+                        self.internal_server_key.message_extract_assign(&mut new_b);
                         new_b
                     },
                 )
@@ -135,7 +135,7 @@ impl TriviumStreamShortint {
                             .unchecked_add_assign(&mut new_c, c5);
                         self.internal_server_key
                             .unchecked_add_assign(&mut new_c, &temp_b);
-                        self.internal_server_key.clear_carry_assign(&mut new_c);
+                        self.internal_server_key.message_extract_assign(&mut new_c);
                         new_c
                     },
                     || {

ci/ec2_products_cost.json

@@ -1,3 +1,4 @@
 {
-  "m6i.metal": 7.168
+  "m6i.metal": 7.168,
+  "hpc7a.96xlarge": 7.7252
 }

ci/lattice_estimator.sage

@@ -36,23 +36,24 @@ def check_security(filename):
         try:
             # The lattice estimator is not able to manage such large dimension.
             # If we have the security for smaller `n` then we have security for larger ones.
-            if param.n == 32768:
+            if param.n > 16384:
                 param = param.updated(n = 16384)
 
             usvp_level = LWE.primal_usvp(param, red_cost_model = model)
             dual_level = LWE.dual_hybrid(param, red_cost_model = model)
 
             estimator_level = log(min(usvp_level["rop"], dual_level["rop"]),2 )
+            security_level = f"security level = {estimator_level} bits"
             if estimator_level < 127:
-                print("FAIL")
-                reason = f"attained security level = {estimator_level} bits target is 128 bits"
+                print("FAIL\t({security_level})")
+                reason = f"attained {security_level} target is 128 bits"
                 to_update.append((param, reason))
                 continue
         except Exception as err:
             print("FAIL")
             to_update.append((param, f"{repr(err)}"))
         else:
-            print("OK")
+            print(f"OK\t({security_level})")
 
     return to_update
 
@@ -72,4 +73,4 @@ if __name__ == "__main__":
             print(f"[{param.tag}] reason: {reason} (param)")
         sys.exit(int(1))  # Explicit conversion is needed to make this call work
     else:
-        print("All parameters passed the security check")
+        print("All parameters passed the security check")

ci/parse_integer_benches_to_csv.py

@@ -20,7 +20,10 @@ def main(args):
             bench_function_id = bench_data["function_id"]
 
             split = bench_function_id.split("::")
-            (_, function_name, parameter_set, bits) = split
+            if split.len() == 5:  # Signed integers
+                (_, _, function_name, parameter_set, bits) = split
+            else:  # Unsigned integers
+                (_, function_name, parameter_set, bits) = split
 
             if "_scalar_" in bits:
                 (bits, scalar) = bits.split("_bits_scalar_")

ci/slab.toml

@@ -1,32 +1,37 @@
 [profile.cpu-big]
 region = "eu-west-3"
-image_id = "ami-0ab73f5bd11708a85"
+image_id = "ami-051942e4055555752"
+instance_type = "m6i.32xlarge"
+
+[profile.cpu-big_fallback]
+region = "us-east-1"
+image_id = "ami-04e3bb9aebb6786df"
 instance_type = "m6i.32xlarge"
 
 [profile.cpu-small]
 region = "eu-west-3"
-image_id = "ami-0ab73f5bd11708a85"
+image_id = "ami-051942e4055555752"
 instance_type = "m6i.4xlarge"
 
 [profile.bench]
-region = "eu-west-3"
-image_id = "ami-0ab73f5bd11708a85"
-instance_type = "m6i.metal"
+region = "eu-west-1"
+image_id = "ami-0e88d98b86aff13de"
+instance_type = "hpc7a.96xlarge"
 
 [command.cpu_test]
 workflow = "aws_tfhe_tests.yml"
 profile = "cpu-big"
 check_run_name = "CPU AWS Tests"
 
-[command.cpu_integer_test]
+[command.cpu_unsigned_integer_test]
 workflow = "aws_tfhe_integer_tests.yml"
 profile = "cpu-big"
-check_run_name = "CPU Integer AWS Tests"
+check_run_name = "CPU Unsigned Integer AWS Tests"
 
-[command.cpu_multi_bit_test]
-workflow = "aws_tfhe_multi_bit_tests.yml"
+[command.cpu_signed_integer_test]
+workflow = "aws_tfhe_signed_integer_tests.yml"
 profile = "cpu-big"
-check_run_name = "CPU AWS Multi Bit Tests"
+check_run_name = "CPU Signed Integer AWS Tests"
 
 [command.cpu_wasm_test]
 workflow = "aws_tfhe_wasm_tests.yml"
@@ -43,6 +48,11 @@ workflow = "integer_full_benchmark.yml"
 profile = "bench"
 check_run_name = "Integer CPU AWS Benchmarks Full Suite"
 
+[command.signed_integer_full_bench]
+workflow = "signed_integer_full_benchmark.yml"
+profile = "bench"
+check_run_name = "Signed Integer CPU AWS Benchmarks Full Suite"
+
 [command.integer_bench]
 workflow = "integer_benchmark.yml"
 profile = "bench"
@@ -53,6 +63,16 @@ workflow = "integer_multi_bit_benchmark.yml"
 profile = "bench"
 check_run_name = "Integer multi bit CPU AWS Benchmarks"
 
+[command.signed_integer_bench]
+workflow = "signed_integer_benchmark.yml"
+profile = "bench"
+check_run_name = "Signed integer CPU AWS Benchmarks"
+
+[command.signed_integer_multi_bit_bench]
+workflow = "signed_integer_multi_bit_benchmark.yml"
+profile = "bench"
+check_run_name = "Signed integer multi bit CPU AWS Benchmarks"
+
 [command.shortint_full_bench]
 workflow = "shortint_full_benchmark.yml"
 profile = "bench"
@@ -77,3 +97,13 @@ check_run_name = "PBS CPU AWS Benchmarks"
 workflow = "wasm_client_benchmark.yml"
 profile = "cpu-small"
 check_run_name = "WASM Client AWS Benchmarks"
+
+[command.csprng_randomness_testing]
+workflow = "csprng_randomness_testing.yml"
+profile = "cpu-small"
+check_run_name = "CSPRNG randomness testing"
+
+[command.code_coverage]
+workflow = "code_coverage.yml"
+profile = "cpu-small"
+check_run_name = "Code coverage"

codecov.yml

@@ -0,0 +1,4 @@
+coverage:
+  status:
+    # Disable patch checks in GitHub until all tfhe-rs layers have coverage implemented.
+    patch: false

concrete-csprng/Cargo.toml

@@ -0,0 +1,53 @@
+[package]
+name = "concrete-csprng"
+version = "0.4.0"
+edition = "2021"
+license = "BSD-3-Clause-Clear"
+description = "Cryptographically Secure PRNG used in the TFHE-rs library."
+homepage = "https://zama.ai/"
+documentation = "https://docs.zama.ai/tfhe-rs"
+repository = "https://github.com/zama-ai/tfhe-rs"
+readme = "README.md"
+keywords = ["fully", "homomorphic", "encryption", "fhe", "cryptography"]
+rust-version = "1.72"
+
+[dependencies]
+aes = "0.8.2"
+rayon = { version = "1.5.0", optional = true }
+
+[target.'cfg(target_os = "macos")'.dependencies]
+libc = "0.2.133"
+
+[dev-dependencies]
+rand = "0.8.3"
+criterion = "0.5.1"
+clap = "=4.4.4"
+
+[features]
+parallel = ["rayon"]
+seeder_x86_64_rdseed = []
+seeder_unix = []
+generator_x86_64_aesni = []
+generator_fallback = []
+generator_aarch64_aes = []
+
+x86_64 = [
+    "parallel",
+    "seeder_x86_64_rdseed",
+    "generator_x86_64_aesni",
+    "generator_fallback",
+]
+x86_64-unix = ["x86_64", "seeder_unix"]
+aarch64 = ["parallel", "generator_aarch64_aes", "generator_fallback"]
+aarch64-unix = ["aarch64", "seeder_unix"]
+
+[[bench]]
+name = "benchmark"
+path = "benches/benchmark.rs"
+harness = false
+required-features = ["seeder_x86_64_rdseed", "generator_x86_64_aesni"]
+
+[[example]]
+name = "generate"
+path = "examples/generate.rs"
+required-features = ["seeder_unix", "generator_fallback"]

concrete-csprng/LICENSE

@@ -0,0 +1,28 @@
+BSD 3-Clause Clear License
+
+Copyright © 2023 ZAMA.
+All rights reserved.
+
+Redistribution and use in source and binary forms, with or without modification,
+are permitted provided that the following conditions are met:
+
+1. Redistributions of source code must retain the above copyright notice, this
+list of conditions and the following disclaimer.
+
+2. Redistributions in binary form must reproduce the above copyright notice, this
+list of conditions and the following disclaimer in the documentation and/or other
+materials provided with the distribution.
+
+3. Neither the name of ZAMA nor the names of its contributors may be used to endorse
+or promote products derived from this software without specific prior written permission.
+
+NO EXPRESS OR IMPLIED LICENSES TO ANY PARTY'S PATENT RIGHTS ARE GRANTED BY THIS LICENSE.
+THIS SOFTWARE IS PROVIDED BY THE ZAMA AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR
+IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
+MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL
+ZAMA OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY,
+OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
+OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
+ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
+NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF
+ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

concrete-csprng/README.md

@@ -0,0 +1,23 @@
+# Concrete CSPRNG
+
+This crate contains a fast *Cryptographically Secure Pseudoramdon Number Generator*, used in the
+['concrete-core'](https://crates.io/crates/concrete-core) library, you can find it [here](../concrete-core/) in this repo.
+
+The implementation is based on the AES blockcipher used in CTR mode, as described in the ISO/IEC
+18033-4 standard.
+
+Two implementations are available, an accelerated one on x86_64 CPUs with the `aes` feature and the `sse2` feature, and a pure software one that can be used on other platforms.
+
+The crate also makes two seeders available, one needing the x86_64 feature `rdseed` and another one based on the Unix random device `/dev/random` the latter requires the user to provide a secret.
+
+## Running the benchmarks
+
+To execute the benchmarks on an x86_64 platform:
+```shell
+RUSTFLAGS="-Ctarget-cpu=native" cargo bench --features=seeder_x86_64_rdseed,generator_x86_64_aesni 
+```
+
+## License
+
+This software is distributed under the BSD-3-Clause-Clear license. If you have any questions,
+please contact us at `hello@zama.ai`.

concrete-csprng/benches/benchmark.rs

@@ -0,0 +1,54 @@
+use concrete_csprng::generators::{
+    AesniRandomGenerator, BytesPerChild, ChildrenCount, RandomGenerator,
+};
+use concrete_csprng::seeders::{RdseedSeeder, Seeder};
+use criterion::{black_box, criterion_group, criterion_main, Criterion};
+
+// The number of bytes to generate during one benchmark iteration.
+const N_GEN: usize = 1_000_000;
+
+fn parent_generate(c: &mut Criterion) {
+    let mut seeder = RdseedSeeder;
+    let mut generator = AesniRandomGenerator::new(seeder.seed());
+    c.bench_function("parent_generate", |b| {
+        b.iter(|| {
+            (0..N_GEN).for_each(|_| {
+                generator.next();
+            })
+        })
+    });
+}
+
+fn child_generate(c: &mut Criterion) {
+    let mut seeder = RdseedSeeder;
+    let mut generator = AesniRandomGenerator::new(seeder.seed());
+    let mut generator = generator
+        .try_fork(ChildrenCount(1), BytesPerChild(N_GEN * 10_000))
+        .unwrap()
+        .next()
+        .unwrap();
+    c.bench_function("child_generate", |b| {
+        b.iter(|| {
+            (0..N_GEN).for_each(|_| {
+                generator.next();
+            })
+        })
+    });
+}
+
+fn fork(c: &mut Criterion) {
+    let mut seeder = RdseedSeeder;
+    let mut generator = AesniRandomGenerator::new(seeder.seed());
+    c.bench_function("fork", |b| {
+        b.iter(|| {
+            black_box(
+                generator
+                    .try_fork(ChildrenCount(2048), BytesPerChild(2048))
+                    .unwrap(),
+            )
+        })
+    });
+}
+
+criterion_group!(benches, parent_generate, child_generate, fork);
+criterion_main!(benches);

concrete-csprng/build.rs

@@ -0,0 +1,112 @@
+// To have clear error messages during compilation about why some piece of code may not be available
+// we decided to check the features compatibility with the target configuration in this script.
+
+use std::collections::HashMap;
+use std::env;
+
+// See https://doc.rust-lang.org/reference/conditional-compilation.html#target_arch for various
+// compilation configuration
+
+// Can be easily extended if needed
+pub struct FeatureRequirement {
+    pub feature_name: &'static str,
+    // target_arch requirement
+    pub feature_req_target_arch: Option<&'static str>,
+    // target_family requirement
+    pub feature_req_target_family: Option<&'static str>,
+}
+
+// We implement a version of default that is const which is not possible through the Default trait
+impl FeatureRequirement {
+    // As we cannot use cfg!(feature = "feature_name") with something else than a literal, we need
+    // a reference to the HashMap we populate with the enabled features
+    fn is_activated(&self, build_activated_features: &HashMap<&'static str, bool>) -> bool {
+        *build_activated_features.get(self.feature_name).unwrap()
+    }
+
+    // panics if the requirements are not met
+    fn check_requirements(&self) {
+        let target_arch = get_target_arch_cfg();
+        if let Some(feature_req_target_arch) = self.feature_req_target_arch {
+            if feature_req_target_arch != target_arch {
+                panic!(
+                    "Feature `{}` requires target_arch `{}`, current cfg: `{}`",
+                    self.feature_name, feature_req_target_arch, target_arch
+                )
+            }
+        }
+
+        let target_family = get_target_family_cfg();
+        if let Some(feature_req_target_family) = self.feature_req_target_family {
+            if feature_req_target_family != target_family {
+                panic!(
+                    "Feature `{}` requires target_family `{}`, current cfg: `{}`",
+                    self.feature_name, feature_req_target_family, target_family
+                )
+            }
+        }
+    }
+}
+
+// const vecs are not yet a thing so use a fixed size array (update the array size when adding
+// requirements)
+static FEATURE_REQUIREMENTS: [FeatureRequirement; 4] = [
+    FeatureRequirement {
+        feature_name: "seeder_x86_64_rdseed",
+        feature_req_target_arch: Some("x86_64"),
+        feature_req_target_family: None,
+    },
+    FeatureRequirement {
+        feature_name: "generator_x86_64_aesni",
+        feature_req_target_arch: Some("x86_64"),
+        feature_req_target_family: None,
+    },
+    FeatureRequirement {
+        feature_name: "seeder_unix",
+        feature_req_target_arch: None,
+        feature_req_target_family: Some("unix"),
+    },
+    FeatureRequirement {
+        feature_name: "generator_aarch64_aes",
+        feature_req_target_arch: Some("aarch64"),
+        feature_req_target_family: None,
+    },
+];
+
+// For a "feature_name" feature_cfg!("feature_name") expands to
+// ("feature_name", cfg!(feature = "feature_name"))
+macro_rules! feature_cfg {
+    ($feat_name:literal) => {
+        ($feat_name, cfg!(feature = $feat_name))
+    };
+}
+
+// Static HashMap would require an additional crate (phf or lazy static e.g.), so we just write a
+// function that returns the HashMap we are interested in
+fn get_feature_enabled_status() -> HashMap<&'static str, bool> {
+    HashMap::from([
+        feature_cfg!("seeder_x86_64_rdseed"),
+        feature_cfg!("generator_x86_64_aesni"),
+        feature_cfg!("seeder_unix"),
+        feature_cfg!("generator_aarch64_aes"),
+    ])
+}
+
+// See https://stackoverflow.com/a/43435335/18088947 for the inspiration of this code
+fn get_target_arch_cfg() -> String {
+    env::var("CARGO_CFG_TARGET_ARCH").expect("CARGO_CFG_TARGET_ARCH is not set")
+}
+
+fn get_target_family_cfg() -> String {
+    env::var("CARGO_CFG_TARGET_FAMILY").expect("CARGO_CFG_TARGET_FAMILY is not set")
+}
+
+fn main() {
+    let feature_enabled_status = get_feature_enabled_status();
+
+    // This will panic if some requirements for a feature are not met
+    FEATURE_REQUIREMENTS
+        .iter()
+        .filter(|&req| FeatureRequirement::is_activated(req, &feature_enabled_status))
+        .for_each(FeatureRequirement::check_requirements);
+}

concrete-csprng/examples/generate.rs

@@ -0,0 +1,113 @@
+//! This program uses the concrete csprng to generate an infinite stream of random bytes on
+//! the program stdout. It can also generate a fixed number of bytes by passing a value along the
+//! optional argument `--bytes_total`. For testing purpose.
+use clap::{value_parser, Arg, Command};
+#[cfg(feature = "generator_x86_64_aesni")]
+use concrete_csprng::generators::AesniRandomGenerator as ActivatedRandomGenerator;
+#[cfg(feature = "generator_aarch64_aes")]
+use concrete_csprng::generators::NeonAesRandomGenerator as ActivatedRandomGenerator;
+#[cfg(all(
+    not(feature = "generator_x86_64_aesni"),
+    not(feature = "generator_aarch64_aes"),
+    feature = "generator_fallback"
+))]
+use concrete_csprng::generators::SoftwareRandomGenerator as ActivatedRandomGenerator;
+
+use concrete_csprng::generators::RandomGenerator;
+
+#[cfg(target_os = "macos")]
+use concrete_csprng::seeders::AppleSecureEnclaveSeeder as ActivatedSeeder;
+#[cfg(all(not(target_os = "macos"), feature = "seeder_x86_64_rdseed"))]
+use concrete_csprng::seeders::RdseedSeeder as ActivatedSeeder;
+#[cfg(all(
+    not(target_os = "macos"),
+    not(feature = "seeder_x86_64_rdseed"),
+    feature = "seeder_unix"
+))]
+use concrete_csprng::seeders::UnixSeeder as ActivatedSeeder;
+
+use concrete_csprng::seeders::Seeder;
+
+use std::io::prelude::*;
+use std::io::{stdout, StdoutLock};
+
+fn write_bytes(
+    buffer: &mut [u8],
+    generator: &mut ActivatedRandomGenerator,
+    stdout: &mut StdoutLock<'_>,
+) -> std::io::Result<()> {
+    buffer.iter_mut().zip(generator).for_each(|(b, g)| *b = g);
+    stdout.write_all(buffer)
+}
+
+fn infinite_bytes_generation(
+    buffer: &mut [u8],
+    generator: &mut ActivatedRandomGenerator,
+    stdout: &mut StdoutLock<'_>,
+) {
+    while write_bytes(buffer, generator, stdout).is_ok() {}
+}
+
+fn bytes_generation(
+    bytes_total: usize,
+    buffer: &mut [u8],
+    generator: &mut ActivatedRandomGenerator,
+    stdout: &mut StdoutLock<'_>,
+) {
+    let quotient = bytes_total / buffer.len();
+    let remaining = bytes_total % buffer.len();
+
+    for _ in 0..quotient {
+        write_bytes(buffer, generator, stdout).unwrap();
+    }
+
+    write_bytes(&mut buffer[0..remaining], generator, stdout).unwrap()
+}
+
+pub fn main() {
+    let matches = Command::new(
+        "Generate a stream of random numbers, specify no flags for infinite generation",
+    )
+    .arg(
+        Arg::new("bytes_total")
+            .short('b')
+            .long("bytes_total")
+            .value_parser(value_parser!(usize))
+            .help("Total number of bytes that has to be generated"),
+    )
+    .get_matches();
+
+    // Ugly hack to be able to use UnixSeeder
+    #[cfg(all(
+        not(target_os = "macos"),
+        not(feature = "seeder_x86_64_rdseed"),
+        feature = "seeder_unix"
+    ))]
+    let new_seeder = || ActivatedSeeder::new(0);
+    #[cfg(not(all(
+        not(target_os = "macos"),
+        not(feature = "seeder_x86_64_rdseed"),
+        feature = "seeder_unix"
+    )))]
+    let new_seeder = || ActivatedSeeder;
+
+    let mut seeder = new_seeder();
+    let seed = seeder.seed();
+    // Don't print on std out
+    eprintln!("seed={seed:?}");
+    let mut generator = ActivatedRandomGenerator::new(seed);
+    let stdout = stdout();
+    let mut buffer = [0u8; 16];
+
+    // lock stdout as there is a single thread running
+    let mut stdout = stdout.lock();
+
+    match matches.get_one::<usize>("bytes_total") {
+        Some(&total) => {
+            bytes_generation(total, &mut buffer, &mut generator, &mut stdout);
+        }
+        None => {
+            infinite_bytes_generation(&mut buffer, &mut generator, &mut stdout);
+        }
+    };
+}

concrete-csprng/src/generators/aes_ctr/block_cipher.rs

@@ -0,0 +1,20 @@
+use crate::generators::aes_ctr::index::AesIndex;
+use crate::generators::aes_ctr::BYTES_PER_BATCH;
+
+/// Represents a key used in the AES block cipher.
+#[derive(Clone, Copy)]
+pub struct AesKey(pub u128);
+
+/// A trait for AES block ciphers.
+///
+/// Note:
+/// -----
+///
+/// The block cipher is used in a batched manner (to reduce amortized cost on special hardware).
+/// For this reason we only expose a `generate_batch` method.
+pub trait AesBlockCipher: Clone + Send + Sync {
+    /// Instantiate a new generator from a secret key.
+    fn new(key: AesKey) -> Self;
+    /// Generates the batch corresponding to the given index.
+    fn generate_batch(&mut self, index: AesIndex) -> [u8; BYTES_PER_BATCH];
+}

concrete-csprng/src/generators/aes_ctr/generic.rs

@@ -0,0 +1,379 @@
+use crate::generators::aes_ctr::block_cipher::{AesBlockCipher, AesKey};
+use crate::generators::aes_ctr::index::TableIndex;
+use crate::generators::aes_ctr::states::{BufferPointer, ShiftAction, State};
+use crate::generators::aes_ctr::BYTES_PER_BATCH;
+use crate::generators::{ByteCount, BytesPerChild, ChildrenCount, ForkError};
+
+// Usually, to work with iterators and parallel iterators, we would use opaque types such as
+// `impl Iterator<..>`. Unfortunately, it is not yet possible to return existential types in
+// traits, which we would need for `RandomGenerator`. For this reason, we have to use the
+// full type name where needed. Hence the following trait aliases definition:
+
+/// A type alias for the children iterator closure type.
+pub type ChildrenClosure<BlockCipher> =
+    fn((usize, (Box<BlockCipher>, TableIndex, BytesPerChild))) -> AesCtrGenerator<BlockCipher>;
+
+/// A type alias for the children iterator type.
+pub type ChildrenIterator<BlockCipher> = std::iter::Map<
+    std::iter::Zip<
+        std::ops::Range<usize>,
+        std::iter::Repeat<(Box<BlockCipher>, TableIndex, BytesPerChild)>,
+    >,
+    ChildrenClosure<BlockCipher>,
+>;
+
+/// A type implementing the `RandomGenerator` api using the AES block cipher in counter mode.
+#[derive(Clone)]
+pub struct AesCtrGenerator<BlockCipher: AesBlockCipher> {
+    // The block cipher used in the background
+    pub(crate) block_cipher: Box<BlockCipher>,
+    // The state corresponding to the latest outputted byte.
+    pub(crate) state: State,
+    // The last legal index. This makes bound check faster.
+    pub(crate) last: TableIndex,
+    // The buffer containing the current batch of aes calls.
+    pub(crate) buffer: [u8; BYTES_PER_BATCH],
+}
+
+#[allow(unused)] // to please clippy when tests are not activated
+impl<BlockCipher: AesBlockCipher> AesCtrGenerator<BlockCipher> {
+    /// Generates a new csprng.
+    ///
+    /// Note :
+    /// ------
+    ///
+    /// The `start_index` given as input, points to the first byte that will be outputted by the
+    /// generator. If not given, this one is automatically set to the second table index. The
+    /// first table index is not used to prevent an edge case from happening: since `state` is
+    /// supposed to contain the index of the previous byte, the initial value must be decremented.
+    /// Using the second value prevents wrapping to the max index, which would make the bound
+    /// checking fail.
+    ///
+    /// The `bound_index` given as input, points to the first byte that can __not__ be legally
+    /// outputted by the generator. If not given, the bound is automatically set to the last
+    /// table index.
+    pub fn new(
+        key: AesKey,
+        start_index: Option<TableIndex>,
+        bound_index: Option<TableIndex>,
+    ) -> AesCtrGenerator<BlockCipher> {
+        AesCtrGenerator::from_block_cipher(
+            Box::new(BlockCipher::new(key)),
+            start_index.unwrap_or(TableIndex::SECOND),
+            bound_index.unwrap_or(TableIndex::LAST),
+        )
+    }
+
+    /// Generates a csprng from an existing block cipher.
+    pub fn from_block_cipher(
+        block_cipher: Box<BlockCipher>,
+        start_index: TableIndex,
+        bound_index: TableIndex,
+    ) -> AesCtrGenerator<BlockCipher> {
+        assert!(start_index < bound_index);
+        let last = bound_index.decremented();
+        let buffer = [0u8; BYTES_PER_BATCH];
+        let state = State::new(start_index);
+        AesCtrGenerator {
+            block_cipher,
+            state,
+            last,
+            buffer,
+        }
+    }
+
+    /// Returns the table index related to the previous random byte.
+    pub fn table_index(&self) -> TableIndex {
+        self.state.table_index()
+    }
+
+    /// Returns the bound of the generator if any.
+    ///
+    /// The bound is the table index of the first byte that can not be outputted by the generator.
+    pub fn get_bound(&self) -> TableIndex {
+        self.last.incremented()
+    }
+
+    /// Returns whether the generator is bounded or not.
+    pub fn is_bounded(&self) -> bool {
+        self.get_bound() != TableIndex::LAST
+    }
+
+    /// Computes the number of bytes that can still be outputted by the generator.
+    ///
+    /// Note :
+    /// ------
+    ///
+    /// Note that `ByteCount` uses the `u128` datatype to store the byte count. Unfortunately, the
+    /// number of remaining bytes is in ⟦0;2¹³² -1⟧. When the number is greater than 2¹²⁸ - 1,
+    /// we saturate the count at 2¹²⁸ - 1.
+    pub fn remaining_bytes(&self) -> ByteCount {
+        TableIndex::distance(&self.last, &self.state.table_index()).unwrap()
+    }
+
+    /// Outputs the next random byte.
+    pub fn generate_next(&mut self) -> u8 {
+        self.next()
+            .expect("Tried to generate a byte after the bound.")
+    }
+
+    /// Tries to fork the current generator into `n_child` generators each able to output
+    /// `child_bytes` random bytes.
+    pub fn try_fork(
+        &mut self,
+        n_children: ChildrenCount,
+        n_bytes: BytesPerChild,
+    ) -> Result<ChildrenIterator<BlockCipher>, ForkError> {
+        if n_children.0 == 0 {
+            return Err(ForkError::ZeroChildrenCount);
+        }
+        if n_bytes.0 == 0 {
+            return Err(ForkError::ZeroBytesPerChild);
+        }
+        if !self.is_fork_in_bound(n_children, n_bytes) {
+            return Err(ForkError::ForkTooLarge);
+        }
+
+        // The state currently stored in the parent generator points to the table index of the last
+        // generated byte. The first index to be generated is the next one:
+        let first_index = self.state.table_index().incremented();
+        let output = (0..n_children.0)
+            .zip(std::iter::repeat((
+                self.block_cipher.clone(),
+                first_index,
+                n_bytes,
+            )))
+            .map(
+                // This map is a little weird because we need to cast the closure to a fn pointer
+                // that matches the signature of `ChildrenIterator<BlockCipher>`.
+                // Unfortunately, the compiler does not manage to coerce this one
+                // automatically.
+                (|(i, (block_cipher, first_index, n_bytes))| {
+                    // The first index to be outputted by the child is the `first_index` shifted by
+                    // the proper amount of `child_bytes`.
+                    let child_first_index = first_index.increased(n_bytes.0 * i);
+                    // The bound of the child is the first index of its next sibling.
+                    let child_bound_index = first_index.increased(n_bytes.0 * (i + 1));
+                    AesCtrGenerator::from_block_cipher(
+                        block_cipher,
+                        child_first_index,
+                        child_bound_index,
+                    )
+                }) as ChildrenClosure<BlockCipher>,
+            );
+        // The parent next index is the bound of the last child.
+        let next_index = first_index.increased(n_bytes.0 * n_children.0);
+        self.state = State::new(next_index);
+
+        Ok(output)
+    }
+
+    pub(crate) fn is_fork_in_bound(
+        &self,
+        n_child: ChildrenCount,
+        child_bytes: BytesPerChild,
+    ) -> bool {
+        let mut end = self.state.table_index();
+        end.increase(n_child.0 * child_bytes.0);
+        end <= self.last
+    }
+}
+
+impl<BlockCipher: AesBlockCipher> Iterator for AesCtrGenerator<BlockCipher> {
+    type Item = u8;
+
+    fn next(&mut self) -> Option<Self::Item> {
+        if self.state.table_index() >= self.last {
+            None
+        } else {
+            match self.state.increment() {
+                ShiftAction::OutputByte(BufferPointer(ptr)) => Some(self.buffer[ptr]),
+                ShiftAction::RefreshBatchAndOutputByte(aes_index, BufferPointer(ptr)) => {
+                    self.buffer = self.block_cipher.generate_batch(aes_index);
+                    Some(self.buffer[ptr])
+                }
+            }
+        }
+    }
+}
+
+#[cfg(test)]
+pub mod aes_ctr_generic_test {
+    #![allow(unused)] // to please clippy when tests are not activated
+
+    use super::*;
+    use crate::generators::aes_ctr::index::{AesIndex, ByteIndex};
+    use crate::generators::aes_ctr::BYTES_PER_AES_CALL;
+    use rand::{thread_rng, Rng};
+
+    const REPEATS: usize = 1_000_000;
+
+    pub fn any_table_index() -> impl Iterator<Item = TableIndex> {
+        std::iter::repeat_with(|| {
+            TableIndex::new(
+                AesIndex(thread_rng().gen()),
+                ByteIndex(thread_rng().gen::<usize>() % BYTES_PER_AES_CALL),
+            )
+        })
+    }
+
+    pub fn any_usize() -> impl Iterator<Item = usize> {
+        std::iter::repeat_with(|| thread_rng().gen())
+    }
+
+    pub fn any_children_count() -> impl Iterator<Item = ChildrenCount> {
+        std::iter::repeat_with(|| ChildrenCount(thread_rng().gen::<usize>() % 2048 + 1))
+    }
+
+    pub fn any_bytes_per_child() -> impl Iterator<Item = BytesPerChild> {
+        std::iter::repeat_with(|| BytesPerChild(thread_rng().gen::<usize>() % 2048 + 1))
+    }
+
+    pub fn any_key() -> impl Iterator<Item = AesKey> {
+        std::iter::repeat_with(|| AesKey(thread_rng().gen()))
+    }
+
+    /// Output a valid fork:
+    ///     a table index t,
+    ///     a number of children nc,
+    ///     a number of bytes per children nb
+    ///     and a positive integer i such that:
+    ///         increase(t, nc*nb+i) < MAX with MAX the largest table index.
+    ///
+    /// Put differently, if we initialize a parent generator at t and fork it with (nc, nb), our
+    /// parent generator current index gets shifted to an index, distant of at least i bytes of
+    /// the max index.
+    pub fn any_valid_fork(
+    ) -> impl Iterator<Item = (TableIndex, ChildrenCount, BytesPerChild, usize)> {
+        any_table_index()
+            .zip(any_children_count())
+            .zip(any_bytes_per_child())
+            .zip(any_usize())
+            .map(|(((t, nc), nb), i)| (t, nc, nb, i))
+            .filter(|(t, nc, nb, i)| {
+                TableIndex::distance(&TableIndex::LAST, t).unwrap().0 > (nc.0 * nb.0 + i) as u128
+            })
+    }
+
+    /// Check the property:
+    ///     On a valid fork, the table index of the first child is the same as the table index of
+    ///     the parent before the fork.
+    pub fn prop_fork_first_state_table_index<G: AesBlockCipher>() {
+        for _ in 0..REPEATS {
+            let (t, nc, nb, i) = any_valid_fork().next().unwrap();
+            let k = any_key().next().unwrap();
+            let original_generator =
+                AesCtrGenerator::<G>::new(k, Some(t), Some(t.increased(nc.0 * nb.0 + i)));
+            let mut forked_generator = original_generator.clone();
+            let first_child = forked_generator.try_fork(nc, nb).unwrap().next().unwrap();
+            assert_eq!(original_generator.table_index(), first_child.table_index());
+        }
+    }
+
+    /// Check the property:
+    ///     On a valid fork, the table index of the first byte outputted by the parent after the
+    ///     fork, is the bound of the last child of the fork.
+    pub fn prop_fork_last_bound_table_index<G: AesBlockCipher>() {
+        for _ in 0..REPEATS {
+            let (t, nc, nb, i) = any_valid_fork().next().unwrap();
+            let k = any_key().next().unwrap();
+            let mut parent_generator =
+                AesCtrGenerator::<G>::new(k, Some(t), Some(t.increased(nc.0 * nb.0 + i)));
+            let last_child = parent_generator.try_fork(nc, nb).unwrap().last().unwrap();
+            assert_eq!(
+                parent_generator.table_index().incremented(),
+                last_child.get_bound()
+            );
+        }
+    }
+
+    /// Check the property:
+    ///     On a valid fork, the bound of the parent does not change.
+    pub fn prop_fork_parent_bound_table_index<G: AesBlockCipher>() {
+        for _ in 0..REPEATS {
+            let (t, nc, nb, i) = any_valid_fork().next().unwrap();
+            let k = any_key().next().unwrap();
+            let original_generator =
+                AesCtrGenerator::<G>::new(k, Some(t), Some(t.increased(nc.0 * nb.0 + i)));
+            let mut forked_generator = original_generator.clone();
+            forked_generator.try_fork(nc, nb).unwrap().last().unwrap();
+            assert_eq!(original_generator.get_bound(), forked_generator.get_bound());
+        }
+    }
+
+    /// Check the property:
+    ///     On a valid fork, the parent table index is increased of the number of children
+    ///     multiplied by the number of bytes per child.
+    pub fn prop_fork_parent_state_table_index<G: AesBlockCipher>() {
+        for _ in 0..REPEATS {
+            let (t, nc, nb, i) = any_valid_fork().next().unwrap();
+            let k = any_key().next().unwrap();
+            let original_generator =
+                AesCtrGenerator::<G>::new(k, Some(t), Some(t.increased(nc.0 * nb.0 + i)));
+            let mut forked_generator = original_generator.clone();
+            forked_generator.try_fork(nc, nb).unwrap().last().unwrap();
+            assert_eq!(
+                forked_generator.table_index(),
+                // Decrement accounts for the fact that the table index stored is the previous one
+                t.increased(nc.0 * nb.0).decremented()
+            );
+        }
+    }
+
+    /// Check the property:
+    ///     On a valid fork, the bytes outputted by the children in the fork order form the same
+    ///     sequence the parent would have had yielded no fork had happened.
+    pub fn prop_fork<G: AesBlockCipher>() {
+        for _ in 0..1000 {
+            let (t, nc, nb, i) = any_valid_fork().next().unwrap();
+            let k = any_key().next().unwrap();
+            let bytes_to_go = nc.0 * nb.0;
+            let original_generator =
+                AesCtrGenerator::<G>::new(k, Some(t), Some(t.increased(nc.0 * nb.0 + i)));
+            let mut forked_generator = original_generator.clone();
+            let initial_output: Vec<u8> = original_generator.take(bytes_to_go).collect();
+            let forked_output: Vec<u8> = forked_generator
+                .try_fork(nc, nb)
+                .unwrap()
+                .flat_map(|child| child.collect::<Vec<_>>())
+                .collect();
+            assert_eq!(initial_output, forked_output);
+        }
+    }
+
+    /// Check the property:
+    ///     On a valid fork, all children got a number of remaining bytes equals to the number of
+    ///     bytes per child given as fork input.
+    pub fn prop_fork_children_remaining_bytes<G: AesBlockCipher>() {
+        for _ in 0..REPEATS {
+            let (t, nc, nb, i) = any_valid_fork().next().unwrap();
+            let k = any_key().next().unwrap();
+            let mut generator =
+                AesCtrGenerator::<G>::new(k, Some(t), Some(t.increased(nc.0 * nb.0 + i)));
+            assert!(generator
+                .try_fork(nc, nb)
+                .unwrap()
+                .all(|c| c.remaining_bytes().0 == nb.0 as u128));
+        }
+    }
+
+    /// Check the property:
+    ///     On a valid fork, the number of remaining bybtes of the parent is reduced by the number
+    ///     of children multiplied by the number of bytes per child.
+    pub fn prop_fork_parent_remaining_bytes<G: AesBlockCipher>() {
+        for _ in 0..REPEATS {
+            let (t, nc, nb, i) = any_valid_fork().next().unwrap();
+            let k = any_key().next().unwrap();
+            let bytes_to_go = nc.0 * nb.0;
+            let mut generator =
+                AesCtrGenerator::<G>::new(k, Some(t), Some(t.increased(nc.0 * nb.0 + i)));
+            let before_remaining_bytes = generator.remaining_bytes();
+            let _ = generator.try_fork(nc, nb).unwrap();
+            let after_remaining_bytes = generator.remaining_bytes();
+            assert_eq!(
+                before_remaining_bytes.0 - after_remaining_bytes.0,
+                bytes_to_go as u128
+            );
+        }
+    }
+}

concrete-csprng/src/generators/aes_ctr/index.rs

@@ -0,0 +1,389 @@
+use crate::generators::aes_ctr::BYTES_PER_AES_CALL;
+use crate::generators::ByteCount;
+use std::cmp::Ordering;
+
+/// A structure representing an [aes index](#coarse-grained-pseudo-random-table-lookup).
+#[derive(Clone, Copy, Debug, PartialOrd, Ord, PartialEq, Eq)]
+pub struct AesIndex(pub u128);
+
+/// A structure representing a [byte index](#fine-grained-pseudo-random-table-lookup).
+#[derive(Clone, Copy, Debug, PartialOrd, Ord, PartialEq, Eq)]
+pub struct ByteIndex(pub usize);
+
+/// A structure representing a [table index](#fine-grained-pseudo-random-table-lookup)
+#[derive(Clone, Copy, Debug)]
+pub struct TableIndex {
+    pub(crate) aes_index: AesIndex,
+    pub(crate) byte_index: ByteIndex,
+}
+
+impl TableIndex {
+    /// The first table index.
+    pub const FIRST: TableIndex = TableIndex {
+        aes_index: AesIndex(0),
+        byte_index: ByteIndex(0),
+    };
+
+    /// The second table index.
+    pub const SECOND: TableIndex = TableIndex {
+        aes_index: AesIndex(0),
+        byte_index: ByteIndex(1),
+    };
+
+    /// The last table index.
+    pub const LAST: TableIndex = TableIndex {
+        aes_index: AesIndex(u128::MAX),
+        byte_index: ByteIndex(BYTES_PER_AES_CALL - 1),
+    };
+
+    /// Creates a table index from an aes index and a byte index.
+    #[allow(unused)] // to please clippy when tests are not activated
+    pub fn new(aes_index: AesIndex, byte_index: ByteIndex) -> Self {
+        assert!(byte_index.0 < BYTES_PER_AES_CALL);
+        TableIndex {
+            aes_index,
+            byte_index,
+        }
+    }
+
+    /// Shifts the table index forward of `shift` bytes.
+    pub fn increase(&mut self, shift: usize) {
+        // Compute full shifts to avoid overflows
+        let full_aes_shifts = shift / BYTES_PER_AES_CALL;
+        let shift_remainder = shift % BYTES_PER_AES_CALL;
+
+        // Get the additional shift if any
+        let new_byte_index = self.byte_index.0 + shift_remainder;
+        let full_aes_shifts = full_aes_shifts + new_byte_index / BYTES_PER_AES_CALL;
+
+        // Store the reaminder in the byte index
+        self.byte_index.0 = new_byte_index % BYTES_PER_AES_CALL;
+
+        self.aes_index.0 = self.aes_index.0.wrapping_add(full_aes_shifts as u128);
+    }
+
+    /// Shifts the table index backward of `shift` bytes.
+    pub fn decrease(&mut self, shift: usize) {
+        let remainder = shift % BYTES_PER_AES_CALL;
+        if remainder <= self.byte_index.0 {
+            self.aes_index.0 = self
+                .aes_index
+                .0
+                .wrapping_sub((shift / BYTES_PER_AES_CALL) as u128);
+            self.byte_index.0 -= remainder;
+        } else {
+            self.aes_index.0 = self
+                .aes_index
+                .0
+                .wrapping_sub((shift / BYTES_PER_AES_CALL) as u128 + 1);
+            self.byte_index.0 += BYTES_PER_AES_CALL - remainder;
+        }
+    }
+
+    /// Shifts the table index forward of one byte.
+    pub fn increment(&mut self) {
+        self.increase(1)
+    }
+
+    /// Shifts the table index backward of one byte.
+    pub fn decrement(&mut self) {
+        self.decrease(1)
+    }
+
+    /// Returns the table index shifted forward by `shift` bytes.
+    pub fn increased(mut self, shift: usize) -> Self {
+        self.increase(shift);
+        self
+    }
+
+    /// Returns the table index shifted backward by `shift` bytes.
+    #[allow(unused)] // to please clippy when tests are not activated
+    pub fn decreased(mut self, shift: usize) -> Self {
+        self.decrease(shift);
+        self
+    }
+
+    /// Returns the table index to the next byte.
+    pub fn incremented(mut self) -> Self {
+        self.increment();
+        self
+    }
+
+    /// Returns the table index to the previous byte.
+    pub fn decremented(mut self) -> Self {
+        self.decrement();
+        self
+    }
+
+    /// Returns the distance between two table indices in bytes.
+    ///
+    /// Note:
+    /// -----
+    ///
+    /// This method assumes that the `larger` input is, well, larger than the `smaller` input. If
+    /// this is not the case, the method returns `None`. Also, note that `ByteCount` uses the
+    /// `u128` datatype to store the byte count. Unfortunately, the number of bytes between two
+    /// table indices is in ⟦0;2¹³² -1⟧. When the distance is greater than 2¹²⁸ - 1, we saturate
+    /// the count at 2¹²⁸ - 1.
+    pub fn distance(larger: &Self, smaller: &Self) -> Option<ByteCount> {
+        match std::cmp::Ord::cmp(larger, smaller) {
+            Ordering::Less => None,
+            Ordering::Equal => Some(ByteCount(0)),
+            Ordering::Greater => {
+                let mut result = larger.aes_index.0 - smaller.aes_index.0;
+                result = result.saturating_mul(BYTES_PER_AES_CALL as u128);
+                result = result.saturating_add(larger.byte_index.0 as u128);
+                result = result.saturating_sub(smaller.byte_index.0 as u128);
+                Some(ByteCount(result))
+            }
+        }
+    }
+}
+
+impl Eq for TableIndex {}
+
+impl PartialEq<Self> for TableIndex {
+    fn eq(&self, other: &Self) -> bool {
+        matches!(self.partial_cmp(other), Some(Ordering::Equal))
+    }
+}
+
+impl PartialOrd<Self> for TableIndex {
+    fn partial_cmp(&self, other: &Self) -> Option<Ordering> {
+        Some(self.cmp(other))
+    }
+}
+
+impl Ord for TableIndex {
+    fn cmp(&self, other: &Self) -> Ordering {
+        match self.aes_index.cmp(&other.aes_index) {
+            Ordering::Equal => self.byte_index.cmp(&other.byte_index),
+            other => other,
+        }
+    }
+}
+
+#[cfg(test)]
+mod test {
+    use super::*;
+    use rand::{thread_rng, Rng};
+
+    const REPEATS: usize = 1_000_000;
+
+    fn any_table_index() -> impl Iterator<Item = TableIndex> {
+        std::iter::repeat_with(|| {
+            TableIndex::new(
+                AesIndex(thread_rng().gen()),
+                ByteIndex(thread_rng().gen::<usize>() % BYTES_PER_AES_CALL),
+            )
+        })
+    }
+
+    fn any_usize() -> impl Iterator<Item = usize> {
+        std::iter::repeat_with(|| thread_rng().gen())
+    }
+
+    #[test]
+    #[should_panic]
+    /// Verifies that the constructor of `TableIndex` panics when the byte index is too large.
+    fn test_table_index_new_panic() {
+        TableIndex::new(AesIndex(12), ByteIndex(144));
+    }
+
+    #[test]
+    /// Verifies that the `TableIndex` wraps nicely with predecessor
+    fn test_table_index_predecessor_edge() {
+        assert_eq!(TableIndex::FIRST.decremented(), TableIndex::LAST);
+    }
+
+    #[test]
+    /// Verifies that the `TableIndex` wraps nicely with successor
+    fn test_table_index_successor_edge() {
+        assert_eq!(TableIndex::LAST.incremented(), TableIndex::FIRST);
+    }
+
+    #[test]
+    /// Check that the table index distance saturates nicely.
+    fn prop_table_index_distance_saturates() {
+        assert_eq!(
+            TableIndex::distance(&TableIndex::LAST, &TableIndex::FIRST)
+                .unwrap()
+                .0,
+            u128::MAX
+        )
+    }
+
+    #[test]
+    /// Check the property:
+    ///     For all table indices t,
+    ///         distance(t, t) = Some(0).
+    fn prop_table_index_distance_zero() {
+        for _ in 0..REPEATS {
+            let t = any_table_index().next().unwrap();
+            assert_eq!(TableIndex::distance(&t, &t), Some(ByteCount(0)));
+        }
+    }
+
+    #[test]
+    /// Check the property:
+    ///     For all table indices t1, t2 such that t1 < t2,
+    ///         distance(t1, t2) = None.
+    fn prop_table_index_distance_wrong_order_none() {
+        for _ in 0..REPEATS {
+            let (t1, t2) = any_table_index()
+                .zip(any_table_index())
+                .find(|(t1, t2)| t1 < t2)
+                .unwrap();
+            assert_eq!(TableIndex::distance(&t1, &t2), None);
+        }
+    }
+
+    #[test]
+    /// Check the property:
+    ///     For all table indices t1, t2 such that t1 > t2,
+    ///         distance(t1, t2) = Some(v) where v is strictly positive.
+    fn prop_table_index_distance_some_positive() {
+        for _ in 0..REPEATS {
+            let (t1, t2) = any_table_index()
+                .zip(any_table_index())
+                .find(|(t1, t2)| t1 > t2)
+                .unwrap();
+            assert!(matches!(TableIndex::distance(&t1, &t2), Some(ByteCount(v)) if v > 0));
+        }
+    }
+
+    #[test]
+    /// Check the property:
+    ///     For all table indices t, positive i such that i < distance (MAX, t) with MAX the largest
+    ///     table index,
+    ///         distance(t.increased(i), t) = Some(i).
+    fn prop_table_index_distance_increase() {
+        for _ in 0..REPEATS {
+            let (t, inc) = any_table_index()
+                .zip(any_usize())
+                .find(|(t, inc)| {
+                    (*inc as u128) < TableIndex::distance(&TableIndex::LAST, t).unwrap().0
+                })
+                .unwrap();
+            assert_eq!(
+                TableIndex::distance(&t.increased(inc), &t).unwrap().0 as usize,
+                inc
+            );
+        }
+    }
+
+    #[test]
+    /// Check the property:
+    ///     For all table indices t, t =? t = true.
+    fn prop_table_index_equality() {
+        for _ in 0..REPEATS {
+            let t = any_table_index().next().unwrap();
+            assert_eq!(
+                std::cmp::PartialOrd::partial_cmp(&t, &t),
+                Some(std::cmp::Ordering::Equal)
+            );
+        }
+    }
+
+    #[test]
+    /// Check the property:
+    ///     For all table indices t, positive i such that i < distance (MAX, t) with MAX the largest
+    ///     table index,
+    ///         t.increased(i) >? t = true.
+    fn prop_table_index_greater() {
+        for _ in 0..REPEATS {
+            let (t, inc) = any_table_index()
+                .zip(any_usize())
+                .find(|(t, inc)| {
+                    (*inc as u128) < TableIndex::distance(&TableIndex::LAST, t).unwrap().0
+                })
+                .unwrap();
+            assert_eq!(
+                std::cmp::PartialOrd::partial_cmp(&t.increased(inc), &t),
+                Some(std::cmp::Ordering::Greater),
+            );
+        }
+    }
+
+    #[test]
+    /// Check the property:
+    ///     For all table indices t, positive i such that i < distance (t, 0) with MAX the largest
+    ///     table index,
+    ///         t.decreased(i) <? t = true.
+    fn prop_table_index_less() {
+        for _ in 0..REPEATS {
+            let (t, inc) = any_table_index()
+                .zip(any_usize())
+                .find(|(t, inc)| {
+                    (*inc as u128) < TableIndex::distance(t, &TableIndex::FIRST).unwrap().0
+                })
+                .unwrap();
+            assert_eq!(
+                std::cmp::PartialOrd::partial_cmp(&t.decreased(inc), &t),
+                Some(std::cmp::Ordering::Less)
+            );
+        }
+    }
+
+    #[test]
+    /// Check the property:
+    ///     For all table indices t,
+    ///         successor(predecessor(t)) = t.
+    fn prop_table_index_decrement_increment() {
+        for _ in 0..REPEATS {
+            let t = any_table_index().next().unwrap();
+            assert_eq!(t.decremented().incremented(), t);
+        }
+    }
+
+    #[test]
+    /// Check the property:
+    ///     For all table indices t,
+    ///         predecessor(successor(t)) = t.
+    fn prop_table_index_increment_decrement() {
+        for _ in 0..REPEATS {
+            let t = any_table_index().next().unwrap();
+            assert_eq!(t.incremented().decremented(), t);
+        }
+    }
+
+    #[test]
+    /// Check the property:
+    ///     For all table indices t, positive integer i,
+    ///         increase(decrease(t, i), i) = t.
+    fn prop_table_index_increase_decrease() {
+        for _ in 0..REPEATS {
+            let (t, i) = any_table_index().zip(any_usize()).next().unwrap();
+            assert_eq!(t.increased(i).decreased(i), t);
+        }
+    }
+
+    #[test]
+    /// Check the property:
+    ///     For all table indices t, positive integer i,
+    ///         decrease(increase(t, i), i) = t.
+    fn prop_table_index_decrease_increase() {
+        for _ in 0..REPEATS {
+            let (t, i) = any_table_index().zip(any_usize()).next().unwrap();
+            assert_eq!(t.decreased(i).increased(i), t);
+        }
+    }
+
+    #[test]
+    /// Check that a big increase does not overflow
+    fn prop_table_increase_max_no_overflow() {
+        let first = TableIndex::FIRST;
+        // Increase so that ByteIndex is at 1usize
+        let second = first.increased(1);
+
+        // Now increase by usize::MAX, as the underlying byte index stores a usize this may overflow
+        // depending on implementation, ensure it does not overflow
+        let big_increase = second.increased(usize::MAX);
+        let total_full_aes_shifts = (1u128 + usize::MAX as u128) / BYTES_PER_AES_CALL as u128;
+
+        assert_eq!(
+            big_increase,
+            TableIndex::new(AesIndex(total_full_aes_shifts), ByteIndex(0))
+        );
+    }
+}

concrete-csprng/src/generators/aes_ctr/mod.rs

@@ -0,0 +1,223 @@
+//! A module implementing the random generator api with batched aes calls.
+//!
+//! This module provides a generic [`AesCtrGenerator`] structure which implements the
+//! [`super::RandomGenerator`] api using the AES block cipher in counter mode. That is, the
+//! generator holds a state (i.e. counter) which is incremented iteratively, to produce the stream
+//! of random values:
+//! ```ascii
+//!        state=0        state=1        state=2
+//!        ╔══↧══╗        ╔══↧══╗        ╔══↧══╗
+//!    key ↦ AES ║    key ↦ AES ║    key ↦ AES ║ ...
+//!        ╚══↧══╝        ╚══↧══╝        ╚══↧══╝
+//!        output0        output1        output2
+//!
+//!          t=0            t=1            t=2
+//! ```
+//!
+//! The [`AesCtrGenerator`] structure is generic over the AES block ciphers, which are
+//! represented by the [`AesBlockCipher`] trait. Consequently, implementers only need to implement
+//! the `AesBlockCipher` trait, to benefit from the whole api of the `AesCtrGenerator` structure.
+//!
+//! In the following section, we give details on the implementation of this generic generator.
+//!
+//! Coarse-grained pseudo-random lookup table
+//! =========================================
+//!
+//! To generate random values, we use the AES block cipher in counter mode. If we denote f the aes
+//! encryption function, we have:
+//! ```ascii
+//!     f: ⟦0;2¹²⁸ -1⟧ X ⟦0;2¹²⁸ -1⟧ ↦ ⟦0;2¹²⁸ -1⟧
+//!     f(secret_key, input) ↦ output
+//! ```
+
+//! If we fix the secret key to a value k, we have a function fₖ from ⟦0;2¹²⁸ -1⟧ to ⟦0;2¹²⁸-1⟧,
+//! transforming the state of the counter into a pseudo random value. Essentially, this fₖ
+//! function can be considered as a the following lookup table, containing 2¹²⁸ pseudo-random
+//! values:
+//! ```ascii  
+//!     ╭──────────────┬──────────────┬─────┬──────────────╮
+//!     │       0      │       1      │     │    2¹²⁸ -1   │
+//!     ├──────────────┼──────────────┼─────┼──────────────┤
+//!     │     fₖ(0)    │     fₖ(1)    │     │  fₖ(2¹²⁸ -1) │
+//!     ╔═══════↧══════╦═══════↧══════╦═════╦═══════↧══════╗
+//!     ║┏━━━━━━━━━━━━┓║┏━━━━━━━━━━━━┓║     ║┏━━━━━━━━━━━━┓║
+//!     ║┃    u128    ┃║┃    u128    ┃║ ... ║┃    u128    ┃║
+//!     ║┗━━━━━━━━━━━━┛║┗━━━━━━━━━━━━┛║     ║┗━━━━━━━━━━━━┛║
+//!     ╚══════════════╩══════════════╩═════╩══════════════╝
+//! ```
+//!
+//! An input to the fₖ function is called an _aes index_ (also called state or counter in the
+//! standards) of the pseudo-random table. The [`AesIndex`] structure defined in this module
+//! represents such an index in the code.
+//!
+//! Fine-grained pseudo-random table lookup
+//! =======================================
+//!
+//! Since we want to deliver the pseudo-random bytes one by one, we have to come with a finer
+//! grained indexing. Fortunately, each `u128` value outputted by fₖ can be seen as a table of 16
+//! `u8`:
+//! ```ascii
+//!     ╭──────────────┬──────────────┬─────┬──────────────╮
+//!     │       0      │       1      │     │    2¹²⁸ -1   │
+//!     ├──────────────┼──────────────┼─────┼──────────────┤
+//!     │     fₖ(0)    │     fₖ(1)    │     │  fₖ(2¹²⁸ -1) │
+//!     ╔═══════↧══════╦═══════↧══════╦═════╦═══════↧══════╗
+//!     ║┏━━━━━━━━━━━━┓║┏━━━━━━━━━━━━┓║     ║┏━━━━━━━━━━━━┓║
+//!     ║┃    u128    ┃║┃    u128    ┃║     ║┃    u128    ┃║
+//!     ║┣━━┯━━┯━━━┯━━┫║┣━━┯━━┯━━━┯━━┫║ ... ║┣━━┯━━┯━━━┯━━┫║
+//!     ║┃u8│u8│...│u8┃║┃u8│u8│...│u8┃║     ║┃u8│u8│...│u8┃║
+//!     ║┗━━┷━━┷━━━┷━━┛║┗━━┷━━┷━━━┷━━┛║     ║┗━━┷━━┷━━━┷━━┛║
+//!     ╚══════════════╩══════════════╩═════╩══════════════╝
+//! ```
+//!
+//! We introduce a second function to select a chunk of 8 bits:
+//! ```ascii
+//!     g: ⟦0;2¹²⁸ -1⟧ X ⟦0;15⟧ ↦ ⟦0;2⁸ -1⟧
+//!     g(big_int, index) ↦ byte
+//! ```
+//!
+//! If we fix the `u128` value to a value e, we have a function gₑ from ⟦0;15⟧ to ⟦0;2⁸ -1⟧
+//! transforming an index into a pseudo-random byte:
+//! ```ascii
+//!     ┏━━━━━━━━┯━━━━━━━━┯━━━┯━━━━━━━━┓
+//!     ┃   u8   │   u8   │...│   u8   ┃
+//!     ┗━━━━━━━━┷━━━━━━━━┷━━━┷━━━━━━━━┛
+//!     │  gₑ(0) │  gₑ(1) │   │ gₑ(15) │
+//!     ╰────────┴─────-──┴───┴────────╯
+//! ```
+//!
+//! We call this input to the gₑ function, a _byte index_ of the pseudo-random table. The
+//! [`ByteIndex`] structure defined in this module represents such an index in the code.
+//!
+//! By using both the g and the fₖ functions, we can define a new function l which allows to index
+//! any byte of the pseudo-random table:
+//! ```ascii
+//!     l: ⟦0;2¹²⁸ -1⟧ X ⟦0;15⟧ ↦ ⟦0;2⁸ -1⟧
+//!     l(aes_index, byte_index) ↦ g(fₖ(aes_index), byte_index)
+//! ```
+//!
+//! In this sense, any member of ⟦0;2¹²⁸ -1⟧ X ⟦0;15⟧ uniquely defines a byte in this pseudo-random
+//! table:
+//! ```ascii
+//!                          e = fₖ(a)
+//!     ╔══════════════╦═══════↧══════╦═════╦══════════════╗
+//!     ║┏━━━━━━━━━━━━┓║┏━━━━━━━━━━━━┓║     ║┏━━━━━━━━━━━━┓║
+//!     ║┃    u128    ┃║┃    u128    ┃║     ║┃    u128    ┃║
+//!     ║┣━━┯━━┯━━━┯━━┫║┣━━┯━━┯━━━┯━━┫║ ... ║┣━━┯━━┯━━━┯━━┫║
+//!     ║┃u8│u8│...│u8┃║┃u8│u8│...│u8┃║     ║┃u8│u8│...│u8┃║
+//!     ║┗━━┷━━┷━━━┷━━┛║┗━━┷↥━┷━━━┷━━┛║     ║┗━━┷━━┷━━━┷━━┛║
+//!     ║              ║│    gₑ(b)   │║     ║              ║
+//!     ║              ║╰───-────────╯║     ║              ║
+//!     ╚══════════════╩══════════════╩═════╩══════════════╝
+//! ```
+//!
+//! We call this input to the l function, a _table index_ of the pseudo-random table. The
+//! [`TableIndex`] structure defined in this module represents such an index in the code.
+//!
+//! Prngs current table index
+//! =========================
+//!
+//! When created, a prng is given an initial _table index_, denoted (a₀, b₀), which identifies the
+//! first byte of the table to be outputted by the prng. Then, each time the prng is queried for a
+//! new value, the byte corresponding to the current _table index_ is returned, and the current
+//! _table index_ is incremented:
+//! ```ascii
+//!       e = fₖ(a₀)                                                  e = fₖ(a₁)
+//!     ╔═════↧═════╦═══════════╦═════╦═══════════╗     ╔═══════════╦═════↧═════╦═════╦═══════════╗
+//!     ║┏━┯━┯━━━┯━┓║┏━┯━┯━━━┯━┓║ ... ║┏━┯━┯━━━┯━┓║     ║┏━┯━┯━━━┯━┓║┏━┯━┯━━━┯━┓║ ... ║┏━┯━┯━━━┯━┓║
+//!     ║┃ │ │...│ ┃║┃ │ │...│ ┃║     ║┃ │ │...│ ┃║     ║┃ │ │...│ ┃║┃ │ │...│ ┃║     ║┃ │ │...│ ┃║
+//!     ║┗━┷━┷━━━┷↥┛║┗━┷━┷━━━┷━┛║     ║┗━┷━┷━━━┷━┛║  →  ║┗━┷━┷━━━┷━┛║┗↥┷━┷━━━┷━┛║     ║┗━┷━┷━━━┷━┛║
+//!     ║│  gₑ(b₀) │║           ║     ║           ║     ║           ║│  gₑ(b₁) │║     ║           ║
+//!     ║╰─────────╯║           ║     ║           ║     ║           ║╰─────────╯║     ║           ║
+//!     ╚═══════════╩═══════════╩═════╩═══════════╝     ╚═══════════╩═══════════╩═════╩═══════════╝
+//! ```
+//!
+//! Prng bound
+//! ==========
+//!
+//! When created, a prng is also given a _bound_ (aₘ, bₘ) , that is a table index which it is not
+//! allowed to exceed:
+//! ```ascii
+//!       e = fₖ(a₀)
+//!     ╔═════↧═════╦═══════════╦═════╦═══════════╗
+//!     ║┏━┯━┯━━━┯━┓║┏━┯━┯━━━┯━┓║ ... ║┏━┯━┯━━━┯━┓║
+//!     ║┃ │ │...│ ┃║┃ │╳│...│╳┃║     ║┃╳│╳│...│╳┃║
+//!     ║┗━┷━┷━━━┷↥┛║┗━┷━┷━━━┷━┛║     ║┗━┷━┷━━━┷━┛║ The current byte can be returned.
+//!     ║│  gₑ(b₀) │║           ║     ║           ║
+//!     ║╰─────────╯║           ║     ║           ║
+//!     ╚═══════════╩═══════════╩═════╩═══════════╝
+//!     
+//!                   e = fₖ(aₘ)
+//!     ╔═══════════╦═════↧═════╦═════╦═══════════╗
+//!     ║┏━┯━┯━━━┯━┓║┏━┯━┯━━━┯━┓║ ... ║┏━┯━┯━━━┯━┓║
+//!     ║┃ │ │...│ ┃║┃ │╳│...│╳┃║     ║┃╳│╳│...│╳┃║ The table index reached the bound,
+//!     ║┗━┷━┷━━━┷━┛║┗━┷↥┷━━━┷━┛║     ║┗━┷━┷━━━┷━┛║ the current byte can not be
+//!     ║           ║│  gₑ(bₘ) │║     ║           ║ returned.
+//!     ║           ║╰─────────╯║     ║           ║
+//!     ╚═══════════╩═══════════╩═════╩═══════════╝
+//! ```
+//!
+//! Buffering
+//! =========
+//!
+//! Calling the aes function every time we need to output a single byte would be a huge waste of
+//! resources. In practice, we call aes 8 times in a row, for 8 successive values of aes index, and
+//! store the results in a buffer. For platforms which have a dedicated aes chip, this allows to
+//! fill the unit pipeline and reduces the amortized cost of the aes function.
+//!
+//! Together with the current table index of the prng, we also store a pointer p (initialized at
+//! p₀=b₀) to the current byte in the buffer. If we denote v the lookup function we have :
+//! ```ascii
+//!                        e = fₖ(a₀)                         Buffer(length=128)
+//!     ╔═════╦═══════════╦═════↧═════╦═══════════╦═════╗  ┏━┯━┯━┯━┯━┯━┯━┯━┯━━━┯━┓
+//!     ║ ... ║┏━┯━┯━━━┯━┓║┏━┯━┯━━━┯━┓║┏━┯━┯━━━┯━┓║ ... ║  ┃▓│▓│▓│▓│▓│▓│▓│▓│...│▓┃
+//!     ║     ║┃ │ │...│ ┃║┃▓│▓│...│▓┃║┃▓│▓│...│▓┃║     ║  ┗━┷↥┷━┷━┷━┷━┷━┷━┷━━━┷━┛
+//!     ║     ║┗━┷━┷━━━┷━┛║┗━┷↥┷━━━┷━┛║┗━┷━┷━━━┷━┛║     ║  │ v(p₀)               │
+//!     ║     ║           ║│  gₑ(b₀) │║           ║     ║  ╰─────────────────────╯
+//!     ║     ║           ║╰─────────╯║           ║     ║
+//!     ╚═════╩═══════════╩═══════════╩═══════════╩═════╝
+//! ```
+//!
+//! We call this input to the v function, a _buffer pointer_. The [`BufferPointer`] structure
+//! defined in this module represents such a pointer in the code.
+//!
+//! When the table index is incremented, the buffer pointer is incremented alongside:
+//! ```ascii
+//!                        e = fₖ(a)                          Buffer(length=128)
+//!     ╔═════╦═══════════╦═════↧═════╦═══════════╦═════╗  ┏━┯━┯━┯━┯━┯━┯━┯━┯━━━┯━┓
+//!     ║ ... ║┏━┯━┯━━━┯━┓║┏━┯━┯━━━┯━┓║┏━┯━┯━━━┯━┓║ ... ║  ┃▓│▓│▓│▓│▓│▓│▓│▓│...│▓┃
+//!     ║     ║┃ │ │...│ ┃║┃▓│▓│...│▓┃║┃▓│▓│...│▓┃║     ║  ┗━┷━┷↥┷━┷━┷━┷━┷━┷━━━┷━┛
+//!     ║     ║┗━┷━┷━━━┷━┛║┗━┷━┷↥━━┷━┛║┗━┷━┷━━━┷━┛║     ║  │   v(p)              │
+//!     ║     ║           ║│  gₑ(b)  │║           ║     ║  ╰─────────────────────╯
+//!     ║     ║           ║╰─────────╯║           ║     ║
+//!     ╚═════╩═══════════╩═══════════╩═══════════╩═════╝
+//! ```
+//!
+//! When the buffer pointer is incremented it is checked against the size of the buffer, and if
+//! necessary, a new batch of aes index values is generated.
+
+pub const AES_CALLS_PER_BATCH: usize = 8;
+pub const BYTES_PER_AES_CALL: usize = 128 / 8;
+pub const BYTES_PER_BATCH: usize = BYTES_PER_AES_CALL * AES_CALLS_PER_BATCH;
+
+/// A module containing structures to manage table indices.
+mod index;
+pub use index::*;
+
+/// A module containing structures to manage table indices and buffer pointers together properly.
+mod states;
+pub use states::*;
+
+/// A module containing an abstraction for aes block ciphers.
+mod block_cipher;
+pub use block_cipher::*;
+
+/// A module containing a generic implementation of a random generator.
+mod generic;
+pub use generic::*;
+
+/// A module extending `generic` to the `rayon` paradigm.
+#[cfg(feature = "parallel")]
+mod parallel;
+#[cfg(feature = "parallel")]
+pub use parallel::*;

concrete-csprng/src/generators/aes_ctr/parallel.rs

@@ -0,0 +1,222 @@
+use crate::generators::aes_ctr::{
+    AesBlockCipher, AesCtrGenerator, ChildrenClosure, State, TableIndex,
+};
+use crate::generators::{BytesPerChild, ChildrenCount, ForkError};
+
+/// A type alias for the parallel children iterator type.
+pub type ParallelChildrenIterator<BlockCipher> = rayon::iter::Map<
+    rayon::iter::Zip<
+        rayon::range::Iter<usize>,
+        rayon::iter::RepeatN<(Box<BlockCipher>, TableIndex, BytesPerChild)>,
+    >,
+    fn((usize, (Box<BlockCipher>, TableIndex, BytesPerChild))) -> AesCtrGenerator<BlockCipher>,
+>;
+
+impl<BlockCipher: AesBlockCipher> AesCtrGenerator<BlockCipher> {
+    /// Tries to fork the current generator into `n_child` generators each able to output
+    /// `child_bytes` random bytes as a parallel iterator.
+    ///
+    /// # Notes
+    ///
+    /// This method necessitate the "multithread" feature.
+    pub fn par_try_fork(
+        &mut self,
+        n_children: ChildrenCount,
+        n_bytes: BytesPerChild,
+    ) -> Result<ParallelChildrenIterator<BlockCipher>, ForkError>
+    where
+        BlockCipher: Send + Sync,
+    {
+        use rayon::prelude::*;
+
+        if n_children.0 == 0 {
+            return Err(ForkError::ZeroChildrenCount);
+        }
+        if n_bytes.0 == 0 {
+            return Err(ForkError::ZeroBytesPerChild);
+        }
+        if !self.is_fork_in_bound(n_children, n_bytes) {
+            return Err(ForkError::ForkTooLarge);
+        }
+
+        // The state currently stored in the parent generator points to the table index of the last
+        // generated byte. The first index to be generated is the next one :
+        let first_index = self.state.table_index().incremented();
+        let output = (0..n_children.0)
+            .into_par_iter()
+            .zip(rayon::iter::repeatn(
+                (self.block_cipher.clone(), first_index, n_bytes),
+                n_children.0,
+            ))
+            .map(
+                // This map is a little weird because we need to cast the closure to a fn pointer
+                // that matches the signature of `ChildrenIterator<BlockCipher>`. Unfortunately,
+                // the compiler does not manage to coerce this one automatically.
+                (|(i, (block_cipher, first_index, n_bytes))| {
+                    // The first index to be outputted by the child is the `first_index` shifted by
+                    // the proper amount of `child_bytes`.
+                    let child_first_index = first_index.increased(n_bytes.0 * i);
+                    // The bound of the child is the first index of its next sibling.
+                    let child_bound_index = first_index.increased(n_bytes.0 * (i + 1));
+                    AesCtrGenerator::from_block_cipher(
+                        block_cipher,
+                        child_first_index,
+                        child_bound_index,
+                    )
+                }) as ChildrenClosure<BlockCipher>,
+            );
+        // The parent next index is the bound of the last child.
+        let next_index = first_index.increased(n_bytes.0 * n_children.0);
+        self.state = State::new(next_index);
+
+        Ok(output)
+    }
+}
+
+#[cfg(test)]
+pub mod aes_ctr_parallel_generic_tests {
+
+    use super::*;
+    use crate::generators::aes_ctr::aes_ctr_generic_test::{any_key, any_valid_fork};
+    use rayon::prelude::*;
+
+    const REPEATS: usize = 1_000_000;
+
+    /// Check the property:
+    ///     On a valid fork, the table index of the first child is the same as the table index of
+    ///     the parent before the fork.
+    pub fn prop_fork_first_state_table_index<G: AesBlockCipher>() {
+        for _ in 0..REPEATS {
+            let (t, nc, nb, i) = any_valid_fork().next().unwrap();
+            let k = any_key().next().unwrap();
+            let original_generator =
+                AesCtrGenerator::<G>::new(k, Some(t), Some(t.increased(nc.0 * nb.0 + i)));
+            let mut forked_generator = original_generator.clone();
+            let first_child = forked_generator
+                .par_try_fork(nc, nb)
+                .unwrap()
+                .find_first(|_| true)
+                .unwrap();
+            assert_eq!(original_generator.table_index(), first_child.table_index());
+        }
+    }
+
+    /// Check the property:
+    ///     On a valid fork, the table index of the first byte outputted by the parent after the
+    ///     fork, is the bound of the last child of the fork.
+    pub fn prop_fork_last_bound_table_index<G: AesBlockCipher>() {
+        for _ in 0..REPEATS {
+            let (t, nc, nb, i) = any_valid_fork().next().unwrap();
+            let k = any_key().next().unwrap();
+            let mut parent_generator =
+                AesCtrGenerator::<G>::new(k, Some(t), Some(t.increased(nc.0 * nb.0 + i)));
+            let last_child = parent_generator
+                .par_try_fork(nc, nb)
+                .unwrap()
+                .find_last(|_| true)
+                .unwrap();
+            assert_eq!(
+                parent_generator.table_index().incremented(),
+                last_child.get_bound()
+            );
+        }
+    }
+
+    /// Check the property:
+    ///     On a valid fork, the bound of the parent does not change.
+    pub fn prop_fork_parent_bound_table_index<G: AesBlockCipher>() {
+        for _ in 0..REPEATS {
+            let (t, nc, nb, i) = any_valid_fork().next().unwrap();
+            let k = any_key().next().unwrap();
+            let original_generator =
+                AesCtrGenerator::<G>::new(k, Some(t), Some(t.increased(nc.0 * nb.0 + i)));
+            let mut forked_generator = original_generator.clone();
+            forked_generator
+                .par_try_fork(nc, nb)
+                .unwrap()
+                .find_last(|_| true)
+                .unwrap();
+            assert_eq!(original_generator.get_bound(), forked_generator.get_bound());
+        }
+    }
+
+    /// Check the property:
+    ///     On a valid fork, the parent table index is increased of the number of children
+    ///     multiplied by the number of bytes per child.
+    pub fn prop_fork_parent_state_table_index<G: AesBlockCipher>() {
+        for _ in 0..REPEATS {
+            let (t, nc, nb, i) = any_valid_fork().next().unwrap();
+            let k = any_key().next().unwrap();
+            let original_generator =
+                AesCtrGenerator::<G>::new(k, Some(t), Some(t.increased(nc.0 * nb.0 + i)));
+            let mut forked_generator = original_generator.clone();
+            forked_generator
+                .par_try_fork(nc, nb)
+                .unwrap()
+                .find_last(|_| true)
+                .unwrap();
+            assert_eq!(
+                forked_generator.table_index(),
+                // Decrement accounts for the fact that the table index stored is the previous one
+                t.increased(nc.0 * nb.0).decremented()
+            );
+        }
+    }
+
+    /// Check the property:
+    ///     On a valid fork, the bytes outputted by the children in the fork order form the same
+    ///     sequence the parent would have had outputted no fork had happened.
+    pub fn prop_fork<G: AesBlockCipher>() {
+        for _ in 0..1000 {
+            let (t, nc, nb, i) = any_valid_fork().next().unwrap();
+            let k = any_key().next().unwrap();
+            let bytes_to_go = nc.0 * nb.0;
+            let original_generator =
+                AesCtrGenerator::<G>::new(k, Some(t), Some(t.increased(nc.0 * nb.0 + i)));
+            let mut forked_generator = original_generator.clone();
+            let initial_output: Vec<u8> = original_generator.take(bytes_to_go).collect();
+            let forked_output: Vec<u8> = forked_generator
+                .par_try_fork(nc, nb)
+                .unwrap()
+                .flat_map(|child| child.collect::<Vec<_>>())
+                .collect();
+            assert_eq!(initial_output, forked_output);
+        }
+    }
+
+    /// Check the property:
+    ///     On a valid fork, all children got a number of remaining bytes equals to the number of
+    ///     bytes per child given as fork input.
+    pub fn prop_fork_children_remaining_bytes<G: AesBlockCipher>() {
+        for _ in 0..REPEATS {
+            let (t, nc, nb, i) = any_valid_fork().next().unwrap();
+            let k = any_key().next().unwrap();
+            let mut generator =
+                AesCtrGenerator::<G>::new(k, Some(t), Some(t.increased(nc.0 * nb.0 + i)));
+            assert!(generator
+                .par_try_fork(nc, nb)
+                .unwrap()
+                .all(|c| c.remaining_bytes().0 == nb.0 as u128));
+        }
+    }
+
+    /// Check the property:
+    ///     On a valid fork, the number of remaining bytes of the parent is reduced by the
+    ///     number of children multiplied by the number of bytes per child.
+    pub fn prop_fork_parent_remaining_bytes<G: AesBlockCipher>() {
+        for _ in 0..REPEATS {
+            let (t, nc, nb, i) = any_valid_fork().next().unwrap();
+            let k = any_key().next().unwrap();
+            let bytes_to_go = nc.0 * nb.0;
+            let mut generator =
+                AesCtrGenerator::<G>::new(k, Some(t), Some(t.increased(nc.0 * nb.0 + i)));
+            let before_remaining_bytes = generator.remaining_bytes();
+            let _ = generator.par_try_fork(nc, nb).unwrap();
+            let after_remaining_bytes = generator.remaining_bytes();
+            assert_eq!(
+                before_remaining_bytes.0 - after_remaining_bytes.0,
+                bytes_to_go as u128
+            );
+        }
+    }
+}

concrete-csprng/src/generators/aes_ctr/states.rs

@@ -0,0 +1,176 @@
+use crate::generators::aes_ctr::index::{AesIndex, TableIndex};
+use crate::generators::aes_ctr::BYTES_PER_BATCH;
+
+/// A pointer to the next byte to be outputted by the generator.
+#[derive(Clone, Copy, Debug, PartialOrd, Ord, PartialEq, Eq)]
+pub struct BufferPointer(pub usize);
+
+/// A structure representing the current state of generator using batched aes-ctr approach.
+#[derive(Debug, Clone, Copy)]
+pub struct State {
+    table_index: TableIndex,
+    buffer_pointer: BufferPointer,
+}
+
+/// A structure representing the action to be taken by the generator after shifting its state.
+#[derive(Clone, Debug, PartialEq, Eq)]
+pub enum ShiftAction {
+    /// Outputs the byte pointed to by the 0-th field.
+    OutputByte(BufferPointer),
+    /// Refresh the buffer starting from the 0-th field, and output the byte pointed to by the 0-th
+    /// field.
+    RefreshBatchAndOutputByte(AesIndex, BufferPointer),
+}
+
+impl State {
+    /// Creates a new state from the initial table index.
+    ///
+    /// Note :
+    /// ------
+    ///
+    /// The `table_index` input, is the __first__ table index that will be outputted on the next
+    /// call to `increment`. Put differently, the current table index of the newly created state
+    /// is the predecessor of this one.
+    pub fn new(table_index: TableIndex) -> Self {
+        // We ensure that the table index is not the first one, to prevent wrapping on `decrement`,
+        // and outputting `RefreshBatchAndOutputByte(AesIndex::MAX, ...)` on the first increment
+        // (which would lead to loading a non continuous batch).
+        assert_ne!(table_index, TableIndex::FIRST);
+        State {
+            // To ensure that the first outputted table index is the proper one, we decrement the
+            // table index.
+            table_index: table_index.decremented(),
+            // To ensure that the first `ShiftAction` will be a `RefreshBatchAndOutputByte`, we set
+            // the buffer to the last allowed value.
+            buffer_pointer: BufferPointer(BYTES_PER_BATCH - 1),
+        }
+    }
+
+    /// Shifts the state forward of `shift` bytes.
+    pub fn increase(&mut self, shift: usize) -> ShiftAction {
+        self.table_index.increase(shift);
+        let total_batch_index = self.buffer_pointer.0 + shift;
+        if total_batch_index > BYTES_PER_BATCH - 1 {
+            self.buffer_pointer.0 = self.table_index.byte_index.0;
+            ShiftAction::RefreshBatchAndOutputByte(self.table_index.aes_index, self.buffer_pointer)
+        } else {
+            self.buffer_pointer.0 = total_batch_index;
+            ShiftAction::OutputByte(self.buffer_pointer)
+        }
+    }
+
+    /// Shifts the state forward of one byte.
+    pub fn increment(&mut self) -> ShiftAction {
+        self.increase(1)
+    }
+
+    /// Returns the current table index.
+    pub fn table_index(&self) -> TableIndex {
+        self.table_index
+    }
+}
+
+impl Default for State {
+    fn default() -> Self {
+        State::new(TableIndex::FIRST)
+    }
+}
+
+#[cfg(test)]
+mod test {
+    use super::*;
+    use crate::generators::aes_ctr::index::ByteIndex;
+    use crate::generators::aes_ctr::BYTES_PER_AES_CALL;
+    use rand::{thread_rng, Rng};
+
+    const REPEATS: usize = 1_000_000;
+
+    fn any_table_index() -> impl Iterator<Item = TableIndex> {
+        std::iter::repeat_with(|| {
+            TableIndex::new(
+                AesIndex(thread_rng().gen()),
+                ByteIndex(thread_rng().gen::<usize>() % BYTES_PER_AES_CALL),
+            )
+        })
+    }
+
+    fn any_usize() -> impl Iterator<Item = usize> {
+        std::iter::repeat_with(|| thread_rng().gen())
+    }
+
+    #[test]
+    /// Check the property:
+    ///     For all table indices t,
+    ///         State::new(t).increment() = RefreshBatchAndOutputByte(t.aes_index, t.byte_index)
+    fn prop_state_new_increment() {
+        for _ in 0..REPEATS {
+            let (t, mut s) = any_table_index()
+                .map(|t| (t, State::new(t)))
+                .next()
+                .unwrap();
+            assert!(matches!(
+                s.increment(),
+                ShiftAction::RefreshBatchAndOutputByte(t_, BufferPointer(p_)) if t_ == t.aes_index && p_ == t.byte_index.0
+            ))
+        }
+    }
+
+    #[test]
+    /// Check the property:
+    ///     For all states s, table indices t, positive integer i
+    ///         if s = State::new(t), then t.increased(i) = s.increased(i-1).table_index().
+    fn prop_state_increase_table_index() {
+        for _ in 0..REPEATS {
+            let (t, mut s, i) = any_table_index()
+                .zip(any_usize())
+                .map(|(t, i)| (t, State::new(t), i))
+                .next()
+                .unwrap();
+            s.increase(i);
+            assert_eq!(s.table_index(), t.increased(i - 1))
+        }
+    }
+
+    #[test]
+    /// Check the property:
+    ///     For all table indices t, positive integer i such as t.byte_index + i < 127,
+    ///         if s = State::new(t), and s.increment() was executed, then
+    ///         s.increase(i) = OutputByte(t.byte_index + i).
+    fn prop_state_increase_small() {
+        for _ in 0..REPEATS {
+            let (t, mut s, i) = any_table_index()
+                .zip(any_usize())
+                .map(|(t, i)| (t, State::new(t), i % BYTES_PER_BATCH))
+                .find(|(t, _, i)| t.byte_index.0 + i < BYTES_PER_BATCH - 1)
+                .unwrap();
+            s.increment();
+            assert!(matches!(
+                s.increase(i),
+                ShiftAction::OutputByte(BufferPointer(p_)) if p_ == t.byte_index.0 + i
+            ));
+        }
+    }
+
+    #[test]
+    /// Check the property:
+    ///     For all table indices t, positive integer i such as t.byte_index + i >= 127,
+    ///         if s = State::new(t), and s.increment() was executed, then
+    ///         s.increase(i) = RefreshBatchAndOutputByte(
+    ///             t.increased(i).aes_index,
+    ///             t.increased(i).byte_index).
+    fn prop_state_increase_large() {
+        for _ in 0..REPEATS {
+            let (t, mut s, i) = any_table_index()
+                .zip(any_usize())
+                .map(|(t, i)| (t, State::new(t), i))
+                .find(|(t, _, i)| t.byte_index.0 + i >= BYTES_PER_BATCH - 1)
+                .unwrap();
+            s.increment();
+            assert!(matches!(
+                s.increase(i),
+                ShiftAction::RefreshBatchAndOutputByte(t_, BufferPointer(p_))
+                    if t_ == t.increased(i).aes_index && p_ == t.increased(i).byte_index.0
+            ));
+        }
+    }
+}

concrete-csprng/src/generators/implem/aarch64/block_cipher.rs

@@ -0,0 +1,184 @@
+use crate::generators::aes_ctr::{AesBlockCipher, AesIndex, AesKey, BYTES_PER_BATCH};
+use core::arch::aarch64::{
+    uint8x16_t, vaeseq_u8, vaesmcq_u8, vdupq_n_u32, vdupq_n_u8, veorq_u8, vgetq_lane_u32,
+    vreinterpretq_u32_u8, vreinterpretq_u8_u32,
+};
+use std::arch::is_aarch64_feature_detected;
+use std::mem::transmute;
+
+const RCONS: [u32; 10] = [0x01, 0x02, 0x04, 0x08, 0x10, 0x20, 0x40, 0x80, 0x1B, 0x36];
+const NUM_WORDS_IN_KEY: usize = 4;
+const NUM_ROUNDS: usize = 10;
+const NUM_ROUND_KEYS: usize = NUM_ROUNDS + 1;
+
+/// An aes block cipher implementation which uses `neon` and `aes` instructions.
+#[derive(Clone)]
+pub struct ArmAesBlockCipher {
+    round_keys: [uint8x16_t; NUM_ROUND_KEYS],
+}
+
+impl AesBlockCipher for ArmAesBlockCipher {
+    fn new(key: AesKey) -> ArmAesBlockCipher {
+        let aes_detected = is_aarch64_feature_detected!("aes");
+        let neon_detected = is_aarch64_feature_detected!("neon");
+
+        if !(aes_detected && neon_detected) {
+            panic!(
+                "The ArmAesBlockCipher requires both aes and neon aarch64 CPU features.\n\
+                aes feature available: {}\nneon feature available: {}\n.",
+                aes_detected, neon_detected
+            )
+        }
+
+        let round_keys = unsafe { generate_round_keys(key) };
+        ArmAesBlockCipher { round_keys }
+    }
+
+    fn generate_batch(&mut self, AesIndex(aes_ctr): AesIndex) -> [u8; BYTES_PER_BATCH] {
+        #[target_feature(enable = "aes,neon")]
+        unsafe fn implementation(
+            this: &ArmAesBlockCipher,
+            AesIndex(aes_ctr): AesIndex,
+        ) -> [u8; BYTES_PER_BATCH] {
+            let mut output = [0u8; BYTES_PER_BATCH];
+            // We want 128 bytes of output, the ctr gives 128 bit message (16 bytes)
+            for (i, out) in output.chunks_exact_mut(16).enumerate() {
+                // Safe because we prevent the user from creating the Generator
+                // on non-supported hardware
+                let encrypted = encrypt(aes_ctr + (i as u128), &this.round_keys);
+                out.copy_from_slice(&encrypted.to_ne_bytes());
+            }
+            output
+        }
+        // SAFETY: we checked for aes and neon availability in `Self::new`
+        unsafe { implementation(self, AesIndex(aes_ctr)) }
+    }
+}
+
+/// Does the AES SubWord operation for the Key Expansion step
+///
+/// # SAFETY
+///
+/// You must make sure the CPU's arch is`aarch64` and has
+/// `neon` and `aes` features.
+#[inline(always)]
+unsafe fn sub_word(word: u32) -> u32 {
+    let data = vreinterpretq_u8_u32(vdupq_n_u32(word));
+    let zero_key = vdupq_n_u8(0u8);
+    let temp = vaeseq_u8(data, zero_key);
+    // vaeseq_u8 does SubBytes(ShiftRow(XOR(data, key))
+    // But because we used a zero aes key,the XOR did not alter data
+    // We now have temp = SubBytes(ShiftRow(data))
+
+    // Since in AES ShiftRow operation, the first row is not shifted
+    // We can just get that one to have our SubWord(word) result
+    vgetq_lane_u32::<0>(vreinterpretq_u32_u8(temp))
+}
+
+#[inline(always)]
+fn uint8x16_t_to_u128(input: uint8x16_t) -> u128 {
+    unsafe { transmute(input) }
+}
+
+#[inline(always)]
+fn u128_to_uint8x16_t(input: u128) -> uint8x16_t {
+    unsafe { transmute(input) }
+}
+
+#[target_feature(enable = "aes,neon")]
+unsafe fn generate_round_keys(key: AesKey) -> [uint8x16_t; NUM_ROUND_KEYS] {
+    let mut round_keys: [uint8x16_t; NUM_ROUND_KEYS] = std::mem::zeroed();
+    round_keys[0] = u128_to_uint8x16_t(key.0);
+
+    let words = std::slice::from_raw_parts_mut(
+        round_keys.as_mut_ptr() as *mut u32,
+        NUM_ROUND_KEYS * NUM_WORDS_IN_KEY,
+    );
+
+    debug_assert_eq!(words.len(), 44);
+
+    // Skip the words of the first key, its already done
+    for i in NUM_WORDS_IN_KEY..words.len() {
+        if (i % NUM_WORDS_IN_KEY) == 0 {
+            words[i] = words[i - NUM_WORDS_IN_KEY]
+                ^ sub_word(words[i - 1]).rotate_right(8)
+                ^ RCONS[(i / NUM_WORDS_IN_KEY) - 1];
+        } else {
+            words[i] = words[i - NUM_WORDS_IN_KEY] ^ words[i - 1];
+        }
+        // Note: there is also a special thing to do when
+        // i mod SElf::NUM_WORDS_IN_KEY == 4 but it cannot happen on 128 bits keys
+    }
+
+    round_keys
+}
+
+/// Encrypts a 128-bit message
+///
+/// # SAFETY
+///
+/// You must make sure the CPU's arch is`aarch64` and has
+/// `neon` and `aes` features.
+#[inline(always)]
+unsafe fn encrypt(message: u128, keys: &[uint8x16_t; NUM_ROUND_KEYS]) -> u128 {
+    // Notes:
+    // According the [ARM Manual](https://developer.arm.com/documentation/ddi0487/gb/):
+    // `vaeseq_u8` is the following AES operations:
+    //      1. AddRoundKey (XOR)
+    //      2. ShiftRows
+    //      3. SubBytes
+    // `vaesmcq_u8` is MixColumns
+    let mut data: uint8x16_t = u128_to_uint8x16_t(message);
+
+    for &key in keys.iter().take(NUM_ROUNDS - 1) {
+        data = vaesmcq_u8(vaeseq_u8(data, key));
+    }
+
+    data = vaeseq_u8(data, keys[NUM_ROUNDS - 1]);
+    data = veorq_u8(data, keys[NUM_ROUND_KEYS - 1]);
+
+    uint8x16_t_to_u128(data)
+}
+
+#[cfg(test)]
+mod test {
+    use super::*;
+
+    // Test vector for aes128, from the FIPS publication 197
+    const CIPHER_KEY: u128 = u128::from_be(0x000102030405060708090a0b0c0d0e0f);
+    const KEY_SCHEDULE: [u128; 11] = [
+        u128::from_be(0x000102030405060708090a0b0c0d0e0f),
+        u128::from_be(0xd6aa74fdd2af72fadaa678f1d6ab76fe),
+        u128::from_be(0xb692cf0b643dbdf1be9bc5006830b3fe),
+        u128::from_be(0xb6ff744ed2c2c9bf6c590cbf0469bf41),
+        u128::from_be(0x47f7f7bc95353e03f96c32bcfd058dfd),
+        u128::from_be(0x3caaa3e8a99f9deb50f3af57adf622aa),
+        u128::from_be(0x5e390f7df7a69296a7553dc10aa31f6b),
+        u128::from_be(0x14f9701ae35fe28c440adf4d4ea9c026),
+        u128::from_be(0x47438735a41c65b9e016baf4aebf7ad2),
+        u128::from_be(0x549932d1f08557681093ed9cbe2c974e),
+        u128::from_be(0x13111d7fe3944a17f307a78b4d2b30c5),
+    ];
+    const PLAINTEXT: u128 = u128::from_be(0x00112233445566778899aabbccddeeff);
+    const CIPHERTEXT: u128 = u128::from_be(0x69c4e0d86a7b0430d8cdb78070b4c55a);
+
+    #[test]
+    fn test_generate_key_schedule() {
+        // Checks that the round keys are correctly generated from the sample key from FIPS
+        let key = AesKey(CIPHER_KEY);
+        let keys = unsafe { generate_round_keys(key) };
+        for (expected, actual) in KEY_SCHEDULE.iter().zip(keys.iter()) {
+            assert_eq!(*expected, uint8x16_t_to_u128(*actual));
+        }
+    }
+
+    #[test]
+    fn test_encrypt_message() {
+        // Checks that encrypting many plaintext at the same time gives the correct output.
+        let message = PLAINTEXT;
+        let key = AesKey(CIPHER_KEY);
+        let keys = unsafe { generate_round_keys(key) };
+        let ciphertext = unsafe { encrypt(message, &keys) };
+        assert_eq!(CIPHERTEXT, ciphertext);
+    }
+}

concrete-csprng/src/generators/implem/aarch64/generator.rs

@@ -0,0 +1,110 @@
+use crate::generators::aes_ctr::{AesCtrGenerator, AesKey, ChildrenIterator};
+use crate::generators::implem::aarch64::block_cipher::ArmAesBlockCipher;
+use crate::generators::{ByteCount, BytesPerChild, ChildrenCount, ForkError, RandomGenerator};
+use crate::seeders::Seed;
+
+/// A random number generator using the `aesni` instructions.
+pub struct NeonAesRandomGenerator(pub(super) AesCtrGenerator<ArmAesBlockCipher>);
+
+/// The children iterator used by [`NeonAesRandomGenerator`].
+///
+/// Outputs children generators one by one.
+pub struct ArmAesChildrenIterator(ChildrenIterator<ArmAesBlockCipher>);
+
+impl Iterator for ArmAesChildrenIterator {
+    type Item = NeonAesRandomGenerator;
+
+    fn next(&mut self) -> Option<Self::Item> {
+        self.0.next().map(NeonAesRandomGenerator)
+    }
+}
+
+impl RandomGenerator for NeonAesRandomGenerator {
+    type ChildrenIter = ArmAesChildrenIterator;
+    fn new(seed: Seed) -> Self {
+        NeonAesRandomGenerator(AesCtrGenerator::new(AesKey(seed.0), None, None))
+    }
+    fn remaining_bytes(&self) -> ByteCount {
+        self.0.remaining_bytes()
+    }
+    fn try_fork(
+        &mut self,
+        n_children: ChildrenCount,
+        n_bytes: BytesPerChild,
+    ) -> Result<Self::ChildrenIter, ForkError> {
+        self.0
+            .try_fork(n_children, n_bytes)
+            .map(ArmAesChildrenIterator)
+    }
+}
+
+impl Iterator for NeonAesRandomGenerator {
+    type Item = u8;
+
+    fn next(&mut self) -> Option<Self::Item> {
+        self.0.next()
+    }
+}
+
+#[cfg(test)]
+mod test {
+    use crate::generators::aes_ctr::aes_ctr_generic_test;
+    use crate::generators::implem::aarch64::block_cipher::ArmAesBlockCipher;
+    use crate::generators::{generator_generic_test, NeonAesRandomGenerator};
+
+    #[test]
+    fn prop_fork_first_state_table_index() {
+        aes_ctr_generic_test::prop_fork_first_state_table_index::<ArmAesBlockCipher>();
+    }
+
+    #[test]
+    fn prop_fork_last_bound_table_index() {
+        aes_ctr_generic_test::prop_fork_last_bound_table_index::<ArmAesBlockCipher>();
+    }
+
+    #[test]
+    fn prop_fork_parent_bound_table_index() {
+        aes_ctr_generic_test::prop_fork_parent_bound_table_index::<ArmAesBlockCipher>();
+    }
+
+    #[test]
+    fn prop_fork_parent_state_table_index() {
+        aes_ctr_generic_test::prop_fork_parent_state_table_index::<ArmAesBlockCipher>();
+    }
+
+    #[test]
+    fn prop_fork() {
+        aes_ctr_generic_test::prop_fork::<ArmAesBlockCipher>();
+    }
+
+    #[test]
+    fn prop_fork_children_remaining_bytes() {
+        aes_ctr_generic_test::prop_fork_children_remaining_bytes::<ArmAesBlockCipher>();
+    }
+
+    #[test]
+    fn prop_fork_parent_remaining_bytes() {
+        aes_ctr_generic_test::prop_fork_parent_remaining_bytes::<ArmAesBlockCipher>();
+    }
+
+    #[test]
+    fn test_roughly_uniform() {
+        generator_generic_test::test_roughly_uniform::<NeonAesRandomGenerator>();
+    }
+
+    #[test]
+    fn test_generator_determinism() {
+        generator_generic_test::test_generator_determinism::<NeonAesRandomGenerator>();
+    }
+
+    #[test]
+    fn test_fork() {
+        generator_generic_test::test_fork_children::<NeonAesRandomGenerator>();
+    }
+
+    #[test]
+    #[should_panic(expected = "expected test panic")]
+    fn test_bounded_panic() {
+        generator_generic_test::test_bounded_none_should_panic::<NeonAesRandomGenerator>();
+    }
+}

concrete-csprng/src/generators/implem/aarch64/mod.rs

@@ -0,0 +1,16 @@
+//! A module implementing a random number generator, using the aarch64 `neon` and `aes`
+//! instructions.
+//!
+//! This module implements a cryptographically secure pseudorandom number generator
+//! (CS-PRNG), using a fast block cipher. The implementation is based on the
+//! [intel aesni white paper 323641-001 revision 3.0](https://www.intel.com/content/dam/doc/white-paper/advanced-encryption-standard-new-instructions-set-paper.pdf).
+
+mod block_cipher;
+
+mod generator;
+pub use generator::*;
+
+#[cfg(feature = "parallel")]
+mod parallel;
+#[cfg(feature = "parallel")]
+pub use parallel::*;

concrete-csprng/src/generators/implem/aarch64/parallel.rs

@@ -0,0 +1,95 @@
+use super::*;
+use crate::generators::aes_ctr::{AesCtrGenerator, ParallelChildrenIterator};
+use crate::generators::implem::aarch64::block_cipher::ArmAesBlockCipher;
+use crate::generators::{BytesPerChild, ChildrenCount, ForkError, ParallelRandomGenerator};
+use rayon::iter::plumbing::{Consumer, ProducerCallback, UnindexedConsumer};
+use rayon::prelude::*;
+
+/// The parallel children iterator used by [`NeonAesRandomGenerator`].
+///
+/// Outputs the children generators one by one.
+#[allow(clippy::type_complexity)]
+pub struct ParallelArmAesChildrenIterator(
+    rayon::iter::Map<
+        ParallelChildrenIterator<ArmAesBlockCipher>,
+        fn(AesCtrGenerator<ArmAesBlockCipher>) -> NeonAesRandomGenerator,
+    >,
+);
+
+impl ParallelIterator for ParallelArmAesChildrenIterator {
+    type Item = NeonAesRandomGenerator;
+    fn drive_unindexed<C>(self, consumer: C) -> C::Result
+    where
+        C: UnindexedConsumer<Self::Item>,
+    {
+        self.0.drive_unindexed(consumer)
+    }
+}
+
+impl IndexedParallelIterator for ParallelArmAesChildrenIterator {
+    fn len(&self) -> usize {
+        self.0.len()
+    }
+    fn drive<C: Consumer<Self::Item>>(self, consumer: C) -> C::Result {
+        self.0.drive(consumer)
+    }
+    fn with_producer<CB: ProducerCallback<Self::Item>>(self, callback: CB) -> CB::Output {
+        self.0.with_producer(callback)
+    }
+}
+
+impl ParallelRandomGenerator for NeonAesRandomGenerator {
+    type ParChildrenIter = ParallelArmAesChildrenIterator;
+
+    fn par_try_fork(
+        &mut self,
+        n_children: ChildrenCount,
+        n_bytes: BytesPerChild,
+    ) -> Result<Self::ParChildrenIter, ForkError> {
+        self.0
+            .par_try_fork(n_children, n_bytes)
+            .map(|iterator| ParallelArmAesChildrenIterator(iterator.map(NeonAesRandomGenerator)))
+    }
+}
+
+#[cfg(test)]
+
+mod test {
+    use crate::generators::aes_ctr::aes_ctr_parallel_generic_tests;
+    use crate::generators::implem::aarch64::block_cipher::ArmAesBlockCipher;
+
+    #[test]
+    fn prop_fork_first_state_table_index() {
+        aes_ctr_parallel_generic_tests::prop_fork_first_state_table_index::<ArmAesBlockCipher>();
+    }
+
+    #[test]
+    fn prop_fork_last_bound_table_index() {
+        aes_ctr_parallel_generic_tests::prop_fork_last_bound_table_index::<ArmAesBlockCipher>();
+    }
+
+    #[test]
+    fn prop_fork_parent_bound_table_index() {
+        aes_ctr_parallel_generic_tests::prop_fork_parent_bound_table_index::<ArmAesBlockCipher>();
+    }
+
+    #[test]
+    fn prop_fork_parent_state_table_index() {
+        aes_ctr_parallel_generic_tests::prop_fork_parent_state_table_index::<ArmAesBlockCipher>();
+    }
+
+    #[test]
+    fn prop_fork_ttt() {
+        aes_ctr_parallel_generic_tests::prop_fork::<ArmAesBlockCipher>();
+    }
+
+    #[test]
+    fn prop_fork_children_remaining_bytes() {
+        aes_ctr_parallel_generic_tests::prop_fork_children_remaining_bytes::<ArmAesBlockCipher>();
+    }
+
+    #[test]
+    fn prop_fork_parent_remaining_bytes() {
+        aes_ctr_parallel_generic_tests::prop_fork_parent_remaining_bytes::<ArmAesBlockCipher>();
+    }
+}

concrete-csprng/src/generators/implem/aesni/block_cipher.rs

@@ -0,0 +1,231 @@
+use crate::generators::aes_ctr::{AesBlockCipher, AesIndex, AesKey, BYTES_PER_BATCH};
+use std::arch::x86_64::{
+    __m128i, _mm_aesenc_si128, _mm_aesenclast_si128, _mm_aeskeygenassist_si128, _mm_shuffle_epi32,
+    _mm_slli_si128, _mm_store_si128, _mm_xor_si128,
+};
+use std::mem::transmute;
+
+/// An aes block cipher implementation which uses `aesni` instructions.
+#[derive(Clone)]
+pub struct AesniBlockCipher {
+    // The set of round keys used for the aes encryption
+    round_keys: [__m128i; 11],
+}
+
+impl AesBlockCipher for AesniBlockCipher {
+    fn new(key: AesKey) -> AesniBlockCipher {
+        let aes_detected = is_x86_feature_detected!("aes");
+        let sse2_detected = is_x86_feature_detected!("sse2");
+
+        if !(aes_detected && sse2_detected) {
+            panic!(
+                "The AesniBlockCipher requires both aes and sse2 x86 CPU features.\n\
+                aes feature available: {}\nsse2 feature available: {}\n.",
+                aes_detected, sse2_detected
+            )
+        }
+
+        // SAFETY: we checked for aes and sse2 availability
+        let round_keys = unsafe { generate_round_keys(key) };
+        AesniBlockCipher { round_keys }
+    }
+
+    fn generate_batch(&mut self, AesIndex(aes_ctr): AesIndex) -> [u8; BYTES_PER_BATCH] {
+        #[target_feature(enable = "sse2,aes")]
+        unsafe fn implementation(
+            this: &AesniBlockCipher,
+            AesIndex(aes_ctr): AesIndex,
+        ) -> [u8; BYTES_PER_BATCH] {
+            si128arr_to_u8arr(aes_encrypt_many(
+                u128_to_si128(aes_ctr),
+                u128_to_si128(aes_ctr + 1),
+                u128_to_si128(aes_ctr + 2),
+                u128_to_si128(aes_ctr + 3),
+                u128_to_si128(aes_ctr + 4),
+                u128_to_si128(aes_ctr + 5),
+                u128_to_si128(aes_ctr + 6),
+                u128_to_si128(aes_ctr + 7),
+                &this.round_keys,
+            ))
+        }
+        // SAFETY: we checked for aes and sse2 availability in `Self::new`
+        unsafe { implementation(self, AesIndex(aes_ctr)) }
+    }
+}
+
+#[target_feature(enable = "sse2,aes")]
+unsafe fn generate_round_keys(key: AesKey) -> [__m128i; 11] {
+    let key = u128_to_si128(key.0);
+    let mut keys: [__m128i; 11] = [u128_to_si128(0); 11];
+    aes_128_key_expansion(key, &mut keys);
+    keys
+}
+
+// Uses aes to encrypt many values at once. This allows a substantial speedup (around 30%)
+// compared to the naive approach.
+#[allow(clippy::too_many_arguments)]
+#[inline(always)]
+fn aes_encrypt_many(
+    message_1: __m128i,
+    message_2: __m128i,
+    message_3: __m128i,
+    message_4: __m128i,
+    message_5: __m128i,
+    message_6: __m128i,
+    message_7: __m128i,
+    message_8: __m128i,
+    keys: &[__m128i; 11],
+) -> [__m128i; 8] {
+    unsafe {
+        let mut tmp_1 = _mm_xor_si128(message_1, keys[0]);
+        let mut tmp_2 = _mm_xor_si128(message_2, keys[0]);
+        let mut tmp_3 = _mm_xor_si128(message_3, keys[0]);
+        let mut tmp_4 = _mm_xor_si128(message_4, keys[0]);
+        let mut tmp_5 = _mm_xor_si128(message_5, keys[0]);
+        let mut tmp_6 = _mm_xor_si128(message_6, keys[0]);
+        let mut tmp_7 = _mm_xor_si128(message_7, keys[0]);
+        let mut tmp_8 = _mm_xor_si128(message_8, keys[0]);
+
+        for key in keys.iter().take(10).skip(1) {
+            tmp_1 = _mm_aesenc_si128(tmp_1, *key);
+            tmp_2 = _mm_aesenc_si128(tmp_2, *key);
+            tmp_3 = _mm_aesenc_si128(tmp_3, *key);
+            tmp_4 = _mm_aesenc_si128(tmp_4, *key);
+            tmp_5 = _mm_aesenc_si128(tmp_5, *key);
+            tmp_6 = _mm_aesenc_si128(tmp_6, *key);
+            tmp_7 = _mm_aesenc_si128(tmp_7, *key);
+            tmp_8 = _mm_aesenc_si128(tmp_8, *key);
+        }
+
+        tmp_1 = _mm_aesenclast_si128(tmp_1, keys[10]);
+        tmp_2 = _mm_aesenclast_si128(tmp_2, keys[10]);
+        tmp_3 = _mm_aesenclast_si128(tmp_3, keys[10]);
+        tmp_4 = _mm_aesenclast_si128(tmp_4, keys[10]);
+        tmp_5 = _mm_aesenclast_si128(tmp_5, keys[10]);
+        tmp_6 = _mm_aesenclast_si128(tmp_6, keys[10]);
+        tmp_7 = _mm_aesenclast_si128(tmp_7, keys[10]);
+        tmp_8 = _mm_aesenclast_si128(tmp_8, keys[10]);
+
+        [tmp_1, tmp_2, tmp_3, tmp_4, tmp_5, tmp_6, tmp_7, tmp_8]
+    }
+}
+
+fn aes_128_assist(temp1: __m128i, temp2: __m128i) -> __m128i {
+    let mut temp3: __m128i;
+    let mut temp2 = temp2;
+    let mut temp1 = temp1;
+    unsafe {
+        temp2 = _mm_shuffle_epi32(temp2, 0xff);
+        temp3 = _mm_slli_si128(temp1, 0x4);
+        temp1 = _mm_xor_si128(temp1, temp3);
+        temp3 = _mm_slli_si128(temp3, 0x4);
+        temp1 = _mm_xor_si128(temp1, temp3);
+        temp3 = _mm_slli_si128(temp3, 0x4);
+        temp1 = _mm_xor_si128(temp1, temp3);
+        temp1 = _mm_xor_si128(temp1, temp2);
+    }
+    temp1
+}
+
+#[inline(always)]
+fn aes_128_key_expansion(key: __m128i, keys: &mut [__m128i; 11]) {
+    let (mut temp1, mut temp2): (__m128i, __m128i);
+    temp1 = key;
+    unsafe {
+        _mm_store_si128(keys.as_mut_ptr(), temp1);
+        temp2 = _mm_aeskeygenassist_si128(temp1, 0x01);
+        temp1 = aes_128_assist(temp1, temp2);
+        _mm_store_si128(keys.as_mut_ptr().offset(1), temp1);
+        temp2 = _mm_aeskeygenassist_si128(temp1, 0x02);
+        temp1 = aes_128_assist(temp1, temp2);
+        _mm_store_si128(keys.as_mut_ptr().offset(2), temp1);
+        temp2 = _mm_aeskeygenassist_si128(temp1, 0x04);
+        temp1 = aes_128_assist(temp1, temp2);
+        _mm_store_si128(keys.as_mut_ptr().offset(3), temp1);
+        temp2 = _mm_aeskeygenassist_si128(temp1, 0x08);
+        temp1 = aes_128_assist(temp1, temp2);
+        _mm_store_si128(keys.as_mut_ptr().offset(4), temp1);
+        temp2 = _mm_aeskeygenassist_si128(temp1, 0x10);
+        temp1 = aes_128_assist(temp1, temp2);
+        _mm_store_si128(keys.as_mut_ptr().offset(5), temp1);
+        temp2 = _mm_aeskeygenassist_si128(temp1, 0x20);
+        temp1 = aes_128_assist(temp1, temp2);
+        _mm_store_si128(keys.as_mut_ptr().offset(6), temp1);
+        temp2 = _mm_aeskeygenassist_si128(temp1, 0x40);
+        temp1 = aes_128_assist(temp1, temp2);
+        _mm_store_si128(keys.as_mut_ptr().offset(7), temp1);
+        temp2 = _mm_aeskeygenassist_si128(temp1, 0x80);
+        temp1 = aes_128_assist(temp1, temp2);
+        _mm_store_si128(keys.as_mut_ptr().offset(8), temp1);
+        temp2 = _mm_aeskeygenassist_si128(temp1, 0x1b);
+        temp1 = aes_128_assist(temp1, temp2);
+        _mm_store_si128(keys.as_mut_ptr().offset(9), temp1);
+        temp2 = _mm_aeskeygenassist_si128(temp1, 0x36);
+        temp1 = aes_128_assist(temp1, temp2);
+        _mm_store_si128(keys.as_mut_ptr().offset(10), temp1);
+    }
+}
+
+#[inline(always)]
+fn u128_to_si128(input: u128) -> __m128i {
+    unsafe { transmute(input) }
+}
+
+#[allow(unused)] // to please clippy when tests are not activated
+fn si128_to_u128(input: __m128i) -> u128 {
+    unsafe { transmute(input) }
+}
+
+#[inline(always)]
+fn si128arr_to_u8arr(input: [__m128i; 8]) -> [u8; BYTES_PER_BATCH] {
+    unsafe { transmute(input) }
+}
+
+#[cfg(test)]
+mod test {
+    use super::*;
+
+    // Test vector for aes128, from the FIPS publication 197
+    const CIPHER_KEY: u128 = u128::from_be(0x000102030405060708090a0b0c0d0e0f);
+    const KEY_SCHEDULE: [u128; 11] = [
+        u128::from_be(0x000102030405060708090a0b0c0d0e0f),
+        u128::from_be(0xd6aa74fdd2af72fadaa678f1d6ab76fe),
+        u128::from_be(0xb692cf0b643dbdf1be9bc5006830b3fe),
+        u128::from_be(0xb6ff744ed2c2c9bf6c590cbf0469bf41),
+        u128::from_be(0x47f7f7bc95353e03f96c32bcfd058dfd),
+        u128::from_be(0x3caaa3e8a99f9deb50f3af57adf622aa),
+        u128::from_be(0x5e390f7df7a69296a7553dc10aa31f6b),
+        u128::from_be(0x14f9701ae35fe28c440adf4d4ea9c026),
+        u128::from_be(0x47438735a41c65b9e016baf4aebf7ad2),
+        u128::from_be(0x549932d1f08557681093ed9cbe2c974e),
+        u128::from_be(0x13111d7fe3944a17f307a78b4d2b30c5),
+    ];
+    const PLAINTEXT: u128 = u128::from_be(0x00112233445566778899aabbccddeeff);
+    const CIPHERTEXT: u128 = u128::from_be(0x69c4e0d86a7b0430d8cdb78070b4c55a);
+
+    #[test]
+    fn test_generate_key_schedule() {
+        // Checks that the round keys are correctly generated from the sample key from FIPS
+        let key = u128_to_si128(CIPHER_KEY);
+        let mut keys: [__m128i; 11] = [u128_to_si128(0); 11];
+        aes_128_key_expansion(key, &mut keys);
+        for (expected, actual) in KEY_SCHEDULE.iter().zip(keys.iter()) {
+            assert_eq!(*expected, si128_to_u128(*actual));
+        }
+    }
+
+    #[test]
+    fn test_encrypt_many_messages() {
+        // Checks that encrypting many plaintext at the same time gives the correct output.
+        let message = u128_to_si128(PLAINTEXT);
+        let key = u128_to_si128(CIPHER_KEY);
+        let mut keys: [__m128i; 11] = [u128_to_si128(0); 11];
+        aes_128_key_expansion(key, &mut keys);
+        let ciphertexts = aes_encrypt_many(
+            message, message, message, message, message, message, message, message, &keys,
+        );
+        for ct in &ciphertexts {
+            assert_eq!(CIPHERTEXT, si128_to_u128(*ct));
+        }
+    }
+}

concrete-csprng/src/generators/implem/aesni/generator.rs

@@ -0,0 +1,110 @@
+use crate::generators::aes_ctr::{AesCtrGenerator, AesKey, ChildrenIterator};
+use crate::generators::implem::aesni::block_cipher::AesniBlockCipher;
+use crate::generators::{ByteCount, BytesPerChild, ChildrenCount, ForkError, RandomGenerator};
+use crate::seeders::Seed;
+
+/// A random number generator using the `aesni` instructions.
+pub struct AesniRandomGenerator(pub(super) AesCtrGenerator<AesniBlockCipher>);
+
+/// The children iterator used by [`AesniRandomGenerator`].
+///
+/// Outputs children generators one by one.
+pub struct AesniChildrenIterator(ChildrenIterator<AesniBlockCipher>);
+
+impl Iterator for AesniChildrenIterator {
+    type Item = AesniRandomGenerator;
+
+    fn next(&mut self) -> Option<Self::Item> {
+        self.0.next().map(AesniRandomGenerator)
+    }
+}
+
+impl RandomGenerator for AesniRandomGenerator {
+    type ChildrenIter = AesniChildrenIterator;
+    fn new(seed: Seed) -> Self {
+        AesniRandomGenerator(AesCtrGenerator::new(AesKey(seed.0), None, None))
+    }
+    fn remaining_bytes(&self) -> ByteCount {
+        self.0.remaining_bytes()
+    }
+    fn try_fork(
+        &mut self,
+        n_children: ChildrenCount,
+        n_bytes: BytesPerChild,
+    ) -> Result<Self::ChildrenIter, ForkError> {
+        self.0
+            .try_fork(n_children, n_bytes)
+            .map(AesniChildrenIterator)
+    }
+}
+
+impl Iterator for AesniRandomGenerator {
+    type Item = u8;
+
+    fn next(&mut self) -> Option<Self::Item> {
+        self.0.next()
+    }
+}
+
+#[cfg(test)]
+mod test {
+    use crate::generators::aes_ctr::aes_ctr_generic_test;
+    use crate::generators::implem::aesni::block_cipher::AesniBlockCipher;
+    use crate::generators::{generator_generic_test, AesniRandomGenerator};
+
+    #[test]
+    fn prop_fork_first_state_table_index() {
+        aes_ctr_generic_test::prop_fork_first_state_table_index::<AesniBlockCipher>();
+    }
+
+    #[test]
+    fn prop_fork_last_bound_table_index() {
+        aes_ctr_generic_test::prop_fork_last_bound_table_index::<AesniBlockCipher>();
+    }
+
+    #[test]
+    fn prop_fork_parent_bound_table_index() {
+        aes_ctr_generic_test::prop_fork_parent_bound_table_index::<AesniBlockCipher>();
+    }
+
+    #[test]
+    fn prop_fork_parent_state_table_index() {
+        aes_ctr_generic_test::prop_fork_parent_state_table_index::<AesniBlockCipher>();
+    }
+
+    #[test]
+    fn prop_fork() {
+        aes_ctr_generic_test::prop_fork::<AesniBlockCipher>();
+    }
+
+    #[test]
+    fn prop_fork_children_remaining_bytes() {
+        aes_ctr_generic_test::prop_fork_children_remaining_bytes::<AesniBlockCipher>();
+    }
+
+    #[test]
+    fn prop_fork_parent_remaining_bytes() {
+        aes_ctr_generic_test::prop_fork_parent_remaining_bytes::<AesniBlockCipher>();
+    }
+
+    #[test]
+    fn test_roughly_uniform() {
+        generator_generic_test::test_roughly_uniform::<AesniRandomGenerator>();
+    }
+
+    #[test]
+    fn test_generator_determinism() {
+        generator_generic_test::test_generator_determinism::<AesniRandomGenerator>();
+    }
+
+    #[test]
+    fn test_fork() {
+        generator_generic_test::test_fork_children::<AesniRandomGenerator>();
+    }
+
+    #[test]
+    #[should_panic(expected = "expected test panic")]
+    fn test_bounded_panic() {
+        generator_generic_test::test_bounded_none_should_panic::<AesniRandomGenerator>();
+    }
+}

concrete-csprng/src/generators/implem/aesni/mod.rs

@@ -0,0 +1,15 @@
+//! A module implementing a random number generator, using the x86_64 `aesni` instructions.
+//!
+//! This module implements a cryptographically secure pseudorandom number generator
+//! (CS-PRNG), using a fast block cipher. The implementation is based on the
+//! [intel aesni white paper 323641-001 revision 3.0](https://www.intel.com/content/dam/doc/white-paper/advanced-encryption-standard-new-instructions-set-paper.pdf).
+
+mod block_cipher;
+
+mod generator;
+pub use generator::*;
+
+#[cfg(feature = "parallel")]
+mod parallel;
+#[cfg(feature = "parallel")]
+pub use parallel::*;

concrete-csprng/src/generators/implem/aesni/parallel.rs

@@ -0,0 +1,95 @@
+use super::*;
+use crate::generators::aes_ctr::{AesCtrGenerator, ParallelChildrenIterator};
+use crate::generators::implem::aesni::block_cipher::AesniBlockCipher;
+use crate::generators::{BytesPerChild, ChildrenCount, ForkError, ParallelRandomGenerator};
+use rayon::iter::plumbing::{Consumer, ProducerCallback, UnindexedConsumer};
+use rayon::prelude::*;
+
+/// The parallel children iterator used by [`AesniRandomGenerator`].
+///
+/// Outputs the children generators one by one.
+#[allow(clippy::type_complexity)]
+pub struct ParallelAesniChildrenIterator(
+    rayon::iter::Map<
+        ParallelChildrenIterator<AesniBlockCipher>,
+        fn(AesCtrGenerator<AesniBlockCipher>) -> AesniRandomGenerator,
+    >,
+);
+
+impl ParallelIterator for ParallelAesniChildrenIterator {
+    type Item = AesniRandomGenerator;
+    fn drive_unindexed<C>(self, consumer: C) -> C::Result
+    where
+        C: UnindexedConsumer<Self::Item>,
+    {
+        self.0.drive_unindexed(consumer)
+    }
+}
+
+impl IndexedParallelIterator for ParallelAesniChildrenIterator {
+    fn len(&self) -> usize {
+        self.0.len()
+    }
+    fn drive<C: Consumer<Self::Item>>(self, consumer: C) -> C::Result {
+        self.0.drive(consumer)
+    }
+    fn with_producer<CB: ProducerCallback<Self::Item>>(self, callback: CB) -> CB::Output {
+        self.0.with_producer(callback)
+    }
+}
+
+impl ParallelRandomGenerator for AesniRandomGenerator {
+    type ParChildrenIter = ParallelAesniChildrenIterator;
+
+    fn par_try_fork(
+        &mut self,
+        n_children: ChildrenCount,
+        n_bytes: BytesPerChild,
+    ) -> Result<Self::ParChildrenIter, ForkError> {
+        self.0
+            .par_try_fork(n_children, n_bytes)
+            .map(|iterator| ParallelAesniChildrenIterator(iterator.map(AesniRandomGenerator)))
+    }
+}
+
+#[cfg(test)]
+
+mod test {
+    use crate::generators::aes_ctr::aes_ctr_parallel_generic_tests;
+    use crate::generators::implem::aesni::block_cipher::AesniBlockCipher;
+
+    #[test]
+    fn prop_fork_first_state_table_index() {
+        aes_ctr_parallel_generic_tests::prop_fork_first_state_table_index::<AesniBlockCipher>();
+    }
+
+    #[test]
+    fn prop_fork_last_bound_table_index() {
+        aes_ctr_parallel_generic_tests::prop_fork_last_bound_table_index::<AesniBlockCipher>();
+    }
+
+    #[test]
+    fn prop_fork_parent_bound_table_index() {
+        aes_ctr_parallel_generic_tests::prop_fork_parent_bound_table_index::<AesniBlockCipher>();
+    }
+
+    #[test]
+    fn prop_fork_parent_state_table_index() {
+        aes_ctr_parallel_generic_tests::prop_fork_parent_state_table_index::<AesniBlockCipher>();
+    }
+
+    #[test]
+    fn prop_fork_ttt() {
+        aes_ctr_parallel_generic_tests::prop_fork::<AesniBlockCipher>();
+    }
+
+    #[test]
+    fn prop_fork_children_remaining_bytes() {
+        aes_ctr_parallel_generic_tests::prop_fork_children_remaining_bytes::<AesniBlockCipher>();
+    }
+
+    #[test]
+    fn prop_fork_parent_remaining_bytes() {
+        aes_ctr_parallel_generic_tests::prop_fork_parent_remaining_bytes::<AesniBlockCipher>();
+    }
+}

concrete-csprng/src/generators/implem/mod.rs

@@ -0,0 +1,14 @@
+#[cfg(feature = "generator_x86_64_aesni")]
+mod aesni;
+#[cfg(feature = "generator_x86_64_aesni")]
+pub use aesni::*;
+
+#[cfg(feature = "generator_aarch64_aes")]
+mod aarch64;
+#[cfg(feature = "generator_aarch64_aes")]
+pub use aarch64::*;
+
+#[cfg(feature = "generator_fallback")]
+mod soft;
+#[cfg(feature = "generator_fallback")]
+pub use soft::*;

concrete-csprng/src/generators/implem/soft/block_cipher.rs

@@ -0,0 +1,114 @@
+use crate::generators::aes_ctr::{
+    AesBlockCipher, AesIndex, AesKey, AES_CALLS_PER_BATCH, BYTES_PER_AES_CALL, BYTES_PER_BATCH,
+};
+use aes::cipher::generic_array::GenericArray;
+use aes::cipher::{BlockEncrypt, KeyInit};
+use aes::Aes128;
+
+#[derive(Clone)]
+pub struct SoftwareBlockCipher {
+    // Aes structure
+    aes: Aes128,
+}
+
+impl AesBlockCipher for SoftwareBlockCipher {
+    fn new(key: AesKey) -> SoftwareBlockCipher {
+        let key: [u8; BYTES_PER_AES_CALL] = key.0.to_ne_bytes();
+        let key = GenericArray::clone_from_slice(&key[..]);
+        let aes = Aes128::new(&key);
+        SoftwareBlockCipher { aes }
+    }
+
+    fn generate_batch(&mut self, AesIndex(aes_ctr): AesIndex) -> [u8; BYTES_PER_BATCH] {
+        aes_encrypt_many(
+            aes_ctr,
+            aes_ctr + 1,
+            aes_ctr + 2,
+            aes_ctr + 3,
+            aes_ctr + 4,
+            aes_ctr + 5,
+            aes_ctr + 6,
+            aes_ctr + 7,
+            &self.aes,
+        )
+    }
+}
+
+// Uses aes to encrypt many values at once. This allows a substantial speedup (around 30%)
+// compared to the naive approach.
+#[allow(clippy::too_many_arguments)]
+fn aes_encrypt_many(
+    message_1: u128,
+    message_2: u128,
+    message_3: u128,
+    message_4: u128,
+    message_5: u128,
+    message_6: u128,
+    message_7: u128,
+    message_8: u128,
+    cipher: &Aes128,
+) -> [u8; BYTES_PER_BATCH] {
+    let mut b1 = GenericArray::clone_from_slice(&message_1.to_ne_bytes()[..]);
+    let mut b2 = GenericArray::clone_from_slice(&message_2.to_ne_bytes()[..]);
+    let mut b3 = GenericArray::clone_from_slice(&message_3.to_ne_bytes()[..]);
+    let mut b4 = GenericArray::clone_from_slice(&message_4.to_ne_bytes()[..]);
+    let mut b5 = GenericArray::clone_from_slice(&message_5.to_ne_bytes()[..]);
+    let mut b6 = GenericArray::clone_from_slice(&message_6.to_ne_bytes()[..]);
+    let mut b7 = GenericArray::clone_from_slice(&message_7.to_ne_bytes()[..]);
+    let mut b8 = GenericArray::clone_from_slice(&message_8.to_ne_bytes()[..]);
+
+    cipher.encrypt_block(&mut b1);
+    cipher.encrypt_block(&mut b2);
+    cipher.encrypt_block(&mut b3);
+    cipher.encrypt_block(&mut b4);
+    cipher.encrypt_block(&mut b5);
+    cipher.encrypt_block(&mut b6);
+    cipher.encrypt_block(&mut b7);
+    cipher.encrypt_block(&mut b8);
+
+    let output_array: [[u8; BYTES_PER_AES_CALL]; AES_CALLS_PER_BATCH] = [
+        b1.into(),
+        b2.into(),
+        b3.into(),
+        b4.into(),
+        b5.into(),
+        b6.into(),
+        b7.into(),
+        b8.into(),
+    ];
+
+    unsafe { *{ output_array.as_ptr() as *const [u8; BYTES_PER_BATCH] } }
+}
+
+#[cfg(test)]
+mod test {
+    use super::*;
+    use std::convert::TryInto;
+
+    // Test vector for aes128, from the FIPS publication 197
+    const CIPHER_KEY: u128 = u128::from_be(0x000102030405060708090a0b0c0d0e0f);
+    const PLAINTEXT: u128 = u128::from_be(0x00112233445566778899aabbccddeeff);
+    const CIPHERTEXT: u128 = u128::from_be(0x69c4e0d86a7b0430d8cdb78070b4c55a);
+
+    #[test]
+    fn test_encrypt_many_messages() {
+        // Checks that encrypting many plaintext at the same time gives the correct output.
+        let key: [u8; BYTES_PER_AES_CALL] = CIPHER_KEY.to_ne_bytes();
+        let aes = Aes128::new(&GenericArray::from(key));
+        let ciphertexts = aes_encrypt_many(
+            PLAINTEXT, PLAINTEXT, PLAINTEXT, PLAINTEXT, PLAINTEXT, PLAINTEXT, PLAINTEXT, PLAINTEXT,
+            &aes,
+        );
+        let ciphertexts: [u8; BYTES_PER_BATCH] = ciphertexts[..].try_into().unwrap();
+        for i in 0..8 {
+            assert_eq!(
+                u128::from_ne_bytes(
+                    ciphertexts[BYTES_PER_AES_CALL * i..BYTES_PER_AES_CALL * (i + 1)]
+                        .try_into()
+                        .unwrap()
+                ),
+                CIPHERTEXT
+            );
+        }
+    }
+}

concrete-csprng/src/generators/implem/soft/generator.rs

@@ -0,0 +1,110 @@
+use crate::generators::aes_ctr::{AesCtrGenerator, AesKey, ChildrenIterator};
+use crate::generators::implem::soft::block_cipher::SoftwareBlockCipher;
+use crate::generators::{ByteCount, BytesPerChild, ChildrenCount, ForkError, RandomGenerator};
+use crate::seeders::Seed;
+
+/// A random number generator using a software implementation.
+pub struct SoftwareRandomGenerator(pub(super) AesCtrGenerator<SoftwareBlockCipher>);
+
+/// The children iterator used by [`SoftwareRandomGenerator`].
+///
+/// Outputs children generators one by one.
+pub struct SoftwareChildrenIterator(ChildrenIterator<SoftwareBlockCipher>);
+
+impl Iterator for SoftwareChildrenIterator {
+    type Item = SoftwareRandomGenerator;
+
+    fn next(&mut self) -> Option<Self::Item> {
+        self.0.next().map(SoftwareRandomGenerator)
+    }
+}
+
+impl RandomGenerator for SoftwareRandomGenerator {
+    type ChildrenIter = SoftwareChildrenIterator;
+    fn new(seed: Seed) -> Self {
+        SoftwareRandomGenerator(AesCtrGenerator::new(AesKey(seed.0), None, None))
+    }
+    fn remaining_bytes(&self) -> ByteCount {
+        self.0.remaining_bytes()
+    }
+    fn try_fork(
+        &mut self,
+        n_children: ChildrenCount,
+        n_bytes: BytesPerChild,
+    ) -> Result<Self::ChildrenIter, ForkError> {
+        self.0
+            .try_fork(n_children, n_bytes)
+            .map(SoftwareChildrenIterator)
+    }
+}
+
+impl Iterator for SoftwareRandomGenerator {
+    type Item = u8;
+
+    fn next(&mut self) -> Option<Self::Item> {
+        self.0.next()
+    }
+}
+
+#[cfg(test)]
+mod test {
+    use super::*;
+    use crate::generators::aes_ctr::aes_ctr_generic_test;
+    use crate::generators::generator_generic_test;
+
+    #[test]
+    fn prop_fork_first_state_table_index() {
+        aes_ctr_generic_test::prop_fork_first_state_table_index::<SoftwareBlockCipher>();
+    }
+
+    #[test]
+    fn prop_fork_last_bound_table_index() {
+        aes_ctr_generic_test::prop_fork_last_bound_table_index::<SoftwareBlockCipher>();
+    }
+
+    #[test]
+    fn prop_fork_parent_bound_table_index() {
+        aes_ctr_generic_test::prop_fork_parent_bound_table_index::<SoftwareBlockCipher>();
+    }
+
+    #[test]
+    fn prop_fork_parent_state_table_index() {
+        aes_ctr_generic_test::prop_fork_parent_state_table_index::<SoftwareBlockCipher>();
+    }
+
+    #[test]
+    fn prop_fork() {
+        aes_ctr_generic_test::prop_fork::<SoftwareBlockCipher>();
+    }
+
+    #[test]
+    fn prop_fork_children_remaining_bytes() {
+        aes_ctr_generic_test::prop_fork_children_remaining_bytes::<SoftwareBlockCipher>();
+    }
+
+    #[test]
+    fn prop_fork_parent_remaining_bytes() {
+        aes_ctr_generic_test::prop_fork_parent_remaining_bytes::<SoftwareBlockCipher>();
+    }
+
+    #[test]
+    fn test_roughly_uniform() {
+        generator_generic_test::test_roughly_uniform::<SoftwareRandomGenerator>();
+    }
+
+    #[test]
+    fn test_fork() {
+        generator_generic_test::test_fork_children::<SoftwareRandomGenerator>();
+    }
+
+    #[test]
+    fn test_generator_determinism() {
+        generator_generic_test::test_generator_determinism::<SoftwareRandomGenerator>();
+    }
+
+    #[test]
+    #[should_panic(expected = "expected test panic")]
+    fn test_bounded_panic() {
+        generator_generic_test::test_bounded_none_should_panic::<SoftwareRandomGenerator>();
+    }
+}

concrete-csprng/src/generators/implem/soft/mod.rs

@@ -0,0 +1,11 @@
+//! A module using a software fallback implementation of random number generator.
+
+mod block_cipher;
+
+mod generator;
+pub use generator::*;
+
+#[cfg(feature = "parallel")]
+mod parallel;
+#[cfg(feature = "parallel")]
+pub use parallel::*;

concrete-csprng/src/generators/implem/soft/parallel.rs

@@ -0,0 +1,94 @@
+use super::*;
+use crate::generators::aes_ctr::{AesCtrGenerator, ParallelChildrenIterator};
+use crate::generators::implem::soft::block_cipher::SoftwareBlockCipher;
+use crate::generators::{BytesPerChild, ChildrenCount, ForkError, ParallelRandomGenerator};
+use rayon::iter::plumbing::{Consumer, ProducerCallback, UnindexedConsumer};
+use rayon::prelude::*;
+
+/// The parallel children iterator used by [`SoftwareRandomGenerator`].
+///
+/// Outputs the children generators one by one.
+#[allow(clippy::type_complexity)]
+pub struct ParallelSoftwareChildrenIterator(
+    rayon::iter::Map<
+        ParallelChildrenIterator<SoftwareBlockCipher>,
+        fn(AesCtrGenerator<SoftwareBlockCipher>) -> SoftwareRandomGenerator,
+    >,
+);
+
+impl ParallelIterator for ParallelSoftwareChildrenIterator {
+    type Item = SoftwareRandomGenerator;
+    fn drive_unindexed<C>(self, consumer: C) -> C::Result
+    where
+        C: UnindexedConsumer<Self::Item>,
+    {
+        self.0.drive_unindexed(consumer)
+    }
+}
+
+impl IndexedParallelIterator for ParallelSoftwareChildrenIterator {
+    fn len(&self) -> usize {
+        self.0.len()
+    }
+    fn drive<C: Consumer<Self::Item>>(self, consumer: C) -> C::Result {
+        self.0.drive(consumer)
+    }
+    fn with_producer<CB: ProducerCallback<Self::Item>>(self, callback: CB) -> CB::Output {
+        self.0.with_producer(callback)
+    }
+}
+
+impl ParallelRandomGenerator for SoftwareRandomGenerator {
+    type ParChildrenIter = ParallelSoftwareChildrenIterator;
+
+    fn par_try_fork(
+        &mut self,
+        n_children: ChildrenCount,
+        n_bytes: BytesPerChild,
+    ) -> Result<Self::ParChildrenIter, ForkError> {
+        self.0
+            .par_try_fork(n_children, n_bytes)
+            .map(|iterator| ParallelSoftwareChildrenIterator(iterator.map(SoftwareRandomGenerator)))
+    }
+}
+
+#[cfg(test)]
+mod test {
+    use super::*;
+    use crate::generators::aes_ctr::aes_ctr_parallel_generic_tests;
+
+    #[test]
+    fn prop_fork_first_state_table_index() {
+        aes_ctr_parallel_generic_tests::prop_fork_first_state_table_index::<SoftwareBlockCipher>();
+    }
+
+    #[test]
+    fn prop_fork_last_bound_table_index() {
+        aes_ctr_parallel_generic_tests::prop_fork_last_bound_table_index::<SoftwareBlockCipher>();
+    }
+
+    #[test]
+    fn prop_fork_parent_bound_table_index() {
+        aes_ctr_parallel_generic_tests::prop_fork_parent_bound_table_index::<SoftwareBlockCipher>();
+    }
+
+    #[test]
+    fn prop_fork_parent_state_table_index() {
+        aes_ctr_parallel_generic_tests::prop_fork_parent_state_table_index::<SoftwareBlockCipher>();
+    }
+
+    #[test]
+    fn prop_fork() {
+        aes_ctr_parallel_generic_tests::prop_fork::<SoftwareBlockCipher>();
+    }
+
+    #[test]
+    fn prop_fork_children_remaining_bytes() {
+        aes_ctr_parallel_generic_tests::prop_fork_children_remaining_bytes::<SoftwareBlockCipher>();
+    }
+
+    #[test]
+    fn prop_fork_parent_remaining_bytes() {
+        aes_ctr_parallel_generic_tests::prop_fork_parent_remaining_bytes::<SoftwareBlockCipher>();
+    }
+}

concrete-csprng/src/generators/mod.rs

@@ -0,0 +1,235 @@
+//! A module containing random generators objects.
+//!
+//! See [crate-level](`crate`) explanations.
+use crate::seeders::Seed;
+use std::error::Error;
+use std::fmt::{Display, Formatter};
+
+/// The number of children created when a generator is forked.
+#[derive(Debug, Copy, Clone)]
+pub struct ChildrenCount(pub usize);
+
+/// The number of bytes each child can generate, when a generator is forked.
+#[derive(Debug, Copy, Clone)]
+pub struct BytesPerChild(pub usize);
+
+/// A structure representing the number of bytes between two table indices.
+#[derive(Clone, Copy, Debug, PartialOrd, Ord, PartialEq, Eq)]
+pub struct ByteCount(pub u128);
+
+/// An error occurring during a generator fork.
+#[derive(Debug)]
+pub enum ForkError {
+    ForkTooLarge,
+    ZeroChildrenCount,
+    ZeroBytesPerChild,
+}
+
+impl Display for ForkError {
+    fn fmt(&self, f: &mut Formatter<'_>) -> std::fmt::Result {
+        match self {
+            ForkError::ForkTooLarge => {
+                write!(
+                    f,
+                    "The children generators would output bytes after the parent bound. "
+                )
+            }
+            ForkError::ZeroChildrenCount => {
+                write!(
+                    f,
+                    "The number of children in the fork must be greater than zero."
+                )
+            }
+            ForkError::ZeroBytesPerChild => {
+                write!(
+                    f,
+                    "The number of bytes per child must be greater than zero."
+                )
+            }
+        }
+    }
+}
+impl Error for ForkError {}
+
+/// A trait for cryptographically secure pseudo-random generators.
+///
+/// See the [crate-level](#crate) documentation for details.
+pub trait RandomGenerator: Iterator<Item = u8> {
+    /// The iterator over children generators, returned by `try_fork` in case of success.
+    type ChildrenIter: Iterator<Item = Self>;
+
+    /// Creates a new generator from a seed.
+    ///
+    /// This operation is usually costly to perform, as the aes round keys need to be generated from
+    /// the seed.
+    fn new(seed: Seed) -> Self;
+
+    /// Returns the number of bytes that can still be outputted by the generator before reaching its
+    /// bound.
+    ///
+    /// Note:
+    /// -----
+    ///
+    /// A fresh generator can generate 2¹³² bytes. Unfortunately, no rust integer type in is able
+    /// to encode such a large number. Consequently [`ByteCount`] uses the largest integer type
+    /// available to encode this value: the `u128` type. For this reason, this method does not
+    /// effectively return the number of remaining bytes, but instead
+    /// `min(2¹²⁸-1, remaining_bytes)`.
+    fn remaining_bytes(&self) -> ByteCount;
+
+    /// Returns the next byte of the stream, if the generator did not yet reach its bound.
+    fn next_byte(&mut self) -> Option<u8> {
+        self.next()
+    }
+
+    /// Tries to fork the generator into an iterator of `n_children` new generators, each able to
+    /// output `n_bytes` bytes.
+    ///
+    /// Note:
+    /// -----
+    ///
+    /// To be successful, the number of remaining bytes for the parent generator must be larger than
+    /// `n_children*n_bytes`.
+    fn try_fork(
+        &mut self,
+        n_children: ChildrenCount,
+        n_bytes: BytesPerChild,
+    ) -> Result<Self::ChildrenIter, ForkError>;
+}
+
+/// A trait extending [`RandomGenerator`] to the parallel iterators of `rayon`.
+#[cfg(feature = "parallel")]
+pub trait ParallelRandomGenerator: RandomGenerator + Send {
+    /// The iterator over children generators, returned by `par_try_fork` in case of success.
+    type ParChildrenIter: rayon::prelude::IndexedParallelIterator<Item = Self>;
+
+    /// Tries to fork the generator into a parallel iterator of `n_children` new generators, each
+    /// able to output `n_bytes` bytes.
+    ///
+    /// Note:
+    /// -----
+    ///
+    /// To be successful, the number of remaining bytes for the parent generator must be larger than
+    /// `n_children*n_bytes`.
+    fn par_try_fork(
+        &mut self,
+        n_children: ChildrenCount,
+        n_bytes: BytesPerChild,
+    ) -> Result<Self::ParChildrenIter, ForkError>;
+}
+
+mod aes_ctr;
+
+mod implem;
+pub use implem::*;
+
+#[cfg(test)]
+pub mod generator_generic_test {
+    #![allow(unused)] // to please clippy when tests are not activated
+    use super::*;
+    use rand::Rng;
+
+    const REPEATS: usize = 1_000;
+
+    fn any_seed() -> impl Iterator<Item = Seed> {
+        std::iter::repeat_with(|| Seed(rand::thread_rng().gen()))
+    }
+
+    fn some_children_count() -> impl Iterator<Item = ChildrenCount> {
+        std::iter::repeat_with(|| ChildrenCount(rand::thread_rng().gen::<usize>() % 16 + 1))
+    }
+
+    fn some_bytes_per_child() -> impl Iterator<Item = BytesPerChild> {
+        std::iter::repeat_with(|| BytesPerChild(rand::thread_rng().gen::<usize>() % 128 + 1))
+    }
+
+    /// Checks that the PRNG roughly generates uniform numbers.
+    ///
+    /// To do that, we perform an histogram of the occurrences of each byte value, over a fixed
+    /// number of samples and check that the empirical probabilities of the bins are close to
+    /// the theoretical probabilities.
+    pub fn test_roughly_uniform<G: RandomGenerator>() {
+        // Number of bins to use for the histogram.
+        const N_BINS: usize = u8::MAX as usize + 1;
+        // Number of samples to use for the histogram.
+        let n_samples = 10_000_000_usize;
+        // Theoretical probability of a each bins.
+        let expected_prob: f64 = 1. / N_BINS as f64;
+        // Absolute error allowed on the empirical probabilities.
+        // This value was tuned to make the test pass on an arguably correct state of
+        // implementation. 10^-4 precision is arguably pretty fine for this rough test, but it would
+        // be interesting to improve this test.
+        let precision = 10f64.powi(-3);
+
+        for _ in 0..REPEATS {
+            // We instantiate a new generator.
+            let seed = any_seed().next().unwrap();
+            let mut generator = G::new(seed);
+            // We create a new histogram
+            let mut counts = [0usize; N_BINS];
+            // We fill the histogram.
+            for _ in 0..n_samples {
+                counts[generator.next_byte().unwrap() as usize] += 1;
+            }
+            // We check that the empirical probabilities are close enough to the theoretical one.
+            counts
+                .iter()
+                .map(|a| (*a as f64) / (n_samples as f64))
+                .for_each(|a| assert!((a - expected_prob).abs() < precision))
+        }
+    }
+
+    /// Checks that given a state and a key, the PRNG is determinist.
+    pub fn test_generator_determinism<G: RandomGenerator>() {
+        for _ in 0..REPEATS {
+            let seed = any_seed().next().unwrap();
+            let mut first_generator = G::new(seed);
+            let mut second_generator = G::new(seed);
+            for _ in 0..1024 {
+                assert_eq!(first_generator.next(), second_generator.next());
+            }
+        }
+    }
+
+    /// Checks that forks returns a bounded child, and that the proper number of bytes can be
+    /// generated.
+    pub fn test_fork_children<G: RandomGenerator>() {
+        for _ in 0..REPEATS {
+            let ((seed, n_children), n_bytes) = any_seed()
+                .zip(some_children_count())
+                .zip(some_bytes_per_child())
+                .next()
+                .unwrap();
+            let mut gen = G::new(seed);
+            let mut bounded = gen.try_fork(n_children, n_bytes).unwrap().next().unwrap();
+            assert_eq!(bounded.remaining_bytes(), ByteCount(n_bytes.0 as u128));
+            for _ in 0..n_bytes.0 {
+                bounded.next().unwrap();
+            }
+
+            // Assert we are at the bound
+            assert!(bounded.next().is_none());
+        }
+    }
+
+    /// Checks that a bounded prng returns none when exceeding the allowed number of bytes.
+    ///
+    /// To properly check for panic use `#[should_panic(expected = "expected test panic")]` as an
+    /// attribute on the test function.
+    pub fn test_bounded_none_should_panic<G: RandomGenerator>() {
+        let ((seed, n_children), n_bytes) = any_seed()
+            .zip(some_children_count())
+            .zip(some_bytes_per_child())
+            .next()
+            .unwrap();
+        let mut gen = G::new(seed);
+        let mut bounded = gen.try_fork(n_children, n_bytes).unwrap().next().unwrap();
+        assert_eq!(bounded.remaining_bytes(), ByteCount(n_bytes.0 as u128));
+        for _ in 0..n_bytes.0 {
+            assert!(bounded.next().is_some());
+        }
+
+        // One call too many, should panic
+        bounded.next().ok_or("expected test panic").unwrap();
+    }
+}

concrete-csprng/src/lib.rs

@@ -0,0 +1,114 @@
+#![deny(rustdoc::broken_intra_doc_links)]
+//! Cryptographically secure pseudo random number generator.
+//!
+//! Welcome to the `concrete-csprng` documentation.
+//!
+//! This crate provides a fast cryptographically secure pseudo-random number generator, suited to
+//! work in a multithreaded setting.
+//!
+//! Random Generators
+//! =================
+//!
+//! The central abstraction of this crate is the [`RandomGenerator`](generators::RandomGenerator)
+//! trait, which is implemented by different types, each supporting a different platform. In
+//! essence, a type implementing [`RandomGenerator`](generators::RandomGenerator) is a type that
+//! outputs a new pseudo-random byte at each call to
+//! [`next_byte`](generators::RandomGenerator::next_byte). Such a generator `g` can be seen as
+//! enclosing a growing index into an imaginary array of pseudo-random bytes:
+//! ```ascii
+//!   0 1 2 3 4 5 6 7 8 9     M-1     │
+//!  ┏━┯━┯━┯━┯━┯━┯━┯━┯━┯━┯━━━┯━┓      │
+//!  ┃ │ │ │ │ │ │ │ │ │ │...│ ┃      │
+//!  ┗↥┷━┷━┷━┷━┷━┷━┷━┷━┷━┷━━━┷━┛      │
+//!   g                               │
+//!                                   │
+//!   g.next_byte()                   │
+//!                                   │
+//!   0 1 2 3 4 5 6 7 8 9     M-1     │
+//!  ┏━┯━┯━┯━┯━┯━┯━┯━┯━┯━┯━━━┯━┓      │
+//!  ┃╳│ │ │ │ │ │ │ │ │ │...│ ┃      │
+//!  ┗━┷↥┷━┷━┷━┷━┷━┷━┷━┷━┷━━━┷━┛      │
+//!     g                             │
+//!                                   │
+//!   g.next_byte()                   │  legend:
+//!                                   │  -------
+//!   0 1 2 3 4 5 6 7 8 9     M-1     │   ↥ : next byte to be outputted by g
+//!  ┏━┯━┯━┯━┯━┯━┯━┯━┯━┯━┯━━━┯━┓      │  │ │: byte not yet outputted by g
+//!  ┃╳│╳│ │ │ │ │ │ │ │ │...│ ┃      │  │╳│: byte already outputted by g
+//!  ┗━┷━┷↥┷━┷━┷━┷━┷━┷━┷━┷━━━┷━┛      │  
+//!       g                           🭭
+//! ```
+//!
+//! While being large, this imaginary array is still bounded to M = 2¹³² bytes. Consequently, a
+//! generator is always bounded to a maximal index. That is, there is always a max amount of
+//! elements of this array that can be outputted by the generator. By default, generators created
+//! via [`new`](generators::RandomGenerator::new) are always bounded to M-1.
+//!
+//! Tree partition of the pseudo-random stream
+//! ==========================================
+//!
+//! One particularity of this implementation is that you can use the
+//! [`try_fork`](generators::RandomGenerator::try_fork) method to create an arbitrary partition tree
+//! of a region of this array. Indeed, calling `try_fork(nc, nb)` outputs `nc` new generators, each
+//! able to output `nb` bytes. The `try_fork` method ensures that the states and bounds of the
+//! parent and children generators are set so as to prevent the same substream to be outputted
+//! twice:
+//! ```ascii
+//!   0 1 2 3 4 5 6 7 8 9     M   │   
+//!  ┏━┯━┯━┯━┯━┯━┯━┯━┯━┯━┯━━━┯━┓  │  
+//!  ┃P│P│P│P│P│P│P│P│P│P│...│P┃  │  
+//!  ┗↥┷━┷━┷━┷━┷━┷━┷━┷━┷━┷━━━┷━┛  │  
+//!   p                           │  
+//!                               │  
+//!   (a,b) = p.fork(2,4)         │  
+//!                               │
+//!   0 1 2 3 4 5 6 7 8 9     M   │
+//!  ┏━┯━┯━┯━┯━┯━┯━┯━┯━┯━┯━━━┯━┓  │
+//!  ┃A│A│A│A│B│B│B│B│P│P│...│P┃  │
+//!  ┗↥┷━┷━┷━┷↥┷━┷━┷━┷↥┷━┷━━━┷━┛  │
+//!   a       b       p           │
+//!                               │  legend:
+//!   (c,d) = b.fork(2, 1)        │  -------
+//!                               │   ↥ : next byte to be outputted by p
+//!   0 1 2 3 4 5 6 7 8 9     M   │  │P│: byte to be outputted by p
+//!  ┏━┯━┯━┯━┯━┯━┯━┯━┯━┯━┯━━━┯━┓  │  │╳│: byte already outputted
+//!  ┃A│A│A│A│C│D│B│B│P│P│...│P┃  │  
+//!  ┗↥┷━┷━┷━┷↥┷↥┷↥┷━┷↥┷━┷━━━┷━┛  │
+//!   a       c d b   p           🭭
+//! ```
+//!
+//! This makes it possible to consume the stream at different places. This is particularly useful in
+//! a multithreaded setting, in which we want to use the same generator from different independent
+//! threads:
+//!
+//! ```ascii
+//!   0 1 2 3 4 5 6 7 8 9     M   │   
+//!  ┏━┯━┯━┯━┯━┯━┯━┯━┯━┯━┯━━━┯━┓  │  
+//!  ┃A│A│A│A│C│D│B│B│P│P│...│P┃  │  
+//!  ┗↥┷━┷━┷━┷↥┷↥┷↥┷━┷↥┷━┷━━━┷━┛  │  
+//!   a       c d b   p           │  
+//!                               │  
+//!   a.next_byte()               │  
+//!                               │
+//!   0 1 2 3 4 5 6 7 8 9     M   │
+//!  ┏━┯━┯━┯━┯━┯━┯━┯━┯━┯━┯━━━┯━┓  │
+//!  ┃╳│A│A│A│C│D│B│B│P│P│...│P┃  │
+//!  ┗━┷↥┷━┷━┷↥┷↥┷↥┷━┷↥┷━┷━━━┷━┛  │
+//!     a     c d b   p           │
+//!                               │  legend:
+//!   b.next_byte()               │  -------
+//!                               │   ↥ : next byte to be outputted by p
+//!   0 1 2 3 4 5 6 7 8 9     M   │  │P│: byte to be outputted by p  
+//!  ┏━┯━┯━┯━┯━┯━┯━┯━┯━┯━┯━━━┯━┓  │  │╳│: byte already outputted
+//!  ┃╳│A│A│A│C│D│╳│B│P│P│...│P┃  │  
+//!  ┗━┷↥┷━┷━┷↥┷↥┷━┷↥┷↥┷━┷━━━┷━┛  │
+//!     a     c d   b p           🭭
+//! ```
+//!
+//! Implementation
+//! ==============
+//!
+//! The implementation is based on the AES blockcipher used in counter (CTR) mode, as presented
+//! in the ISO/IEC 18033-4 document.
+pub mod generators;
+pub mod seeders;

concrete-csprng/src/seeders/implem/apple_secure_enclave_seeder.rs

@@ -0,0 +1,141 @@
+use crate::seeders::{Seed, Seeder};
+use libc;
+use std::cmp::Ordering;
+
+/// There is no `rseed` equivalent in the ARM specification until `ARMv8.5-A`.
+/// However it seems that these instructions are not exposed in `core::arch::aarch64`.
+///
+/// Our primary interest for supporting aarch64 targets is AppleSilicon support
+/// which for the M1 macs available, they are based on the `ARMv8.4-A` set.
+///
+/// So we fall back to using a function from Apple's API which
+/// uses the [Secure Enclave] to generate cryptographically secure random bytes.
+///
+/// [Secure Enclave]: https://support.apple.com/fr-fr/guide/security/sec59b0b31ff/web
+mod secure_enclave {
+    pub enum __SecRandom {}
+    pub type SecRandomRef = *const __SecRandom;
+    use libc::{c_int, c_void};
+
+    #[link(name = "Security", kind = "framework")]
+    extern "C" {
+        pub static kSecRandomDefault: SecRandomRef;
+
+        pub fn SecRandomCopyBytes(rnd: SecRandomRef, count: usize, bytes: *mut c_void) -> c_int;
+    }
+
+    pub fn generate_random_bytes(bytes: &mut [u8]) -> std::io::Result<()> {
+        // As per Apple's documentation:
+        // - https://developer.apple.com/documentation/security/randomization_services?language=objc
+        // - https://developer.apple.com/documentation/security/1399291-secrandomcopybytes?language=objc
+        //
+        // The `SecRandomCopyBytes` "Generate cryptographically secure random numbers"
+        unsafe {
+            let res = SecRandomCopyBytes(
+                kSecRandomDefault,
+                bytes.len(),
+                bytes.as_mut_ptr() as *mut c_void,
+            );
+            if res != 0 {
+                Err(std::io::Error::last_os_error())
+            } else {
+                Ok(())
+            }
+        }
+    }
+}
+
+/// A seeder which uses the `SecRandomCopyBytes` function from Apple's `Security` framework.
+///
+/// <https://developer.apple.com/documentation/security/1399291-secrandomcopybytes?language=objc>
+pub struct AppleSecureEnclaveSeeder;
+
+impl Seeder for AppleSecureEnclaveSeeder {
+    fn seed(&mut self) -> Seed {
+        // 16 bytes == 128 bits
+        let mut bytes = [0u8; 16];
+        secure_enclave::generate_random_bytes(&mut bytes)
+            .expect("Failure while using Apple secure enclave: {err:?}");
+
+        Seed(u128::from_le_bytes(bytes))
+    }
+
+    fn is_available() -> bool {
+        let os_version_sysctl_name = match std::ffi::CString::new("kern.osproductversion") {
+            Ok(c_str) => c_str,
+            _ => return false,
+        };
+
+        // Big enough buffer to get a version output as an ASCII string
+        const OUTPUT_BUFFER_SIZE: usize = 64;
+        let mut output_buffer_size = OUTPUT_BUFFER_SIZE;
+        let mut output_buffer = [0u8; OUTPUT_BUFFER_SIZE];
+        let res = unsafe {
+            libc::sysctlbyname(
+                os_version_sysctl_name.as_ptr() as *const _ as *const _,
+                &mut output_buffer as *mut _ as *mut _,
+                &mut output_buffer_size as *mut _ as *mut _,
+                std::ptr::null_mut(),
+                0,
+            )
+        };
+
+        if res != 0 {
+            return false;
+        }
+
+        let result_c_str =
+            match std::ffi::CStr::from_bytes_with_nul(&output_buffer[..output_buffer_size]) {
+                Ok(c_str) => c_str,
+                _ => return false,
+            };
+
+        let result_string = match result_c_str.to_str() {
+            Ok(str) => str,
+            _ => return false,
+        };
+
+        // Normally we get a major version and minor version
+        let split_string: Vec<&str> = result_string.split('.').collect();
+
+        let mut major = -1;
+        let mut minor = -1;
+
+        // Major part of the version string
+        if !split_string.is_empty() {
+            major = match split_string[0].parse() {
+                Ok(major_from_str) => major_from_str,
+                _ => return false,
+            };
+        }
+
+        // SecRandomCopyBytes is available starting with mac OS 10.7
+        // https://developer.apple.com/documentation/security/1399291-secrandomcopybytes?language=objc
+        // This match pattern is recommended by clippy, so we oblige here
+        match major.cmp(&10) {
+            Ordering::Greater => true,
+            Ordering::Equal => {
+                // Minor part of the version string
+                if split_string.len() >= 2 {
+                    minor = match split_string[1].parse() {
+                        Ok(minor_from_str) => minor_from_str,
+                        _ => return false,
+                    };
+                }
+                minor >= 7
+            }
+            Ordering::Less => false,
+        }
+    }
+}
+
+#[cfg(test)]
+mod test {
+    use super::*;
+    use crate::seeders::generic_tests::check_seeder_fixed_sequences_different;
+
+    #[test]
+    fn check_bounded_sequence_difference() {
+        check_seeder_fixed_sequences_different(|_| AppleSecureEnclaveSeeder);
+    }
+}

concrete-csprng/src/seeders/implem/mod.rs

@@ -0,0 +1,14 @@
+#[cfg(target_os = "macos")]
+mod apple_secure_enclave_seeder;
+#[cfg(target_os = "macos")]
+pub use apple_secure_enclave_seeder::AppleSecureEnclaveSeeder;
+
+#[cfg(feature = "seeder_x86_64_rdseed")]
+mod rdseed;
+#[cfg(feature = "seeder_x86_64_rdseed")]
+pub use rdseed::RdseedSeeder;
+
+#[cfg(feature = "seeder_unix")]
+mod unix;
+#[cfg(feature = "seeder_unix")]
+pub use unix::UnixSeeder;

concrete-csprng/src/seeders/implem/rdseed.rs

@@ -0,0 +1,51 @@
+use crate::seeders::{Seed, Seeder};
+
+/// A seeder which uses the `rdseed` x86_64 instruction.
+///
+/// The `rdseed` instruction allows to deliver seeds from a hardware source of entropy see
+/// <https://www.felixcloutier.com/x86/rdseed> .
+pub struct RdseedSeeder;
+
+impl Seeder for RdseedSeeder {
+    fn seed(&mut self) -> Seed {
+        Seed(unsafe { rdseed_random_m128() })
+    }
+
+    fn is_available() -> bool {
+        is_x86_feature_detected!("rdseed")
+    }
+}
+
+// Generates a random 128 bits value from rdseed
+#[target_feature(enable = "rdseed")]
+unsafe fn rdseed_random_m128() -> u128 {
+    let mut rand1: u64 = 0;
+    let mut rand2: u64 = 0;
+    let mut output_bytes = [0u8; 16];
+    unsafe {
+        loop {
+            if core::arch::x86_64::_rdseed64_step(&mut rand1) == 1 {
+                break;
+            }
+        }
+        loop {
+            if core::arch::x86_64::_rdseed64_step(&mut rand2) == 1 {
+                break;
+            }
+        }
+    }
+    output_bytes[0..8].copy_from_slice(&rand1.to_ne_bytes());
+    output_bytes[8..16].copy_from_slice(&rand2.to_ne_bytes());
+    u128::from_ne_bytes(output_bytes)
+}
+
+#[cfg(test)]
+mod test {
+    use super::*;
+    use crate::seeders::generic_tests::check_seeder_fixed_sequences_different;
+
+    #[test]
+    fn check_bounded_sequence_difference() {
+        check_seeder_fixed_sequences_different(|_| RdseedSeeder);
+    }
+}

concrete-csprng/src/seeders/implem/unix.rs

@@ -0,0 +1,72 @@
+use crate::seeders::{Seed, Seeder};
+use std::fs::File;
+use std::io::Read;
+
+/// A seeder which uses the `/dev/random` source on unix-like systems.
+pub struct UnixSeeder {
+    counter: u128,
+    secret: u128,
+    file: File,
+}
+
+impl UnixSeeder {
+    /// Creates a new seeder from a user defined secret.
+    ///
+    /// Important:
+    /// ----------
+    ///
+    /// This secret is used to ensure the quality of the seed in scenarios where `/dev/random` may
+    /// be compromised.
+    ///
+    /// The attack hypotheses are as follow:
+    /// - `/dev/random` output can be predicted by a process running on the machine by just
+    ///   observing various states of the machine
+    /// - The attacker cannot read data from the process where `concrete-csprng` is running
+    ///
+    /// Using a secret in `concrete-csprng` allows to generate values that the attacker cannot
+    /// predict, making this seeder secure on systems were `/dev/random` outputs can be
+    /// predicted.
+    pub fn new(secret: u128) -> UnixSeeder {
+        let file = std::fs::File::open("/dev/random").expect("Failed to open /dev/random .");
+        let counter = std::time::UNIX_EPOCH
+            .elapsed()
+            .expect("Failed to initialize unix seeder.")
+            .as_nanos();
+        UnixSeeder {
+            secret,
+            counter,
+            file,
+        }
+    }
+}
+
+impl Seeder for UnixSeeder {
+    fn seed(&mut self) -> Seed {
+        let output = self.secret ^ self.counter ^ dev_random(&mut self.file);
+        self.counter = self.counter.wrapping_add(1);
+        Seed(output)
+    }
+
+    fn is_available() -> bool {
+        cfg!(target_family = "unix")
+    }
+}
+
+fn dev_random(random: &mut File) -> u128 {
+    let mut buf = [0u8; 16];
+    random
+        .read_exact(&mut buf[..])
+        .expect("Failed to read from /dev/random .");
+    u128::from_ne_bytes(buf)
+}
+
+#[cfg(test)]
+mod test {
+    use super::*;
+    use crate::seeders::generic_tests::check_seeder_fixed_sequences_different;
+
+    #[test]
+    fn check_bounded_sequence_difference() {
+        check_seeder_fixed_sequences_different(UnixSeeder::new);
+    }
+}

concrete-csprng/src/seeders/mod.rs

@@ -0,0 +1,50 @@
+//! A module containing seeders objects.
+//!
+//! When initializing a generator, one needs to provide a [`Seed`], which is then used as key to the
+//! AES blockcipher. As a consequence, the quality of the outputs of the generator is directly
+//! conditioned by the quality of this seed. This module proposes different mechanisms to deliver
+//! seeds that can accommodate varying scenarios.
+
+/// A seed value, used to initialize a generator.
+#[derive(Debug, Copy, Clone, PartialEq, Eq)]
+pub struct Seed(pub u128);
+
+/// A trait representing a seeding strategy.
+pub trait Seeder {
+    /// Generates a new seed.
+    fn seed(&mut self) -> Seed;
+
+    /// Check whether the seeder can be used on the current machine. This function may check if some
+    /// required CPU features are available or if some OS features are available for example.
+    fn is_available() -> bool
+    where
+        Self: Sized;
+}
+
+mod implem;
+// This import statement can be empty if seeder features are disabled, rustc's behavior changed to
+// warn of empty modules, we know this can happen, so allow it.
+#[allow(unused_imports)]
+pub use implem::*;
+
+#[cfg(test)]
+mod generic_tests {
+    use crate::seeders::Seeder;
+
+    /// Naively verifies that two fixed-size sequences generated by repeatedly calling the seeder
+    /// are different.
+    #[allow(unused)] // to please clippy when tests are not activated
+    pub fn check_seeder_fixed_sequences_different<S: Seeder, F: Fn(u128) -> S>(
+        construct_seeder: F,
+    ) {
+        const SEQUENCE_SIZE: usize = 500;
+        const REPEATS: usize = 10_000;
+        for i in 0..REPEATS {
+            let mut seeder = construct_seeder(i as u128);
+            let orig_seed = seeder.seed();
+            for _ in 0..SEQUENCE_SIZE {
+                assert_ne!(seeder.seed(), orig_seed);
+            }
+        }
+    }
+}

docker/Dockerfile.wasm_tests

@@ -11,6 +11,7 @@ RUN sed -i 's|^deb http://archive.ubuntu.com/ubuntu/|deb http://mirror.ubuntu.ik
 ENV CARGO_TARGET_DIR=/root/tfhe-rs-target
 
 ARG RUST_TOOLCHAIN="stable"
+ARG NODE_VERSION
 
 WORKDIR /tfhe-wasm-tests
 
@@ -34,6 +35,6 @@ RUN curl --proto '=https' --tlsv1.2 -sSf https://sh.rustup.rs > install-rustup.s
     chmod +x install-node.sh && \
     ./install-node.sh && \
     . "$HOME/.nvm/nvm.sh" && \
-    bash -i -c 'nvm install node && nvm use node'
+    bash -i -c 'nvm install ${NODE_VERSION} && nvm use ${NODE_VERSION}'
 
 WORKDIR /tfhe-wasm-tests/tfhe-rs/

scripts/dieharder_test.sh

@@ -0,0 +1,77 @@
+#!/usr/bin/env bash
+
+# dieharder does not support running a subset of its tests, so we'll check which ones are not good
+# and ignore the output from those tests in the final log
+
+set -e
+
+DIEHARDER_RUN_LOG_FILE="dieharder_run.log"
+
+bad_tests="$(dieharder -l | \
+# select lines with the -d
+grep -w '\-d' | \
+# forget about the good tests
+grep -v -i 'good' | \
+# get the test id
+cut -d ' ' -f 4 | \
+# nice formatting
+xargs)"
+
+
+bad_test_filter=""
+for bad_test in ${bad_tests}; do
+    bad_test_filter="${bad_test_filter:+${bad_test_filter}|}$(dieharder -d "${bad_test}" -t 1 -p 1 -D test_name | xargs)"
+done
+
+echo "The following tests will be ignored as they are marked as either 'suspect' or 'do not use': "
+echo ""
+echo "${bad_test_filter}"
+echo ""
+
+# by default we may have no pv just forward the input
+pv="cat"
+if which pv > /dev/null; then
+    pv="pv -t -a -b"
+fi
+
+rm -f "${DIEHARDER_RUN_LOG_FILE}"
+
+# ignore potential errors and parse the log afterwards
+set +e
+
+# We are writing in both cases
+# shellcheck disable=SC2094
+./target/release/examples/generate 2>"${DIEHARDER_RUN_LOG_FILE}" | \
+$pv | \
+# -a: all tests
+# -g 200: get random bytes from input
+# -Y 1: disambiguate results, i.e. if a weak result appear check if it's a random failure/weakness
+# -k 2: better maths formulas to determine some test statistics
+dieharder -a -g 200 -Y 1 -k 2 | \
+tee -a "${DIEHARDER_RUN_LOG_FILE}"
+set -e
+
+printf "\n\n"
+
+cat "${DIEHARDER_RUN_LOG_FILE}"
+
+if ! grep -q -i "failed" < "${DIEHARDER_RUN_LOG_FILE}"; then
+    echo "All tests passed!"
+    exit 0
+fi
+
+printf "\n\n"
+
+failed_tests="$(grep -i "failed" < "${DIEHARDER_RUN_LOG_FILE}")"
+true_failed_test="$(grep -i "failed" < "${DIEHARDER_RUN_LOG_FILE}" | { grep -v -E "${bad_test_filter}" || true; } | sed -z '$ s/\n$//')"
+
+if [[ "${true_failed_test}" == "" ]]; then
+    echo "There were test failures, but the tests were either marked as 'suspect' or 'do not use'"
+    echo "${failed_tests}"
+    exit 0
+fi
+
+echo "The following tests failed:"
+echo "${true_failed_test}"
+
+exit 1

scripts/integer-tests.sh

@@ -8,14 +8,21 @@ function usage() {
     echo "--help                    Print this message"
     echo "--rust-toolchain          The toolchain to run the tests with default: stable"
     echo "--multi-bit               Run multi-bit tests only: default off"
+    echo "--unsigned-only           Run only unsigned integer tests, by default both signed and unsigned tests are run"
+    echo "--signed-only             Run only signed integer tests, by default both signed and unsigned tests are run"
     echo "--cargo-profile           The cargo profile used to build tests"
+    echo "--avx512-support          Set to ON to enable avx512"
     echo
 }
 
 RUST_TOOLCHAIN="+stable"
 multi_bit=""
 not_multi_bit="_multi_bit"
+# Run signed test by default
+signed=""
+not_signed=""
 cargo_profile="release"
+avx512_feature=""
 
 while [ -n "$1" ]
 do
@@ -35,11 +42,28 @@ do
             not_multi_bit=""
             ;;
 
+        "--unsigned-only" )
+            signed=""
+            not_signed="_signed"
+            ;;
+
+        "--signed-only" )
+            signed="_signed"
+            not_signed=""
+            ;;
+
         "--cargo-profile" )
             shift
             cargo_profile="$1"
             ;;
 
+        "--avx512-support" )
+            shift
+            if [[ "$1" == "ON" ]]; then
+                avx512_feature=nightly-avx512
+            fi
+            ;;
+
         *)
             echo "Unknown param : $1"
             exit 1
@@ -62,104 +86,86 @@ if [[ $(uname) == "Darwin" ]]; then
     nproc_bin="sysctl -n hw.logicalcpu"
 fi
 
-n_threads="$(${nproc_bin})"
+# TODO autodetect/have a finer CPU count depending on memory
+num_cpu_threads="$(${nproc_bin})"
 
 if uname -a | grep "arm64"; then
     if [[ $(uname) == "Darwin" ]]; then
         # Keys are 4.7 gigs at max, CI M1 macs only has 8 gigs of RAM
-        n_threads=1
+        small_instance_n_threads=1
     fi
 else
     # Keys are 4.7 gigs at max, test machine has 32 gigs of RAM
-    n_threads=6
+    small_instance_n_threads=6
 fi
 
-if [[ "${BIG_TESTS_INSTANCE}" != TRUE ]]; then
-    if [[ "${FAST_TESTS}" != TRUE ]]; then
-        # block pbs are too slow for high params
-        # mul_crt_4_4 is extremely flaky (~80% failure)
-        # test_wopbs_bivariate_crt_wopbs_param_message generate tables that are too big at the moment
-        # test_integer_smart_mul_param_message_4_carry_4_ks_pbs is too slow
-        # so is test_integer_default_add_sequence_multi_thread_param_message_4_carry_4_ks_pbs
-        filter_expression="""\
-test(/^integer::.*${multi_bit}/) \
-${not_multi_bit:+"and not test(~${not_multi_bit})"} \
-and not test(/.*_block_pbs(_base)?_param_message_[34]_carry_[34]_ks_pbs$/) \
-and not test(~mul_crt_param_message_4_carry_4_ks_pbs) \
-and not test(/.*test_wopbs_bivariate_crt_wopbs_param_message_[34]_carry_[34]_ks_pbs$/) \
-and not test(/.*test_integer_smart_mul_param_message_4_carry_4_ks_pbs$/) \
-and not test(/.*test_integer_default_add_sequence_multi_thread_param_message_4_carry_4_ks_pbs$/)"""
-    else
-        # test only fast default operations with only two set of parameters
-        filter_expression="""\
-test(/^integer::.*${multi_bit}/) \
-${not_multi_bit:+"and not test(~${not_multi_bit})"} \
-and test(/.*_default_.*?_param${multi_bit}_message_[2-3]_carry_[2-3]${multi_bit:+"_group_2"}_ks_pbs/) \
-and not test(/.*_param_message_[14]_carry_[14]_ks_pbs$/) \
-and not test(/.*default_add_sequence_multi_thread_param_message_3_carry_3_ks_pbs$/)"""
-    fi
-
-    cargo "${RUST_TOOLCHAIN}" nextest run \
-        --tests \
-        --cargo-profile "${cargo_profile}" \
-        --package tfhe \
-        --profile ci \
-        --features="${ARCH_FEATURE}",integer,internal-keycache \
-        --test-threads "${n_threads}" \
-        -E "$filter_expression"
-
-    if [[ "${multi_bit}" == "" ]]; then
-        cargo "${RUST_TOOLCHAIN}" test \
-            --profile "${cargo_profile}" \
-            --package tfhe \
-            --features="${ARCH_FEATURE}",integer,internal-keycache \
-            --doc \
-            -- integer::
-    fi
+if [[ "${BIG_TESTS_INSTANCE}" == TRUE ]]; then
+    test_threads="$((num_cpu_threads * 1 / 2))"
+    doctest_threads="${num_cpu_threads}"
 else
-    if [[ "${FAST_TESTS}" != TRUE ]]; then
-        # block pbs are too slow for high params
-        # mul_crt_4_4 is extremely flaky (~80% failure)
-        # test_wopbs_bivariate_crt_wopbs_param_message generate tables that are too big at the moment
-        # test_integer_smart_mul_param_message_4_carry_4_ks_pbs is too slow
-        # so is test_integer_default_add_sequence_multi_thread_param_message_4_carry_4_ks_pbs
-        filter_expression="""\
+    test_threads="${small_instance_n_threads}"
+    doctest_threads="${num_cpu_threads}"
+fi
+
+# block pbs are too slow for high params
+# mul_crt_4_4 is extremely flaky (~80% failure)
+# test_wopbs_bivariate_crt_wopbs_param_message generate tables that are too big at the moment
+# test_integer_smart_mul_param_message_4_carry_4_ks_pbs is too slow
+# so is test_integer_default_add_sequence_multi_thread_param_message_4_carry_4_ks_pbs
+# we skip smart_div, smart_rem which are already covered by the smar_div_rem test
+# we similarly skip default_div, default_rem which are covered by default_div_rem
+full_test_filter_expression="""\
 test(/^integer::.*${multi_bit}/) \
+${signed:+"and test(/^integer::.*${signed}/)"} \
 ${not_multi_bit:+"and not test(~${not_multi_bit})"} \
+${not_signed:+"and not test(~${not_signed})"} \
+and not test(/.*integer_smart_div_param/) \
+and not test(/.*integer_smart_rem_param/) \
+and not test(/.*integer_default_div_param/) \
+and not test(/.*integer_default_rem_param/) \
 and not test(/.*_block_pbs(_base)?_param_message_[34]_carry_[34]_ks_pbs$/) \
 and not test(~mul_crt_param_message_4_carry_4_ks_pbs) \
 and not test(/.*test_wopbs_bivariate_crt_wopbs_param_message_[34]_carry_[34]_ks_pbs$/) \
 and not test(/.*test_integer_smart_mul_param_message_4_carry_4_ks_pbs$/) \
 and not test(/.*test_integer_default_add_sequence_multi_thread_param_message_4_carry_4_ks_pbs$/)"""
-    else
-        # test only fast default operations with only two set of parameters
-        filter_expression="""\
+
+# test only fast default operations with only two set of parameters
+# we skip default_div, default_rem which are covered by default_div_rem
+fast_test_filter_expression="""\
 test(/^integer::.*${multi_bit}/) \
+${signed:+"and test(/^integer::.*${signed}/)"} \
 ${not_multi_bit:+"and not test(~${not_multi_bit})"} \
+${not_signed:+"and not test(~${not_signed})"} \
 and test(/.*_default_.*?_param${multi_bit}_message_[2-3]_carry_[2-3]${multi_bit:+"_group_2"}_ks_pbs/) \
+and not test(/.*integer_default_div_param/) \
+and not test(/.*integer_default_rem_param/) \
 and not test(/.*_param_message_[14]_carry_[14]_ks_pbs$/) \
 and not test(/.*default_add_sequence_multi_thread_param_message_3_carry_3_ks_pbs$/)"""
-    fi
 
-    num_cpu_threads="$(${nproc_bin})"
-    num_threads=$((num_cpu_threads * 1 / 2))
-    cargo "${RUST_TOOLCHAIN}" nextest run \
-        --tests \
-        --cargo-profile "${cargo_profile}" \
+if [[ "${FAST_TESTS}" == "TRUE" ]]; then
+    echo "Running 'fast' test set'"
+    filter_expression="${fast_test_filter_expression}"
+else
+    echo "Running 'slow' test set"
+    filter_expression="${full_test_filter_expression}"
+fi
+
+cargo "${RUST_TOOLCHAIN}" nextest run \
+    --tests \
+    --cargo-profile "${cargo_profile}" \
+    --package tfhe \
+    --profile ci \
+    --features="${ARCH_FEATURE}",integer,internal-keycache,"${avx512_feature}" \
+    --test-threads "${test_threads}" \
+    -E "$filter_expression"
+
+if [[ "${multi_bit}" == "" ]]; then
+    cargo "${RUST_TOOLCHAIN}" test \
+        --profile "${cargo_profile}" \
         --package tfhe \
-        --profile ci \
-        --features="${ARCH_FEATURE}",integer,internal-keycache \
-        --test-threads $num_threads \
-        -E "$filter_expression"
-
-    if [[ "${multi_bit}" == "" ]]; then
-        cargo "${RUST_TOOLCHAIN}" test \
-            --profile "${cargo_profile}" \
-            --package tfhe \
-            --features="${ARCH_FEATURE}",integer,internal-keycache \
-            --doc \
-            -- --test-threads="$(${nproc_bin})" integer::
-    fi
+        --features="${ARCH_FEATURE}",integer,internal-keycache,"${avx512_feature}" \
+        --doc \
+        -- --test-threads="${doctest_threads}" integer::
 fi
 
 echo "Test ran in $SECONDS seconds"

scripts/shortint-tests.sh

@@ -94,6 +94,7 @@ or test(/^shortint::.*_param${multi_bit}_message_2_carry_3${multi_bit:+"_group_[
 or test(/^shortint::.*_param${multi_bit}_message_3_carry_1${multi_bit:+"_group_[0-9]"}(_compact_pk)?_ks_pbs/) \
 or test(/^shortint::.*_param${multi_bit}_message_3_carry_2${multi_bit:+"_group_[0-9]"}(_compact_pk)?_ks_pbs/) \
 or test(/^shortint::.*_param${multi_bit}_message_3_carry_3${multi_bit:+"_group_[0-9]"}(_compact_pk)?_ks_pbs/) \
+or test(/^shortint::.*_ci_run_filter/) \
 )\
 and not test(~smart_add_and_mul)""" # This test is too slow
     else
@@ -159,6 +160,7 @@ or test(/^shortint::.*_param${multi_bit}_message_3_carry_1${multi_bit:+"_group_[
 or test(/^shortint::.*_param${multi_bit}_message_3_carry_2${multi_bit:+"_group_[0-9]"}(_compact_pk)?_ks_pbs/) \
 or test(/^shortint::.*_param${multi_bit}_message_3_carry_3${multi_bit:+"_group_[0-9]"}(_compact_pk)?_ks_pbs/) \
 or test(/^shortint::.*_param${multi_bit}_message_4_carry_4${multi_bit:+"_group_[0-9]"}(_compact_pk)?_ks_pbs/) \
+or test(/^shortint::.*_ci_run_filter/) \
 )\
 and not test(~smart_add_and_mul)""" # This test is too slow
     else

tasks/Cargo.toml

@@ -6,7 +6,7 @@ edition = "2021"
 # See more keys and their definitions at https://doc.rust-lang.org/cargo/reference/manifest.html
 
 [dependencies]
-clap = "3.1"
+clap = "=4.4.4"
 lazy_static = "1.4"
 log = "0.4"
 simplelog = "0.12"

tfhe/.gitignore

@@ -1 +1 @@
-build/
+build/

tfhe/Cargo.toml

@@ -1,6 +1,6 @@
 [package]
 name = "tfhe"
-version = "0.4.0"
+version = "0.5.0"
 edition = "2021"
 readme = "../README.md"
 keywords = ["fully", "homomorphic", "encryption", "fhe", "cryptography"]
@@ -17,7 +17,7 @@ exclude = [
     "/js_on_wasm_tests/",
     "/web_wasm_parallel_tests/",
 ]
-rust-version = "1.67"
+rust-version = "1.72"
 
 # See more keys and their definitions at https://doc.rust-lang.org/cargo/reference/manifest.html
 
@@ -26,29 +26,30 @@ rand = "0.8.5"
 rand_distr = "0.4.3"
 paste = "1.0.7"
 lazy_static = { version = "1.4.0" }
-criterion = "0.4.0"
+criterion = "0.5.1"
 doc-comment = "0.3.3"
 serde_json = "1.0.94"
 # clap has to be pinned as its minimum supported rust version
 # changes often between minor releases, which breaks our CI
-clap = { version = "=4.2.7", features = ["derive"] }
+clap = { version = "=4.4.4", features = ["derive"] }
 # Used in user documentation
 bincode = "1.3.3"
 fs2 = { version = "0.4.3" }
-itertools = "0.10.5"
-num_cpus = "1.15"
+itertools = "0.11.0"
 # For erf and normality test
 libm = "0.2.6"
+# Begin regex-engine deps
 test-case = "3.1.0"
 combine = "4.6.6"
 env_logger = "0.10.0"
 log = "0.4.19"
+# End regex-engine deps
 
 [build-dependencies]
-cbindgen = { version = "0.24.3", optional = true }
+cbindgen = { version = "0.26.0", optional = true }
 
 [dependencies]
-concrete-csprng = { version = "0.3.0", features = [
+concrete-csprng = { version = "0.4.0", path = "../concrete-csprng", features = [
     "generator_fallback",
     "parallel",
 ] }
@@ -56,15 +57,14 @@ lazy_static = { version = "1.4.0", optional = true }
 serde = { version = "1.0", features = ["derive"] }
 rayon = { version = "1.5.0" }
 bincode = { version = "1.3.3", optional = true }
-concrete-fft = { version = "0.2.1", features = ["serde", "fft128"] }
-pulp = "0.11"
+concrete-fft = { version = "0.3.0", features = ["serde", "fft128"] }
+pulp = "0.13"
 aligned-vec = { version = "0.5", features = ["serde"] }
 dyn-stack = { version = "0.9" }
-once_cell = "1.13"
-paste = "1.0.7"
+paste = { version = "1.0.7", optional = true }
 fs2 = { version = "0.4.3", optional = true }
 # While we wait for repeat_n in rust standard library
-itertools = "0.10.5"
+itertools = "0.11.0"
 
 # wasm deps
 wasm-bindgen = { version = "0.2.86", features = [
@@ -73,40 +73,45 @@ wasm-bindgen = { version = "0.2.86", features = [
 wasm-bindgen-rayon = { version = "1.0", optional = true }
 js-sys = { version = "0.3", optional = true }
 console_error_panic_hook = { version = "0.1.7", optional = true }
-serde-wasm-bindgen = { version = "0.4", optional = true }
+serde-wasm-bindgen = { version = "0.6.0", optional = true }
 getrandom = { version = "0.2.8", optional = true }
 bytemuck = "1.13.1"
 
 [features]
-boolean = []
-shortint = []
-integer = ["shortint"]
-internal-keycache = ["lazy_static", "fs2", "bincode"]
+# paste is used by the HL API
+boolean = ["dep:paste"]
+shortint = ["dep:paste"]
+integer = ["shortint", "dep:paste"]
+internal-keycache = ["dep:lazy_static", "dep:fs2", "dep:bincode", "dep:paste"]
+safe-deserialization = ["dep:bincode"]
 
 # Experimental section
 experimental = []
 experimental-force_fft_algo_dif4 = []
 # End experimental section
 
-__c_api = ["cbindgen", "bincode"]
+__c_api = ["dep:cbindgen", "dep:bincode", "dep:paste"]
+# For the semver trick to skip the build.rs
+__force_skip_cbindgen = []
 boolean-c-api = ["boolean", "__c_api"]
 shortint-c-api = ["shortint", "__c_api"]
 high-level-c-api = ["boolean-c-api", "shortint-c-api", "integer", "__c_api"]
 
 __wasm_api = [
-    "wasm-bindgen",
-    "js-sys",
-    "console_error_panic_hook",
-    "serde-wasm-bindgen",
-    "getrandom",
+    "dep:wasm-bindgen",
+    "dep:js-sys",
+    "dep:console_error_panic_hook",
+    "dep:serde-wasm-bindgen",
+    "dep:getrandom",
     "getrandom/js",
-    "bincode",
+    "dep:bincode",
+    "safe-deserialization",
 ]
 boolean-client-js-wasm-api = ["boolean", "__wasm_api"]
 shortint-client-js-wasm-api = ["shortint", "__wasm_api"]
 integer-client-js-wasm-api = ["integer", "__wasm_api"]
 high-level-client-js-wasm-api = ["boolean", "shortint", "integer", "__wasm_api"]
-parallel-wasm-api = ["wasm-bindgen-rayon"]
+parallel-wasm-api = ["dep:wasm-bindgen-rayon"]
 
 nightly-avx512 = ["concrete-fft/nightly", "pulp/nightly"]
 
@@ -120,6 +125,7 @@ generator_aarch64_aes = ["concrete-csprng/generator_aarch64_aes"]
 
 # Private features
 __profiling = []
+__coverage = []
 
 seeder_unix = ["concrete-csprng/seeder_unix"]
 seeder_x86_64_rdseed = ["concrete-csprng/seeder_x86_64_rdseed"]
@@ -145,6 +151,12 @@ rustdoc-args = ["--html-in-header", "katex-header.html"]
 #         #
 ###########
 
+[[bench]]
+name = "ks-bench"
+path = "benches/core_crypto/ks_bench.rs"
+harness = false
+required-features = ["shortint", "internal-keycache"]
+
 [[bench]]
 name = "pbs-bench"
 path = "benches/core_crypto/pbs_bench.rs"
@@ -180,6 +192,12 @@ path = "benches/integer/bench.rs"
 harness = false
 required-features = ["integer", "internal-keycache"]
 
+[[bench]]
+name = "integer-signed-bench"
+path = "benches/integer/signed_bench.rs"
+harness = false
+required-features = ["integer", "internal-keycache"]
+
 [[bench]]
 name = "keygen"
 path = "benches/keygen/bench.rs"
@@ -202,7 +220,7 @@ required-features = ["shortint", "internal-keycache"]
 [[example]]
 name = "generates_test_keys"
 path = "examples/utilities/generates_test_keys.rs"
-required-features = ["shortint", "internal-keycache"]
+required-features = ["boolean", "shortint", "internal-keycache"]
 
 [[example]]
 name = "boolean_key_sizes"

tfhe/benches/boolean/bench.rs

@@ -43,7 +43,7 @@ pub fn write_to_json_boolean<T: Into<CryptoParametersRecord<u32>>>(
 
 // Put all `bench_function` in one place
 // so the keygen is only run once per parameters saving time.
-fn benchs(c: &mut Criterion, params: BooleanParameters, parameter_name: &str) {
+fn benches(c: &mut Criterion, params: BooleanParameters, parameter_name: &str) {
     let mut bench_group = c.benchmark_group("gates_benches");
 
     let cks = ClientKey::new(&params);
@@ -83,15 +83,15 @@ fn benchs(c: &mut Criterion, params: BooleanParameters, parameter_name: &str) {
 }
 
 fn bench_default_parameters(c: &mut Criterion) {
-    benchs(c, DEFAULT_PARAMETERS, "DEFAULT_PARAMETERS");
+    benches(c, DEFAULT_PARAMETERS, "DEFAULT_PARAMETERS");
 }
 
 fn bench_default_parameters_ks_pbs(c: &mut Criterion) {
-    benchs(c, DEFAULT_PARAMETERS_KS_PBS, "DEFAULT_PARAMETERS_KS_PBS");
+    benches(c, DEFAULT_PARAMETERS_KS_PBS, "DEFAULT_PARAMETERS_KS_PBS");
 }
 
 fn bench_low_prob_parameters(c: &mut Criterion) {
-    benchs(
+    benches(
         c,
         PARAMETERS_ERROR_PROB_2_POW_MINUS_165,
         "PARAMETERS_ERROR_PROB_2_POW_MINUS_165",
@@ -99,7 +99,7 @@ fn bench_low_prob_parameters(c: &mut Criterion) {
 }
 
 fn bench_low_prob_parameters_ks_pbs(c: &mut Criterion) {
-    benchs(
+    benches(
         c,
         PARAMETERS_ERROR_PROB_2_POW_MINUS_165_KS_PBS,
         "PARAMETERS_ERROR_PROB_2_POW_MINUS_165_KS_PBS",
@@ -107,5 +107,5 @@ fn bench_low_prob_parameters_ks_pbs(c: &mut Criterion) {
 }
 
 fn bench_tfhe_lib_parameters(c: &mut Criterion) {
-    benchs(c, TFHE_LIB_PARAMETERS, " TFHE_LIB_PARAMETERS");
+    benches(c, TFHE_LIB_PARAMETERS, " TFHE_LIB_PARAMETERS");
 }

tfhe/benches/core_crypto/dev_bench.rs

@@ -130,7 +130,7 @@ fn multi_bit_pbs<Scalar: UnsignedTorus + CastInto<usize> + CastFrom<usize> + Syn
     );
 
     let id = format!("Multi Bit PBS {}", Scalar::BITS);
-    #[allow(clippy::unit_arg)]
+
     {
         c.bench_function(&id, |b| {
             b.iter(|| {

tfhe/benches/core_crypto/ks_bench.rs

@@ -0,0 +1,91 @@
+use criterion::{criterion_group, criterion_main, Criterion};
+use tfhe::core_crypto::prelude::*;
+use tfhe::keycache::NamedParam;
+use tfhe::shortint::prelude::*;
+
+fn criterion_bench(criterion: &mut Criterion) {
+    type Scalar = u64;
+
+    let mut bench_group = criterion.benchmark_group("KS");
+    bench_group
+        .sample_size(15)
+        .measurement_time(std::time::Duration::from_secs(60));
+
+    for params in [
+        PARAM_MESSAGE_1_CARRY_1_KS_PBS,
+        PARAM_MESSAGE_2_CARRY_2_KS_PBS,
+        PARAM_MESSAGE_3_CARRY_3_KS_PBS,
+        PARAM_MESSAGE_4_CARRY_4_KS_PBS,
+    ]
+    .into_iter()
+    {
+        let lwe_dimension = params.lwe_dimension;
+        let lwe_modular_std_dev = params.lwe_modular_std_dev;
+        let ciphertext_modulus = params.ciphertext_modulus;
+        let encoding_with_padding = if ciphertext_modulus.is_native_modulus() {
+            Scalar::ONE << (Scalar::BITS - 1)
+        } else {
+            Scalar::cast_from(ciphertext_modulus.get_custom_modulus() / 2)
+        };
+        let glwe_dimension = params.glwe_dimension;
+        let polynomial_size = params.polynomial_size;
+        let ks_decomp_base_log = params.ks_base_log;
+        let ks_decomp_level_count = params.ks_level;
+        let msg_modulus: Scalar = params.message_modulus.0.cast_into();
+        let total_modulus: Scalar = (params.message_modulus.0 * params.carry_modulus.0).cast_into();
+
+        let msg = msg_modulus - 1;
+        let delta: Scalar = encoding_with_padding / total_modulus;
+
+        // Create the PRNG
+        let mut seeder = new_seeder();
+        let seeder = seeder.as_mut();
+        let mut encryption_generator =
+            EncryptionRandomGenerator::<ActivatedRandomGenerator>::new(seeder.seed(), seeder);
+        let mut secret_generator =
+            SecretRandomGenerator::<ActivatedRandomGenerator>::new(seeder.seed());
+
+        let lwe_sk =
+            allocate_and_generate_new_binary_lwe_secret_key(lwe_dimension, &mut secret_generator);
+
+        let glwe_sk = allocate_and_generate_new_binary_glwe_secret_key(
+            glwe_dimension,
+            polynomial_size,
+            &mut secret_generator,
+        );
+        let big_lwe_sk = glwe_sk.into_lwe_secret_key();
+        let ksk_big_to_small = allocate_and_generate_new_lwe_keyswitch_key(
+            &big_lwe_sk,
+            &lwe_sk,
+            ks_decomp_base_log,
+            ks_decomp_level_count,
+            lwe_modular_std_dev,
+            ciphertext_modulus,
+            &mut encryption_generator,
+        );
+
+        let plaintext = Plaintext(msg * delta);
+        let ct = allocate_and_encrypt_new_lwe_ciphertext(
+            &big_lwe_sk,
+            plaintext,
+            lwe_modular_std_dev,
+            ciphertext_modulus,
+            &mut encryption_generator,
+        );
+
+        let mut output_ct = LweCiphertext::new(
+            Scalar::ZERO,
+            lwe_sk.lwe_dimension().to_lwe_size(),
+            ciphertext_modulus,
+        );
+
+        bench_group.bench_function(&params.name(), |bencher| {
+            bencher.iter(|| {
+                keyswitch_lwe_ciphertext(&ksk_big_to_small, &ct, &mut output_ct);
+            })
+        });
+    }
+}
+
+criterion_group!(benches, criterion_bench);
+criterion_main!(benches);

tfhe/benches/core_crypto/pbs_bench.rs

@@ -66,8 +66,8 @@ criterion_group!(
 
 criterion_main!(pbs_group, multi_bit_pbs_group, pbs_throughput_group);
 
-fn benchmark_parameters<Scalar: UnsignedInteger>(
-) -> Vec<(&'static str, CryptoParametersRecord<Scalar>)> {
+fn benchmark_parameters<Scalar: UnsignedInteger>() -> Vec<(String, CryptoParametersRecord<Scalar>)>
+{
     if Scalar::BITS == 64 {
         SHORTINT_BENCH_PARAMS
             .iter()
@@ -83,7 +83,7 @@ fn benchmark_parameters<Scalar: UnsignedInteger>(
     } else if Scalar::BITS == 32 {
         BOOLEAN_BENCH_PARAMS
             .iter()
-            .map(|(name, params)| (*name, params.to_owned().into()))
+            .map(|(name, params)| (name.to_string(), params.to_owned().into()))
             .collect()
     } else {
         vec![]
@@ -91,7 +91,7 @@ fn benchmark_parameters<Scalar: UnsignedInteger>(
 }
 
 fn throughput_benchmark_parameters<Scalar: UnsignedInteger>(
-) -> Vec<(&'static str, CryptoParametersRecord<Scalar>)> {
+) -> Vec<(String, CryptoParametersRecord<Scalar>)> {
     if Scalar::BITS == 64 {
         vec![
             PARAM_MESSAGE_1_CARRY_1_KS_PBS,
@@ -111,18 +111,15 @@ fn throughput_benchmark_parameters<Scalar: UnsignedInteger>(
     } else if Scalar::BITS == 32 {
         BOOLEAN_BENCH_PARAMS
             .iter()
-            .map(|(name, params)| (*name, params.to_owned().into()))
+            .map(|(name, params)| (name.to_string(), params.to_owned().into()))
             .collect()
     } else {
         vec![]
     }
 }
 
-fn multi_bit_benchmark_parameters<Scalar: UnsignedInteger + Default>() -> Vec<(
-    &'static str,
-    CryptoParametersRecord<Scalar>,
-    LweBskGroupingFactor,
-)> {
+fn multi_bit_benchmark_parameters<Scalar: UnsignedInteger + Default>(
+) -> Vec<(String, CryptoParametersRecord<Scalar>, LweBskGroupingFactor)> {
     if Scalar::BITS == 64 {
         vec![
             PARAM_MULTI_BIT_MESSAGE_1_CARRY_1_GROUP_2_KS_PBS,
@@ -243,7 +240,7 @@ fn mem_optimized_pbs<Scalar: UnsignedTorus + CastInto<usize> + Serialize>(c: &mu
         write_to_json(
             &id,
             *params,
-            *name,
+            name,
             "pbs",
             &OperatorType::Atomic,
             bit_size,
@@ -332,7 +329,7 @@ fn multi_bit_pbs<
         write_to_json(
             &id,
             *params,
-            *name,
+            name,
             "pbs",
             &OperatorType::Atomic,
             bit_size,
@@ -421,7 +418,7 @@ fn multi_bit_deterministic_pbs<
         write_to_json(
             &id,
             *params,
-            *name,
+            name,
             "pbs",
             &OperatorType::Atomic,
             bit_size,
@@ -541,7 +538,7 @@ fn pbs_throughput<Scalar: UnsignedTorus + CastInto<usize> + Sync + Send + Serial
             write_to_json(
                 &id,
                 *params,
-                *name,
+                name,
                 "pbs",
                 &OperatorType::Atomic,
                 bit_size,