diff --git a/.noir-sync-commit b/.noir-sync-commit index 9bbde85e56b..b6e1166fe48 100644 --- a/.noir-sync-commit +++ b/.noir-sync-commit @@ -1 +1 @@ -68c32b4ffd9b069fe4b119327dbf4018c17ab9d4 +31640e91ba75b9c5200ea66d1f54cc5185e0d196 diff --git a/noir/noir-repo/.github/workflows/formatting.yml b/noir/noir-repo/.github/workflows/formatting.yml index 08c02af519f..ab92d452c79 100644 --- a/noir/noir-repo/.github/workflows/formatting.yml +++ b/noir/noir-repo/.github/workflows/formatting.yml @@ -15,18 +15,11 @@ concurrency: jobs: clippy: name: cargo clippy - runs-on: ${{ matrix.runner }} + runs-on: ubuntu-latest timeout-minutes: 30 env: RUSTFLAGS: -Dwarnings - strategy: - fail-fast: false - matrix: - include: - - runner: ubuntu-latest - target: x86_64-unknown-linux-gnu - steps: - name: Checkout uses: actions/checkout@v4 @@ -34,18 +27,41 @@ jobs: - name: Setup toolchain uses: dtolnay/rust-toolchain@1.74.1 with: - targets: ${{ matrix.target }} + targets: x86_64-unknown-linux-gnu components: clippy, rustfmt - uses: Swatinem/rust-cache@v2 with: - key: ${{ matrix.target }} + key: x86_64-unknown-linux-gnu cache-on-failure: true save-if: ${{ github.event_name != 'merge_group' }} - name: Run `cargo clippy` run: cargo clippy --all-targets --workspace --locked --release + rustfmt: + name: cargo fmt + runs-on: ubuntu-latest + timeout-minutes: 30 + env: + RUSTFLAGS: -Dwarnings + + steps: + - name: Checkout + uses: actions/checkout@v4 + + - name: Setup toolchain + uses: dtolnay/rust-toolchain@1.74.1 + with: + targets: x86_64-unknown-linux-gnu + components: clippy, rustfmt + + - uses: Swatinem/rust-cache@v2 + with: + key: x86_64-unknown-linux-gnu + cache-on-failure: true + save-if: ${{ github.event_name != 'merge_group' }} + - name: Run `cargo fmt` run: cargo fmt --all --check @@ -88,7 +104,6 @@ jobs: run: | mkdir dist cp ./target/release/nargo ./dist/nargo - 7z a -ttar -so -an ./dist/* | 7z a -si ./nargo-x86_64-unknown-linux-gnu.tar.gz - name: Upload artifact uses: actions/upload-artifact@v4 diff --git a/noir/noir-repo/.github/workflows/gates_report.yml b/noir/noir-repo/.github/workflows/gates_report.yml deleted file mode 100644 index 0b0a527b69e..00000000000 --- a/noir/noir-repo/.github/workflows/gates_report.yml +++ /dev/null @@ -1,94 +0,0 @@ -name: Report gates diff - -on: - push: - branches: - - master - pull_request: - -jobs: - build-nargo: - runs-on: ubuntu-latest - strategy: - matrix: - target: [x86_64-unknown-linux-gnu] - - steps: - - name: Checkout Noir repo - uses: actions/checkout@v4 - - - name: Setup toolchain - uses: dtolnay/rust-toolchain@1.74.1 - - - uses: Swatinem/rust-cache@v2 - with: - key: ${{ matrix.target }} - cache-on-failure: true - save-if: ${{ github.event_name != 'merge_group' }} - - - name: Build Nargo - run: cargo build --package nargo_cli --release - - - name: Package artifacts - run: | - mkdir dist - cp ./target/release/nargo ./dist/nargo - 7z a -ttar -so -an ./dist/* | 7z a -si ./nargo-x86_64-unknown-linux-gnu.tar.gz - - - name: Upload artifact - uses: actions/upload-artifact@v4 - with: - name: nargo - path: ./dist/* - retention-days: 3 - - - compare_gates_reports: - needs: [build-nargo] - runs-on: ubuntu-latest - permissions: - pull-requests: write - - steps: - - uses: actions/checkout@v4 - - - name: Install `bb` - run: | - ./scripts/install_bb.sh - echo "$HOME/.bb/" >> $GITHUB_PATH - - - name: Download nargo binary - uses: actions/download-artifact@v4 - with: - name: nargo - path: ./nargo - - - name: Set nargo on PATH - run: | - nargo_binary="${{ github.workspace }}/nargo/nargo" - chmod +x $nargo_binary - echo "$(dirname $nargo_binary)" >> $GITHUB_PATH - export PATH="$PATH:$(dirname $nargo_binary)" - nargo -V - - - name: Generate gates report - working-directory: ./test_programs - run: | - ./rebuild.sh - ./gates_report.sh - mv gates_report.json ../gates_report.json - - - name: Compare gates reports - id: gates_diff - uses: noir-lang/noir-gates-diff@1931aaaa848a1a009363d6115293f7b7fc72bb87 - with: - report: gates_report.json - summaryQuantile: 0.9 # only display the 10% most significant circuit size diffs in the summary (defaults to 20%) - - - name: Add gates diff to sticky comment - if: github.event_name == 'pull_request' || github.event_name == 'pull_request_target' - uses: marocchino/sticky-pull-request-comment@v2 - with: - # delete the comment in case changes no longer impact circuit sizes - delete: ${{ !steps.gates_diff.outputs.markdown }} - message: ${{ steps.gates_diff.outputs.markdown }} diff --git a/noir/noir-repo/.github/workflows/gates_report_brillig.yml b/noir/noir-repo/.github/workflows/gates_report_brillig.yml deleted file mode 100644 index e7ec30923f0..00000000000 --- a/noir/noir-repo/.github/workflows/gates_report_brillig.yml +++ /dev/null @@ -1,92 +0,0 @@ -name: Report Brillig bytecode size diff - -on: - push: - branches: - - master - pull_request: - -jobs: - build-nargo: - runs-on: ubuntu-latest - strategy: - matrix: - target: [x86_64-unknown-linux-gnu] - - steps: - - name: Checkout Noir repo - uses: actions/checkout@v4 - - - name: Setup toolchain - uses: dtolnay/rust-toolchain@1.74.1 - - - uses: Swatinem/rust-cache@v2 - with: - key: ${{ matrix.target }} - cache-on-failure: true - save-if: ${{ github.event_name != 'merge_group' }} - - - name: Build Nargo - run: cargo build --package nargo_cli --release - - - name: Package artifacts - run: | - mkdir dist - cp ./target/release/nargo ./dist/nargo - 7z a -ttar -so -an ./dist/* | 7z a -si ./nargo-x86_64-unknown-linux-gnu.tar.gz - - - name: Upload artifact - uses: actions/upload-artifact@v4 - with: - name: nargo - path: ./dist/* - retention-days: 3 - - compare_brillig_bytecode_size_reports: - needs: [build-nargo] - runs-on: ubuntu-latest - permissions: - pull-requests: write - - steps: - - uses: actions/checkout@v4 - - - name: Download nargo binary - uses: actions/download-artifact@v4 - with: - name: nargo - path: ./nargo - - - name: Set nargo on PATH - run: | - nargo_binary="${{ github.workspace }}/nargo/nargo" - chmod +x $nargo_binary - echo "$(dirname $nargo_binary)" >> $GITHUB_PATH - export PATH="$PATH:$(dirname $nargo_binary)" - nargo -V - - - name: Generate Brillig bytecode size report - working-directory: ./test_programs - run: | - chmod +x gates_report_brillig.sh - ./gates_report_brillig.sh - mv gates_report_brillig.json ../gates_report_brillig.json - - - name: Compare Brillig bytecode size reports - id: brillig_bytecode_diff - uses: noir-lang/noir-gates-diff@d88f7523b013b9edd3f31c5cfddaef87a3fe1b48 - with: - report: gates_report_brillig.json - header: | - # Changes to Brillig bytecode sizes - brillig_report: true - summaryQuantile: 0.9 # only display the 10% most significant bytecode size diffs in the summary (defaults to 20%) - - - name: Add bytecode size diff to sticky comment - if: github.event_name == 'pull_request' || github.event_name == 'pull_request_target' - uses: marocchino/sticky-pull-request-comment@v2 - with: - header: brillig - # delete the comment in case changes no longer impact brillig bytecode sizes - delete: ${{ !steps.brillig_bytecode_diff.outputs.markdown }} - message: ${{ steps.brillig_bytecode_diff.outputs.markdown }} \ No newline at end of file diff --git a/noir/noir-repo/.github/workflows/gates_report_brillig_execution.yml b/noir/noir-repo/.github/workflows/gates_report_brillig_execution.yml deleted file mode 100644 index 0ef98f5045b..00000000000 --- a/noir/noir-repo/.github/workflows/gates_report_brillig_execution.yml +++ /dev/null @@ -1,92 +0,0 @@ -name: Report Brillig opcodes executed diff - -on: - push: - branches: - - master - pull_request: - -jobs: - build-nargo: - runs-on: ubuntu-latest - strategy: - matrix: - target: [x86_64-unknown-linux-gnu] - - steps: - - name: Checkout Noir repo - uses: actions/checkout@v4 - - - name: Setup toolchain - uses: dtolnay/rust-toolchain@1.74.1 - - - uses: Swatinem/rust-cache@v2 - with: - key: ${{ matrix.target }} - cache-on-failure: true - save-if: ${{ github.event_name != 'merge_group' }} - - - name: Build Nargo - run: cargo build --package nargo_cli --release - - - name: Package artifacts - run: | - mkdir dist - cp ./target/release/nargo ./dist/nargo - 7z a -ttar -so -an ./dist/* | 7z a -si ./nargo-x86_64-unknown-linux-gnu.tar.gz - - - name: Upload artifact - uses: actions/upload-artifact@v4 - with: - name: nargo - path: ./dist/* - retention-days: 3 - - compare_brillig_execution_reports: - needs: [build-nargo] - runs-on: ubuntu-latest - permissions: - pull-requests: write - - steps: - - uses: actions/checkout@v4 - - - name: Download nargo binary - uses: actions/download-artifact@v4 - with: - name: nargo - path: ./nargo - - - name: Set nargo on PATH - run: | - nargo_binary="${{ github.workspace }}/nargo/nargo" - chmod +x $nargo_binary - echo "$(dirname $nargo_binary)" >> $GITHUB_PATH - export PATH="$PATH:$(dirname $nargo_binary)" - nargo -V - - - name: Generate Brillig execution report - working-directory: ./test_programs - run: | - chmod +x gates_report_brillig_execution.sh - ./gates_report_brillig_execution.sh - mv gates_report_brillig_execution.json ../gates_report_brillig_execution.json - - - name: Compare Brillig execution reports - id: brillig_execution_diff - uses: noir-lang/noir-gates-diff@d88f7523b013b9edd3f31c5cfddaef87a3fe1b48 - with: - report: gates_report_brillig_execution.json - header: | - # Changes to number of Brillig opcodes executed - brillig_report: true - summaryQuantile: 0.9 # only display the 10% most significant bytecode size diffs in the summary (defaults to 20%) - - - name: Add bytecode size diff to sticky comment - if: github.event_name == 'pull_request' || github.event_name == 'pull_request_target' - uses: marocchino/sticky-pull-request-comment@v2 - with: - header: brillig_execution - # delete the comment in case changes no longer impact brillig bytecode sizes - delete: ${{ !steps.brillig_execution_diff.outputs.markdown }} - message: ${{ steps.brillig_execution_diff.outputs.markdown }} \ No newline at end of file diff --git a/noir/noir-repo/.github/workflows/lockfile.yml b/noir/noir-repo/.github/workflows/lockfile.yml deleted file mode 100644 index 190e01745af..00000000000 --- a/noir/noir-repo/.github/workflows/lockfile.yml +++ /dev/null @@ -1,23 +0,0 @@ -name: Lockfile check - -on: - pull_request: - -# This will cancel previous runs when a branch or PR is updated -concurrency: - group: ${{ github.workflow }}-${{ github.head_ref || github.ref || github.run_id }} - cancel-in-progress: true - -jobs: - yarn-lock: - runs-on: ubuntu-latest - timeout-minutes: 30 - - steps: - - name: Checkout - uses: actions/checkout@v4 - - # Errors if installation would result in modifications to yarn.lock - - name: Install - run: yarn --immutable - shell: bash diff --git a/noir/noir-repo/.github/workflows/release.yml b/noir/noir-repo/.github/workflows/release.yml index 7e0909224e5..59c3d9a1415 100644 --- a/noir/noir-repo/.github/workflows/release.yml +++ b/noir/noir-repo/.github/workflows/release.yml @@ -15,7 +15,7 @@ jobs: steps: - name: Run release-please id: release - uses: google-github-actions/release-please-action@v4 + uses: googleapis/release-please-action@v4 with: token: ${{ secrets.NOIR_RELEASES_TOKEN }} diff --git a/noir/noir-repo/.github/workflows/reports.yml b/noir/noir-repo/.github/workflows/reports.yml new file mode 100644 index 00000000000..8f8aeabb65e --- /dev/null +++ b/noir/noir-repo/.github/workflows/reports.yml @@ -0,0 +1,235 @@ +name: Reports + +on: + push: + branches: + - master + pull_request: + +jobs: + build-nargo: + runs-on: ubuntu-latest + + steps: + - name: Checkout Noir repo + uses: actions/checkout@v4 + + - name: Setup toolchain + uses: dtolnay/rust-toolchain@1.74.1 + + - uses: Swatinem/rust-cache@v2 + with: + key: x86_64-unknown-linux-gnu + cache-on-failure: true + save-if: ${{ github.event_name != 'merge_group' }} + + - name: Build Nargo + run: cargo build --package nargo_cli --release + + - name: Package artifacts + run: | + mkdir dist + cp ./target/release/nargo ./dist/nargo + + - name: Upload artifact + uses: actions/upload-artifact@v4 + with: + name: nargo + path: ./dist/* + retention-days: 3 + + + compare_gates_reports: + name: Circuit sizes + needs: [build-nargo] + runs-on: ubuntu-latest + permissions: + pull-requests: write + + steps: + - uses: actions/checkout@v4 + + - name: Install `bb` + run: | + ./scripts/install_bb.sh + echo "$HOME/.bb/" >> $GITHUB_PATH + + - name: Download nargo binary + uses: actions/download-artifact@v4 + with: + name: nargo + path: ./nargo + + - name: Set nargo on PATH + run: | + nargo_binary="${{ github.workspace }}/nargo/nargo" + chmod +x $nargo_binary + echo "$(dirname $nargo_binary)" >> $GITHUB_PATH + export PATH="$PATH:$(dirname $nargo_binary)" + nargo -V + + - name: Generate gates report + working-directory: ./test_programs + run: | + ./rebuild.sh + ./gates_report.sh + mv gates_report.json ../gates_report.json + + - name: Compare gates reports + id: gates_diff + uses: noir-lang/noir-gates-diff@1931aaaa848a1a009363d6115293f7b7fc72bb87 + with: + report: gates_report.json + summaryQuantile: 0.9 # only display the 10% most significant circuit size diffs in the summary (defaults to 20%) + + - name: Add gates diff to sticky comment + if: github.event_name == 'pull_request' || github.event_name == 'pull_request_target' + uses: marocchino/sticky-pull-request-comment@v2 + with: + # delete the comment in case changes no longer impact circuit sizes + delete: ${{ !steps.gates_diff.outputs.markdown }} + message: ${{ steps.gates_diff.outputs.markdown }} + + compare_brillig_bytecode_size_reports: + name: Brillig bytecode sizes + needs: [build-nargo] + runs-on: ubuntu-latest + permissions: + pull-requests: write + + steps: + - uses: actions/checkout@v4 + + - name: Download nargo binary + uses: actions/download-artifact@v4 + with: + name: nargo + path: ./nargo + + - name: Set nargo on PATH + run: | + nargo_binary="${{ github.workspace }}/nargo/nargo" + chmod +x $nargo_binary + echo "$(dirname $nargo_binary)" >> $GITHUB_PATH + export PATH="$PATH:$(dirname $nargo_binary)" + nargo -V + + - name: Generate Brillig bytecode size report + working-directory: ./test_programs + run: | + ./gates_report_brillig.sh + mv gates_report_brillig.json ../gates_report_brillig.json + + - name: Compare Brillig bytecode size reports + id: brillig_bytecode_diff + uses: noir-lang/noir-gates-diff@d88f7523b013b9edd3f31c5cfddaef87a3fe1b48 + with: + report: gates_report_brillig.json + header: | + # Changes to Brillig bytecode sizes + brillig_report: true + summaryQuantile: 0.9 # only display the 10% most significant bytecode size diffs in the summary (defaults to 20%) + + - name: Add bytecode size diff to sticky comment + if: github.event_name == 'pull_request' || github.event_name == 'pull_request_target' + uses: marocchino/sticky-pull-request-comment@v2 + with: + header: brillig + # delete the comment in case changes no longer impact brillig bytecode sizes + delete: ${{ !steps.brillig_bytecode_diff.outputs.markdown }} + message: ${{ steps.brillig_bytecode_diff.outputs.markdown }} + + compare_brillig_execution_reports: + name: Brillig execution trace sizes + needs: [build-nargo] + runs-on: ubuntu-latest + permissions: + pull-requests: write + + steps: + - uses: actions/checkout@v4 + + - name: Download nargo binary + uses: actions/download-artifact@v4 + with: + name: nargo + path: ./nargo + + - name: Set nargo on PATH + run: | + nargo_binary="${{ github.workspace }}/nargo/nargo" + chmod +x $nargo_binary + echo "$(dirname $nargo_binary)" >> $GITHUB_PATH + export PATH="$PATH:$(dirname $nargo_binary)" + nargo -V + + - name: Generate Brillig execution report + working-directory: ./test_programs + run: | + ./gates_report_brillig_execution.sh + mv gates_report_brillig_execution.json ../gates_report_brillig_execution.json + + - name: Compare Brillig execution reports + id: brillig_execution_diff + uses: noir-lang/noir-gates-diff@d88f7523b013b9edd3f31c5cfddaef87a3fe1b48 + with: + report: gates_report_brillig_execution.json + header: | + # Changes to number of Brillig opcodes executed + brillig_report: true + summaryQuantile: 0.9 # only display the 10% most significant bytecode size diffs in the summary (defaults to 20%) + + - name: Add bytecode size diff to sticky comment + if: github.event_name == 'pull_request' || github.event_name == 'pull_request_target' + uses: marocchino/sticky-pull-request-comment@v2 + with: + header: brillig_execution + # delete the comment in case changes no longer impact brillig bytecode sizes + delete: ${{ !steps.brillig_execution_diff.outputs.markdown }} + message: ${{ steps.brillig_execution_diff.outputs.markdown }} + + generate_memory_report: + name: Peak memory usage + needs: [build-nargo] + runs-on: ubuntu-latest + permissions: + pull-requests: write + + steps: + - uses: actions/checkout@v4 + + - name: Download nargo binary + uses: actions/download-artifact@v4 + with: + name: nargo + path: ./nargo + + - name: Set nargo on PATH + run: | + nargo_binary="${{ github.workspace }}/nargo/nargo" + chmod +x $nargo_binary + echo "$(dirname $nargo_binary)" >> $GITHUB_PATH + export PATH="$PATH:$(dirname $nargo_binary)" + nargo -V + + - name: Generate Memory report + working-directory: ./test_programs + run: | + ./memory_report.sh + mv memory_report.json ../memory_report.json + + - name: Parse memory report + id: memory_report + uses: noir-lang/noir-bench-report@ccb0d806a91d3bd86dba0ba3d580a814eed5673c + with: + report: memory_report.json + header: | + # Memory Report + memory_report: true + + - name: Add memory report to sticky comment + if: github.event_name == 'pull_request' || github.event_name == 'pull_request_target' + uses: marocchino/sticky-pull-request-comment@v2 + with: + header: memory + message: ${{ steps.memory_report.outputs.markdown }} diff --git a/noir/noir-repo/.github/workflows/test-js-packages.yml b/noir/noir-repo/.github/workflows/test-js-packages.yml index 4a5d0b8179b..6a9a918b955 100644 --- a/noir/noir-repo/.github/workflows/test-js-packages.yml +++ b/noir/noir-repo/.github/workflows/test-js-packages.yml @@ -13,6 +13,19 @@ concurrency: cancel-in-progress: true jobs: + yarn-lock: + runs-on: ubuntu-latest + timeout-minutes: 30 + + steps: + - name: Checkout + uses: actions/checkout@v4 + + # Errors if installation would result in modifications to yarn.lock + - name: Install + run: yarn --immutable + shell: bash + build-nargo: runs-on: ubuntu-22.04 timeout-minutes: 30 @@ -78,7 +91,6 @@ jobs: ./tooling/noirc_abi_wasm/web retention-days: 10 - build-noir-wasm: runs-on: ubuntu-latest timeout-minutes: 30 @@ -519,12 +531,25 @@ jobs: fail-fast: false matrix: project: - # Disabled as these are currently failing with many visibility errors - - { repo: AztecProtocol/aztec-nr, path: ./ } + - { repo: noir-lang/ec, path: ./ } + - { repo: noir-lang/eddsa, path: ./ } + - { repo: noir-lang/mimc, path: ./ } + - { repo: noir-lang/noir_sort, path: ./ } + - { repo: noir-lang/noir-edwards, path: ./ } + - { repo: noir-lang/noir-bignum, path: ./ } + - { repo: noir-lang/noir_bigcurve, path: ./ } + - { repo: noir-lang/noir_base64, path: ./ } + - { repo: noir-lang/noir_string_search, path: ./ } + - { repo: noir-lang/sparse_array, path: ./ } + - { repo: noir-lang/noir_rsa, path: ./lib } + - { repo: AztecProtocol/aztec-packages, path: ./noir-projects/aztec-nr } - { repo: AztecProtocol/aztec-packages, path: ./noir-projects/noir-contracts } - # Disabled as aztec-packages requires a setup-step in order to generate a `Nargo.toml` - #- { repo: AztecProtocol/aztec-packages, path: ./noir-projects/noir-protocol-circuits } - - { repo: noir-lang/noir-edwards, path: ./, ref: 3188ea74fe3b059219a2ea87899589c266256d74 } + - { repo: AztecProtocol/aztec-packages, path: ./noir-projects/noir-protocol-circuits/crates/parity-lib } + - { repo: AztecProtocol/aztec-packages, path: ./noir-projects/noir-protocol-circuits/crates/private-kernel-lib } + - { repo: AztecProtocol/aztec-packages, path: ./noir-projects/noir-protocol-circuits/crates/reset-kernel-lib } + - { repo: AztecProtocol/aztec-packages, path: ./noir-projects/noir-protocol-circuits/crates/rollup-lib } + - { repo: AztecProtocol/aztec-packages, path: ./noir-projects/noir-protocol-circuits/crates/types } + name: Check external repo - ${{ matrix.project.repo }} steps: - name: Checkout @@ -554,9 +579,12 @@ jobs: # Github actions seems to not expand "**" in globs by default. shopt -s globstar sed -i '/^compiler_version/d' ./**/Nargo.toml - - name: Run nargo check + + - name: Run nargo test working-directory: ./test-repo/${{ matrix.project.path }} - run: nargo check + run: nargo test --silence-warnings + env: + NARGO_IGNORE_TEST_FAILURES_FROM_FOREIGN_CALLS: true # This is a job which depends on all test jobs and reports the overall status. # This allows us to add/remove test jobs without having to update the required workflows. @@ -566,6 +594,7 @@ jobs: # We want this job to always run (even if the dependant jobs fail) as we want this job to fail rather than skipping. if: ${{ always() }} needs: + - yarn-lock - test-acvm_js-node - test-acvm_js-browser - test-noirc-abi diff --git a/noir/noir-repo/Cargo.lock b/noir/noir-repo/Cargo.lock index 94a84b89d05..96ceb94fcdd 100644 --- a/noir/noir-repo/Cargo.lock +++ b/noir/noir-repo/Cargo.lock @@ -13,7 +13,7 @@ dependencies = [ "criterion", "flate2", "fxhash", - "pprof 0.13.0", + "pprof", "serde", "serde-big-array", "serde-generate", @@ -158,6 +158,15 @@ dependencies = [ "memchr", ] +[[package]] +name = "aligned-vec" +version = "0.6.1" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "7e0966165eaf052580bd70eb1b32cb3d6245774c0104d1b2793e9650bf83b52a" +dependencies = [ + "equator", +] + [[package]] name = "android-tzdata" version = "0.1.1" @@ -604,7 +613,7 @@ dependencies = [ "lazy_static", "noir_grumpkin", "num-bigint", - "pprof 0.12.1", + "pprof", ] [[package]] @@ -1417,6 +1426,26 @@ dependencies = [ "log", ] +[[package]] +name = "equator" +version = "0.2.2" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "c35da53b5a021d2484a7cc49b2ac7f2d840f8236a286f84202369bd338d761ea" +dependencies = [ + "equator-macro", +] + +[[package]] +name = "equator-macro" +version = "0.2.1" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "3bf679796c0322556351f287a51b49e48f7c4986e727b5dd78c972d30e2e16cc" +dependencies = [ + "proc-macro2", + "quote", + "syn 2.0.87", +] + [[package]] name = "equivalent" version = "1.0.1" @@ -2789,6 +2818,7 @@ dependencies = [ "dirs", "file-lock", "fm", + "fxhash", "iai", "iter-extended", "lazy_static", @@ -2806,7 +2836,7 @@ dependencies = [ "notify", "notify-debouncer-full", "paste", - "pprof 0.13.0", + "pprof", "predicates 2.1.5", "prettytable-rs", "proptest", @@ -3151,6 +3181,7 @@ dependencies = [ "serde_json", "serde_with", "similar-asserts", + "test-case", "thiserror", "tracing", ] @@ -3577,32 +3608,11 @@ checksum = "439ee305def115ba05938db6eb1644ff94165c5ab5e9420d1c1bcedbba909391" [[package]] name = "pprof" -version = "0.12.1" -source = "registry+https://github.com/rust-lang/crates.io-index" -checksum = "978385d59daf9269189d052ca8a84c1acfd0715c0599a5d5188d4acc078ca46a" -dependencies = [ - "backtrace", - "cfg-if 1.0.0", - "criterion", - "findshlibs", - "inferno", - "libc", - "log", - "nix 0.26.4", - "once_cell", - "parking_lot 0.12.3", - "smallvec", - "symbolic-demangle", - "tempfile", - "thiserror", -] - -[[package]] -name = "pprof" -version = "0.13.0" +version = "0.14.0" source = "registry+https://github.com/rust-lang/crates.io-index" -checksum = "ef5c97c51bd34c7e742402e216abdeb44d415fbe6ae41d56b114723e953711cb" +checksum = "ebbe2f8898beba44815fdc9e5a4ae9c929e21c5dc29b0c774a15555f7f58d6d0" dependencies = [ + "aligned-vec", "backtrace", "cfg-if 1.0.0", "criterion", diff --git a/noir/noir-repo/Cargo.toml b/noir/noir-repo/Cargo.toml index 94ebe54fde1..4ce0ddd999f 100644 --- a/noir/noir-repo/Cargo.toml +++ b/noir/noir-repo/Cargo.toml @@ -126,7 +126,7 @@ codespan-reporting = "0.11.1" criterion = "0.5.0" # Note that using the "frame-pointer" feature breaks framegraphs on linux # https://github.com/tikv/pprof-rs/pull/172 -pprof = { version = "0.13", features = ["flamegraph", "criterion"] } +pprof = { version = "0.14", features = ["flamegraph", "criterion"] } cfg-if = "1.0.0" dirs = "4" diff --git a/noir/noir-repo/acvm-repo/acir/src/circuit/brillig.rs b/noir/noir-repo/acvm-repo/acir/src/circuit/brillig.rs index a9714ce29b2..ef75d088f8c 100644 --- a/noir/noir-repo/acvm-repo/acir/src/circuit/brillig.rs +++ b/noir/noir-repo/acvm-repo/acir/src/circuit/brillig.rs @@ -5,7 +5,7 @@ use serde::{Deserialize, Serialize}; /// Inputs for the Brillig VM. These are the initial inputs /// that the Brillig VM will use to start. -#[derive(Clone, PartialEq, Eq, Serialize, Deserialize, Debug)] +#[derive(Clone, PartialEq, Eq, Serialize, Deserialize, Debug, Hash)] pub enum BrilligInputs { Single(Expression), Array(Vec>), @@ -14,7 +14,7 @@ pub enum BrilligInputs { /// Outputs for the Brillig VM. Once the VM has completed /// execution, this will be the object that is returned. -#[derive(Clone, PartialEq, Eq, Serialize, Deserialize, Debug)] +#[derive(Clone, PartialEq, Eq, Serialize, Deserialize, Debug, Hash)] pub enum BrilligOutputs { Simple(Witness), Array(Vec), @@ -23,7 +23,7 @@ pub enum BrilligOutputs { /// This is purely a wrapper struct around a list of Brillig opcode's which represents /// a full Brillig function to be executed by the Brillig VM. /// This is stored separately on a program and accessed through a [BrilligPointer]. -#[derive(Clone, PartialEq, Eq, Serialize, Deserialize, Default, Debug)] +#[derive(Clone, PartialEq, Eq, Serialize, Deserialize, Default, Debug, Hash)] pub struct BrilligBytecode { pub bytecode: Vec>, } diff --git a/noir/noir-repo/acvm-repo/acir/src/circuit/mod.rs b/noir/noir-repo/acvm-repo/acir/src/circuit/mod.rs index 6282a33af6b..4ff581bf17a 100644 --- a/noir/noir-repo/acvm-repo/acir/src/circuit/mod.rs +++ b/noir/noir-repo/acvm-repo/acir/src/circuit/mod.rs @@ -25,7 +25,7 @@ use self::{brillig::BrilligBytecode, opcodes::BlockId}; /// Bounded Expressions are useful if you are eventually going to pass the ACIR /// into a proving system which supports PLONK, where arithmetic expressions have a /// finite fan-in. -#[derive(Debug, Clone, Copy, PartialEq, Eq, Serialize, Deserialize, Default)] +#[derive(Debug, Clone, Copy, PartialEq, Eq, Serialize, Deserialize, Default, Hash)] pub enum ExpressionWidth { #[default] Unbounded, @@ -36,13 +36,13 @@ pub enum ExpressionWidth { /// A program represented by multiple ACIR circuits. The execution trace of these /// circuits is dictated by construction of the [crate::native_types::WitnessStack]. -#[derive(Clone, PartialEq, Eq, Serialize, Deserialize, Default)] +#[derive(Clone, PartialEq, Eq, Serialize, Deserialize, Default, Hash)] pub struct Program { pub functions: Vec>, pub unconstrained_functions: Vec>, } -#[derive(Clone, PartialEq, Eq, Serialize, Deserialize, Default)] +#[derive(Clone, PartialEq, Eq, Serialize, Deserialize, Default, Hash)] pub struct Circuit { // current_witness_index is the highest witness index in the circuit. The next witness to be added to this circuit // will take on this value. (The value is cached here as an optimization.) @@ -69,13 +69,13 @@ pub struct Circuit { pub assert_messages: Vec<(OpcodeLocation, AssertionPayload)>, } -#[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize)] +#[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize, Hash)] pub enum ExpressionOrMemory { Expression(Expression), Memory(BlockId), } -#[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize)] +#[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize, Hash)] pub struct AssertionPayload { pub error_selector: u64, pub payload: Vec>, @@ -355,7 +355,7 @@ impl std::fmt::Debug for Program { } } -#[derive(Clone, Debug, PartialEq, Eq, Serialize, Deserialize, Default)] +#[derive(Clone, Debug, PartialEq, Eq, Serialize, Deserialize, Default, Hash)] pub struct PublicInputs(pub BTreeSet); impl PublicInputs { diff --git a/noir/noir-repo/acvm-repo/acir/src/circuit/opcodes.rs b/noir/noir-repo/acvm-repo/acir/src/circuit/opcodes.rs index 06effd3c5b6..f47c40b0dd7 100644 --- a/noir/noir-repo/acvm-repo/acir/src/circuit/opcodes.rs +++ b/noir/noir-repo/acvm-repo/acir/src/circuit/opcodes.rs @@ -15,7 +15,7 @@ pub use black_box_function_call::{ }; pub use memory_operation::{BlockId, MemOp}; -#[derive(Clone, PartialEq, Eq, Serialize, Deserialize)] +#[derive(Clone, PartialEq, Eq, Serialize, Deserialize, Hash)] pub enum BlockType { Memory, CallData(u32), @@ -29,7 +29,7 @@ impl BlockType { } #[allow(clippy::large_enum_variant)] -#[derive(Clone, PartialEq, Eq, Serialize, Deserialize)] +#[derive(Clone, PartialEq, Eq, Serialize, Deserialize, Hash)] pub enum Opcode { /// An `AssertZero` opcode adds the constraint that `P(w) = 0`, where /// `w=(w_1,..w_n)` is a tuple of `n` witnesses, and `P` is a multi-variate diff --git a/noir/noir-repo/acvm-repo/acir/src/circuit/opcodes/black_box_function_call.rs b/noir/noir-repo/acvm-repo/acir/src/circuit/opcodes/black_box_function_call.rs index dfdf9616306..9cf31e94eb4 100644 --- a/noir/noir-repo/acvm-repo/acir/src/circuit/opcodes/black_box_function_call.rs +++ b/noir/noir-repo/acvm-repo/acir/src/circuit/opcodes/black_box_function_call.rs @@ -9,13 +9,13 @@ use thiserror::Error; // Note: Some functions will not use all of the witness // So we need to supply how many bits of the witness is needed -#[derive(Clone, Copy, Debug, PartialEq, Eq, Serialize, Deserialize)] +#[derive(Clone, Copy, Debug, PartialEq, Eq, Serialize, Deserialize, Hash)] pub enum ConstantOrWitnessEnum { Constant(F), Witness(Witness), } -#[derive(Clone, Copy, Debug, PartialEq, Eq, Serialize, Deserialize)] +#[derive(Clone, Copy, Debug, PartialEq, Eq, Serialize, Deserialize, Hash)] pub struct FunctionInput { input: ConstantOrWitnessEnum, num_bits: u32, @@ -79,7 +79,7 @@ impl std::fmt::Display for FunctionInput { } } -#[derive(Clone, PartialEq, Eq, Serialize, Deserialize)] +#[derive(Clone, PartialEq, Eq, Serialize, Deserialize, Hash)] pub enum BlackBoxFuncCall { AES128Encrypt { inputs: Vec>, diff --git a/noir/noir-repo/acvm-repo/acir/src/circuit/opcodes/memory_operation.rs b/noir/noir-repo/acvm-repo/acir/src/circuit/opcodes/memory_operation.rs index 90e3ee0563a..c9a78983204 100644 --- a/noir/noir-repo/acvm-repo/acir/src/circuit/opcodes/memory_operation.rs +++ b/noir/noir-repo/acvm-repo/acir/src/circuit/opcodes/memory_operation.rs @@ -7,7 +7,7 @@ pub struct BlockId(pub u32); /// Operation on a block of memory /// We can either write or read at an index in memory -#[derive(Clone, PartialEq, Eq, Serialize, Deserialize, Debug)] +#[derive(Clone, PartialEq, Eq, Serialize, Deserialize, Debug, Hash)] pub struct MemOp { /// A constant expression that can be 0 (read) or 1 (write) pub operation: Expression, diff --git a/noir/noir-repo/acvm-repo/bn254_blackbox_solver/Cargo.toml b/noir/noir-repo/acvm-repo/bn254_blackbox_solver/Cargo.toml index 8829692b9b4..825a0ef0481 100644 --- a/noir/noir-repo/acvm-repo/bn254_blackbox_solver/Cargo.toml +++ b/noir/noir-repo/acvm-repo/bn254_blackbox_solver/Cargo.toml @@ -30,7 +30,7 @@ num-bigint.workspace = true [dev-dependencies] ark-std.workspace = true criterion = "0.5.0" -pprof = { version = "0.12", features = [ +pprof = { version = "0.14", features = [ "flamegraph", "frame-pointer", "criterion", diff --git a/noir/noir-repo/acvm-repo/brillig/src/black_box.rs b/noir/noir-repo/acvm-repo/brillig/src/black_box.rs index 9cc5349e45b..f185b36e6c8 100644 --- a/noir/noir-repo/acvm-repo/brillig/src/black_box.rs +++ b/noir/noir-repo/acvm-repo/brillig/src/black_box.rs @@ -3,7 +3,7 @@ use serde::{Deserialize, Serialize}; /// These opcodes provide an equivalent of ACIR blackbox functions. /// They are implemented as native functions in the VM. -#[derive(Debug, Clone, Copy, PartialEq, Eq, Serialize, Deserialize)] +#[derive(Debug, Clone, Copy, PartialEq, Eq, Serialize, Deserialize, Hash)] pub enum BlackBoxOp { /// Encrypts a message using AES128. AES128Encrypt { diff --git a/noir/noir-repo/acvm-repo/brillig/src/opcodes.rs b/noir/noir-repo/acvm-repo/brillig/src/opcodes.rs index 8b72b5a9b41..1cb31ca3d0a 100644 --- a/noir/noir-repo/acvm-repo/brillig/src/opcodes.rs +++ b/noir/noir-repo/acvm-repo/brillig/src/opcodes.rs @@ -56,7 +56,7 @@ impl MemoryAddress { } /// Describes the memory layout for an array/vector element -#[derive(Debug, Clone, Eq, PartialEq, Serialize, Deserialize)] +#[derive(Debug, Clone, Eq, PartialEq, Serialize, Deserialize, Hash)] pub enum HeapValueType { // A single field element is enough to represent the value with a given bit size Simple(BitSize), @@ -81,7 +81,7 @@ impl HeapValueType { } /// A fixed-sized array starting from a Brillig memory location. -#[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize, Copy)] +#[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize, Copy, Hash)] pub struct HeapArray { pub pointer: MemoryAddress, pub size: usize, @@ -94,13 +94,13 @@ impl Default for HeapArray { } /// A memory-sized vector passed starting from a Brillig memory location and with a memory-held size -#[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize, Copy)] +#[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize, Copy, Hash)] pub struct HeapVector { pub pointer: MemoryAddress, pub size: MemoryAddress, } -#[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize, Copy, PartialOrd, Ord)] +#[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize, Copy, PartialOrd, Ord, Hash)] pub enum IntegerBitSize { U1, U8, @@ -152,7 +152,7 @@ impl std::fmt::Display for IntegerBitSize { } } -#[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize, Copy, PartialOrd, Ord)] +#[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize, Copy, PartialOrd, Ord, Hash)] pub enum BitSize { Field, Integer(IntegerBitSize), @@ -181,7 +181,7 @@ impl BitSize { /// While we are usually agnostic to how memory is passed within Brillig, /// this needs to be encoded somehow when dealing with an external system. /// For simplicity, the extra type information is given right in the ForeignCall instructions. -#[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize, Copy)] +#[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize, Copy, Hash)] pub enum ValueOrArray { /// A single value passed to or from an external call /// It is an 'immediate' value - used without dereferencing. @@ -198,7 +198,7 @@ pub enum ValueOrArray { HeapVector(HeapVector), } -#[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize)] +#[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize, Hash)] pub enum BrilligOpcode { /// Takes the fields in addresses `lhs` and `rhs` /// Performs the specified binary operation @@ -314,7 +314,7 @@ pub enum BrilligOpcode { } /// Binary fixed-length field expressions -#[derive(Debug, Clone, Copy, PartialEq, Eq, Serialize, Deserialize)] +#[derive(Debug, Clone, Copy, PartialEq, Eq, Serialize, Deserialize, Hash)] pub enum BinaryFieldOp { Add, Sub, @@ -332,7 +332,7 @@ pub enum BinaryFieldOp { } /// Binary fixed-length integer expressions -#[derive(Debug, Clone, Copy, PartialEq, Eq, Serialize, Deserialize)] +#[derive(Debug, Clone, Copy, PartialEq, Eq, Serialize, Deserialize, Hash)] pub enum BinaryIntOp { Add, Sub, diff --git a/noir/noir-repo/compiler/noirc_driver/src/debug.rs b/noir/noir-repo/compiler/noirc_driver/src/debug.rs index f5eaede89b2..6044e6c0e65 100644 --- a/noir/noir-repo/compiler/noirc_driver/src/debug.rs +++ b/noir/noir-repo/compiler/noirc_driver/src/debug.rs @@ -8,7 +8,7 @@ use std::{ /// For a given file, we store the source code and the path to the file /// so consumers of the debug artifact can reconstruct the original source code structure. -#[derive(Clone, Debug, Serialize, Deserialize)] +#[derive(Clone, Debug, Serialize, Deserialize, Hash)] pub struct DebugFile { pub source: String, pub path: PathBuf, diff --git a/noir/noir-repo/compiler/noirc_driver/src/lib.rs b/noir/noir-repo/compiler/noirc_driver/src/lib.rs index 72ea464805f..5bedefaf563 100644 --- a/noir/noir-repo/compiler/noirc_driver/src/lib.rs +++ b/noir/noir-repo/compiler/noirc_driver/src/lib.rs @@ -13,7 +13,7 @@ use noirc_abi::{AbiParameter, AbiType, AbiValue}; use noirc_errors::{CustomDiagnostic, FileDiagnostic}; use noirc_evaluator::create_program; use noirc_evaluator::errors::RuntimeError; -use noirc_evaluator::ssa::SsaProgramArtifact; +use noirc_evaluator::ssa::{SsaLogging, SsaProgramArtifact}; use noirc_frontend::debug::build_debug_crate_file; use noirc_frontend::hir::def_map::{Contract, CrateDefMap}; use noirc_frontend::hir::Context; @@ -70,6 +70,11 @@ pub struct CompileOptions { #[arg(long, hide = true)] pub show_ssa: bool, + /// Only show SSA passes whose name contains the provided string. + /// This setting takes precedence over `show_ssa` if it's not empty. + #[arg(long, hide = true)] + pub show_ssa_pass_name: Option, + /// Emit the unoptimized SSA IR to file. /// The IR will be dumped into the workspace target directory, /// under `[compiled-package].ssa.json`. @@ -126,11 +131,19 @@ pub struct CompileOptions { #[arg(long)] pub skip_underconstrained_check: bool, - /// Setting to decide on an inlining strategy for brillig functions. + /// Setting to decide on an inlining strategy for Brillig functions. /// A more aggressive inliner should generate larger programs but more optimized /// A less aggressive inliner should generate smaller programs #[arg(long, hide = true, allow_hyphen_values = true, default_value_t = i64::MAX)] pub inliner_aggressiveness: i64, + + /// Setting the maximum acceptable increase in Brillig bytecode size due to + /// unrolling small loops. When left empty, any change is accepted as long + /// as it required fewer SSA instructions. + /// A higher value results in fewer jumps but a larger program. + /// A lower value keeps the original program if it was smaller, even if it has more jumps. + #[arg(long, hide = true, allow_hyphen_values = true)] + pub max_bytecode_increase_percent: Option, } pub fn parse_expression_width(input: &str) -> Result { @@ -321,6 +334,8 @@ pub fn compute_function_abi( /// /// On success this returns the compiled program alongside any warnings that were found. /// On error this returns the non-empty list of warnings and errors. +/// +/// See [compile_no_check] for further information about the use of `cached_program`. pub fn compile_main( context: &mut Context, crate_id: CrateId, @@ -542,6 +557,15 @@ pub const DEFAULT_EXPRESSION_WIDTH: ExpressionWidth = ExpressionWidth::Bounded { /// Compile the current crate using `main_function` as the entrypoint. /// /// This function assumes [`check_crate`] is called beforehand. +/// +/// If the program is not returned from cache, it is backend-agnostic and must go through a transformation +/// pass before usage in proof generation; if it's returned from cache these transformations might have +/// already been applied. +/// +/// The transformations are _not_ covered by the check that decides whether we can use the cached artifact. +/// That comparison is based on on [CompiledProgram::hash] which is a persisted version of the hash of the input +/// [`ast::Program`][noirc_frontend::monomorphization::ast::Program], whereas the output [`circuit::Program`][acir::circuit::Program] +/// contains the final optimized ACIR opcodes, including the transformation done after this compilation. #[tracing::instrument(level = "trace", skip_all, fields(function_name = context.function_name(&main_function)))] pub fn compile_no_check( context: &mut Context, @@ -556,8 +580,6 @@ pub fn compile_no_check( monomorphize(main_function, &mut context.def_interner)? }; - let hash = fxhash::hash64(&program); - let hashes_match = cached_program.as_ref().map_or(false, |program| program.hash == hash); if options.show_monomorphized { println!("{program}"); } @@ -571,13 +593,28 @@ pub fn compile_no_check( || options.show_ssa || options.emit_ssa; - if !force_compile && hashes_match { - info!("Program matches existing artifact, returning early"); - return Ok(cached_program.expect("cache must exist for hashes to match")); + // Hash the AST program, which is going to be used to fingerprint the compilation artifact. + let hash = fxhash::hash64(&program); + + if let Some(cached_program) = cached_program { + if !force_compile && cached_program.hash == hash { + info!("Program matches existing artifact, returning early"); + return Ok(cached_program); + } } + let return_visibility = program.return_visibility; let ssa_evaluator_options = noirc_evaluator::ssa::SsaEvaluatorOptions { - enable_ssa_logging: options.show_ssa, + ssa_logging: match &options.show_ssa_pass_name { + Some(string) => SsaLogging::Contains(string.clone()), + None => { + if options.show_ssa { + SsaLogging::All + } else { + SsaLogging::None + } + } + }, enable_brillig_logging: options.show_brillig, force_brillig_output: options.force_brillig, print_codegen_timings: options.benchmark_codegen, @@ -589,6 +626,7 @@ pub fn compile_no_check( emit_ssa: if options.emit_ssa { Some(context.package_build_path.clone()) } else { None }, skip_underconstrained_check: options.skip_underconstrained_check, inliner_aggressiveness: options.inliner_aggressiveness, + max_bytecode_increase_percent: options.max_bytecode_increase_percent, }; let SsaProgramArtifact { program, debug, warnings, names, brillig_names, error_types, .. } = diff --git a/noir/noir-repo/compiler/noirc_driver/src/program.rs b/noir/noir-repo/compiler/noirc_driver/src/program.rs index 88460482928..4b4d6662e8e 100644 --- a/noir/noir-repo/compiler/noirc_driver/src/program.rs +++ b/noir/noir-repo/compiler/noirc_driver/src/program.rs @@ -9,7 +9,7 @@ use serde::{Deserialize, Serialize}; use super::debug::DebugFile; -#[derive(Debug, Serialize, Deserialize, Clone)] +#[derive(Debug, Serialize, Deserialize, Clone, Hash)] pub struct CompiledProgram { pub noir_version: String, /// Hash of the [`Program`][noirc_frontend::monomorphization::ast::Program] from which this [`CompiledProgram`] diff --git a/noir/noir-repo/compiler/noirc_errors/src/debug_info.rs b/noir/noir-repo/compiler/noirc_errors/src/debug_info.rs index 77028f739bd..a5e12b37712 100644 --- a/noir/noir-repo/compiler/noirc_errors/src/debug_info.rs +++ b/noir/noir-repo/compiler/noirc_errors/src/debug_info.rs @@ -94,7 +94,7 @@ impl ProgramDebugInfo { } #[serde_as] -#[derive(Default, Debug, Clone, Deserialize, Serialize)] +#[derive(Default, Debug, Clone, Deserialize, Serialize, Hash)] pub struct DebugInfo { /// Map opcode index of an ACIR circuit into the source code location /// Serde does not support mapping keys being enums for json, so we indicate diff --git a/noir/noir-repo/compiler/noirc_evaluator/Cargo.toml b/noir/noir-repo/compiler/noirc_evaluator/Cargo.toml index e25b5bf855a..bb8c62cfd95 100644 --- a/noir/noir-repo/compiler/noirc_evaluator/Cargo.toml +++ b/noir/noir-repo/compiler/noirc_evaluator/Cargo.toml @@ -33,6 +33,7 @@ cfg-if.workspace = true proptest.workspace = true similar-asserts.workspace = true num-traits.workspace = true +test-case.workspace = true [features] bn254 = ["noirc_frontend/bn254"] diff --git a/noir/noir-repo/compiler/noirc_evaluator/src/acir/acir_variable.rs b/noir/noir-repo/compiler/noirc_evaluator/src/acir/acir_variable.rs index a42426e6c04..9f2c649ee3e 100644 --- a/noir/noir-repo/compiler/noirc_evaluator/src/acir/acir_variable.rs +++ b/noir/noir-repo/compiler/noirc_evaluator/src/acir/acir_variable.rs @@ -92,7 +92,7 @@ impl<'a> From<&'a SsaType> for AcirType { SsaType::Numeric(numeric_type) => AcirType::NumericType(*numeric_type), SsaType::Array(elements, size) => { let elements = elements.iter().map(|e| e.into()).collect(); - AcirType::Array(elements, *size) + AcirType::Array(elements, *size as usize) } _ => unreachable!("The type {value} cannot be represented in ACIR"), } diff --git a/noir/noir-repo/compiler/noirc_evaluator/src/acir/mod.rs b/noir/noir-repo/compiler/noirc_evaluator/src/acir/mod.rs index 69679495b92..76f0dea95bb 100644 --- a/noir/noir-repo/compiler/noirc_evaluator/src/acir/mod.rs +++ b/noir/noir-repo/compiler/noirc_evaluator/src/acir/mod.rs @@ -571,7 +571,7 @@ impl<'a> Context<'a> { AcirValue::Array(_) => { let block_id = self.block_id(param_id); let len = if matches!(typ, Type::Array(_, _)) { - typ.flattened_size() + typ.flattened_size() as usize } else { return Err(InternalError::Unexpected { expected: "Block params should be an array".to_owned(), @@ -816,7 +816,9 @@ impl<'a> Context<'a> { let inputs = vecmap(arguments, |arg| self.convert_value(*arg, dfg)); let output_count = result_ids .iter() - .map(|result_id| dfg.type_of_value(*result_id).flattened_size()) + .map(|result_id| { + dfg.type_of_value(*result_id).flattened_size() as usize + }) .sum(); let Some(acir_function_id) = @@ -948,7 +950,7 @@ impl<'a> Context<'a> { let block_id = self.block_id(&array_id); let array_typ = dfg.type_of_value(array_id); let len = if matches!(array_typ, Type::Array(_, _)) { - array_typ.flattened_size() + array_typ.flattened_size() as usize } else { Self::flattened_value_size(&output) }; @@ -1444,7 +1446,7 @@ impl<'a> Context<'a> { // a separate SSA value and restrictions on slice indices should be generated elsewhere in the SSA. let array_typ = dfg.type_of_value(array); let array_len = if !array_typ.contains_slice_element() { - array_typ.flattened_size() + array_typ.flattened_size() as usize } else { self.flattened_slice_size(array, dfg) }; @@ -1539,7 +1541,7 @@ impl<'a> Context<'a> { let value = self.convert_value(array, dfg); let array_typ = dfg.type_of_value(array); let len = if !array_typ.contains_slice_element() { - array_typ.flattened_size() + array_typ.flattened_size() as usize } else { self.flattened_slice_size(array, dfg) }; @@ -1810,7 +1812,7 @@ impl<'a> Context<'a> { return_values .iter() - .fold(0, |acc, value_id| acc + dfg.type_of_value(*value_id).flattened_size()) + .fold(0, |acc, value_id| acc + dfg.type_of_value(*value_id).flattened_size() as usize) } /// Converts an SSA terminator's return values into their ACIR representations @@ -2156,7 +2158,7 @@ impl<'a> Context<'a> { let inputs = vecmap(&arguments_no_slice_len, |arg| self.convert_value(*arg, dfg)); let output_count = result_ids.iter().fold(0usize, |sum, result_id| { - sum + dfg.try_get_array_length(*result_id).unwrap_or(1) + sum + dfg.try_get_array_length(*result_id).unwrap_or(1) as usize }); let vars = self.acir_context.black_box_function(black_box, inputs, output_count)?; @@ -2180,7 +2182,7 @@ impl<'a> Context<'a> { endian, field, radix, - array_length as u32, + array_length, result_type[0].clone().into(), ) .map(|array| vec![array]) @@ -2194,12 +2196,7 @@ impl<'a> Context<'a> { }; self.acir_context - .bit_decompose( - endian, - field, - array_length as u32, - result_type[0].clone().into(), - ) + .bit_decompose(endian, field, array_length, result_type[0].clone().into()) .map(|array| vec![array]) } Intrinsic::ArrayLen => { @@ -2220,7 +2217,7 @@ impl<'a> Context<'a> { let acir_value = self.convert_value(slice_contents, dfg); let array_len = if !slice_typ.contains_slice_element() { - slice_typ.flattened_size() + slice_typ.flattened_size() as usize } else { self.flattened_slice_size(slice_contents, dfg) }; diff --git a/noir/noir-repo/compiler/noirc_evaluator/src/brillig/brillig_gen/brillig_block.rs b/noir/noir-repo/compiler/noirc_evaluator/src/brillig/brillig_gen/brillig_block.rs index 1fa4985295a..9c88c559b59 100644 --- a/noir/noir-repo/compiler/noirc_evaluator/src/brillig/brillig_gen/brillig_block.rs +++ b/noir/noir-repo/compiler/noirc_evaluator/src/brillig/brillig_gen/brillig_block.rs @@ -1823,7 +1823,7 @@ impl<'block> BrilligBlock<'block> { Type::Array(_, nested_size) => { let inner_array = BrilligArray { pointer: self.brillig_context.allocate_register(), - size: *nested_size, + size: *nested_size as usize, }; self.allocate_foreign_call_result_array(element_type, inner_array); diff --git a/noir/noir-repo/compiler/noirc_evaluator/src/brillig/brillig_gen/brillig_block_variables.rs b/noir/noir-repo/compiler/noirc_evaluator/src/brillig/brillig_gen/brillig_block_variables.rs index 393d4c967c2..bf0a1bc7347 100644 --- a/noir/noir-repo/compiler/noirc_evaluator/src/brillig/brillig_gen/brillig_block_variables.rs +++ b/noir/noir-repo/compiler/noirc_evaluator/src/brillig/brillig_gen/brillig_block_variables.rs @@ -142,7 +142,7 @@ pub(crate) fn allocate_value( } Type::Array(item_typ, elem_count) => BrilligVariable::BrilligArray(BrilligArray { pointer: brillig_context.allocate_register(), - size: compute_array_length(&item_typ, elem_count), + size: compute_array_length(&item_typ, elem_count as usize), }), Type::Slice(_) => BrilligVariable::BrilligVector(BrilligVector { pointer: brillig_context.allocate_register(), diff --git a/noir/noir-repo/compiler/noirc_evaluator/src/brillig/brillig_gen/brillig_fn.rs b/noir/noir-repo/compiler/noirc_evaluator/src/brillig/brillig_gen/brillig_fn.rs index 2779be103cd..3dea7b3e7f5 100644 --- a/noir/noir-repo/compiler/noirc_evaluator/src/brillig/brillig_gen/brillig_fn.rs +++ b/noir/noir-repo/compiler/noirc_evaluator/src/brillig/brillig_gen/brillig_fn.rs @@ -59,7 +59,7 @@ impl FunctionContext { vecmap(item_type.iter(), |item_typ| { FunctionContext::ssa_type_to_parameter(item_typ) }), - *size, + *size as usize, ), Type::Slice(_) => { panic!("ICE: Slice parameters cannot be derived from type information") diff --git a/noir/noir-repo/compiler/noirc_evaluator/src/brillig/brillig_ir/brillig_variable.rs b/noir/noir-repo/compiler/noirc_evaluator/src/brillig/brillig_ir/brillig_variable.rs index 81d61e05cc4..0bb18448670 100644 --- a/noir/noir-repo/compiler/noirc_evaluator/src/brillig/brillig_ir/brillig_variable.rs +++ b/noir/noir-repo/compiler/noirc_evaluator/src/brillig/brillig_ir/brillig_variable.rs @@ -88,7 +88,7 @@ pub(crate) fn type_to_heap_value_type(typ: &Type) -> HeapValueType { ), Type::Array(elem_type, size) => HeapValueType::Array { value_types: elem_type.as_ref().iter().map(type_to_heap_value_type).collect(), - size: typ.element_size() * size, + size: typ.element_size() * *size as usize, }, Type::Slice(elem_type) => HeapValueType::Vector { value_types: elem_type.as_ref().iter().map(type_to_heap_value_type).collect(), diff --git a/noir/noir-repo/compiler/noirc_evaluator/src/brillig/brillig_ir/procedures/array_copy.rs b/noir/noir-repo/compiler/noirc_evaluator/src/brillig/brillig_ir/procedures/array_copy.rs index 67f7cf2dc34..0a6e8824223 100644 --- a/noir/noir-repo/compiler/noirc_evaluator/src/brillig/brillig_ir/procedures/array_copy.rs +++ b/noir/noir-repo/compiler/noirc_evaluator/src/brillig/brillig_ir/procedures/array_copy.rs @@ -69,6 +69,8 @@ pub(super) fn compile_array_copy_procedure( BRILLIG_MEMORY_ADDRESSING_BIT_SIZE, 1_usize.into(), ); + // Decrease the original ref count now that this copy is no longer pointing to it + ctx.codegen_usize_op(rc.address, rc.address, BrilligBinaryOp::Sub, 1); } }); } diff --git a/noir/noir-repo/compiler/noirc_evaluator/src/brillig/mod.rs b/noir/noir-repo/compiler/noirc_evaluator/src/brillig/mod.rs index 1b61ae1a864..cb8c35cd8e0 100644 --- a/noir/noir-repo/compiler/noirc_evaluator/src/brillig/mod.rs +++ b/noir/noir-repo/compiler/noirc_evaluator/src/brillig/mod.rs @@ -12,7 +12,7 @@ use self::{ }, }; use crate::ssa::{ - ir::function::{Function, FunctionId, RuntimeType}, + ir::function::{Function, FunctionId}, ssa_gen::Ssa, }; use fxhash::FxHashMap as HashMap; @@ -59,7 +59,7 @@ impl std::ops::Index for Brillig { } impl Ssa { - /// Compile to brillig brillig functions and ACIR functions reachable from them + /// Compile Brillig functions and ACIR functions reachable from them #[tracing::instrument(level = "trace", skip_all)] pub(crate) fn to_brillig(&self, enable_debug_trace: bool) -> Brillig { // Collect all the function ids that are reachable from brillig @@ -67,9 +67,7 @@ impl Ssa { let brillig_reachable_function_ids = self .functions .iter() - .filter_map(|(id, func)| { - matches!(func.runtime(), RuntimeType::Brillig(_)).then_some(*id) - }) + .filter_map(|(id, func)| func.runtime().is_brillig().then_some(*id)) .collect::>(); let mut brillig = Brillig::default(); diff --git a/noir/noir-repo/compiler/noirc_evaluator/src/errors.rs b/noir/noir-repo/compiler/noirc_evaluator/src/errors.rs index 994e97eabb8..75a3ceb3a72 100644 --- a/noir/noir-repo/compiler/noirc_evaluator/src/errors.rs +++ b/noir/noir-repo/compiler/noirc_evaluator/src/errors.rs @@ -63,7 +63,7 @@ pub enum RuntimeError { UnknownReference { call_stack: CallStack }, } -#[derive(Debug, Clone, Serialize, Deserialize)] +#[derive(Debug, Clone, Serialize, Deserialize, Hash)] pub enum SsaReport { Warning(InternalWarning), Bug(InternalBug), @@ -107,7 +107,7 @@ impl From for FileDiagnostic { } } -#[derive(Debug, PartialEq, Eq, Clone, Error, Serialize, Deserialize)] +#[derive(Debug, PartialEq, Eq, Clone, Error, Serialize, Deserialize, Hash)] pub enum InternalWarning { #[error("Return variable contains a constant value")] ReturnConstant { call_stack: CallStack }, @@ -115,7 +115,7 @@ pub enum InternalWarning { VerifyProof { call_stack: CallStack }, } -#[derive(Debug, PartialEq, Eq, Clone, Error, Serialize, Deserialize)] +#[derive(Debug, PartialEq, Eq, Clone, Error, Serialize, Deserialize, Hash)] pub enum InternalBug { #[error("Input to brillig function is in a separate subgraph to output")] IndependentSubgraph { call_stack: CallStack }, diff --git a/noir/noir-repo/compiler/noirc_evaluator/src/ssa.rs b/noir/noir-repo/compiler/noirc_evaluator/src/ssa.rs index 97c1760d87c..8f31023f790 100644 --- a/noir/noir-repo/compiler/noirc_evaluator/src/ssa.rs +++ b/noir/noir-repo/compiler/noirc_evaluator/src/ssa.rs @@ -44,9 +44,16 @@ mod opt; pub(crate) mod parser; pub mod ssa_gen; +#[derive(Debug, Clone)] +pub enum SsaLogging { + None, + All, + Contains(String), +} + pub struct SsaEvaluatorOptions { /// Emit debug information for the intermediate SSA IR - pub enable_ssa_logging: bool, + pub ssa_logging: SsaLogging, pub enable_brillig_logging: bool, @@ -67,6 +74,11 @@ pub struct SsaEvaluatorOptions { /// The higher the value, the more inlined brillig functions will be. pub inliner_aggressiveness: i64, + + /// Maximum accepted percentage increase in the Brillig bytecode size after unrolling loops. + /// When `None` the size increase check is skipped altogether and any decrease in the SSA + /// instruction count is accepted. + pub max_bytecode_increase_percent: Option, } pub(crate) struct ArtifactsAndWarnings(Artifacts, Vec); @@ -85,46 +97,49 @@ pub(crate) fn optimize_into_acir( let mut ssa = SsaBuilder::new( program, - options.enable_ssa_logging, + options.ssa_logging.clone(), options.force_brillig_output, options.print_codegen_timings, &options.emit_ssa, )? - .run_pass(Ssa::defunctionalize, "After Defunctionalization:") - .run_pass(Ssa::remove_paired_rc, "After Removing Paired rc_inc & rc_decs:") - .run_pass(Ssa::separate_runtime, "After Runtime Separation:") - .run_pass(Ssa::resolve_is_unconstrained, "After Resolving IsUnconstrained:") - .run_pass(|ssa| ssa.inline_functions(options.inliner_aggressiveness), "After Inlining (1st):") + .run_pass(Ssa::defunctionalize, "Defunctionalization") + .run_pass(Ssa::remove_paired_rc, "Removing Paired rc_inc & rc_decs") + .run_pass(Ssa::separate_runtime, "Runtime Separation") + .run_pass(Ssa::resolve_is_unconstrained, "Resolving IsUnconstrained") + .run_pass(|ssa| ssa.inline_functions(options.inliner_aggressiveness), "Inlining (1st)") // Run mem2reg with the CFG separated into blocks - .run_pass(Ssa::mem2reg, "After Mem2Reg (1st):") - .run_pass(Ssa::simplify_cfg, "After Simplifying (1st):") - .run_pass(Ssa::as_slice_optimization, "After `as_slice` optimization") + .run_pass(Ssa::mem2reg, "Mem2Reg (1st)") + .run_pass(Ssa::simplify_cfg, "Simplifying (1st)") + .run_pass(Ssa::as_slice_optimization, "`as_slice` optimization") .try_run_pass( Ssa::evaluate_static_assert_and_assert_constant, - "After `static_assert` and `assert_constant`:", + "`static_assert` and `assert_constant`", + )? + .run_pass(Ssa::loop_invariant_code_motion, "Loop Invariant Code Motion") + .try_run_pass( + |ssa| ssa.unroll_loops_iteratively(options.max_bytecode_increase_percent), + "Unrolling", )? - .run_pass(Ssa::loop_invariant_code_motion, "After Loop Invariant Code Motion:") - .try_run_pass(Ssa::unroll_loops_iteratively, "After Unrolling:")? - .run_pass(Ssa::simplify_cfg, "After Simplifying (2nd):") - .run_pass(Ssa::flatten_cfg, "After Flattening:") - .run_pass(Ssa::remove_bit_shifts, "After Removing Bit Shifts:") + .run_pass(Ssa::simplify_cfg, "Simplifying (2nd)") + .run_pass(Ssa::flatten_cfg, "Flattening") + .run_pass(Ssa::remove_bit_shifts, "After Removing Bit Shifts") // Run mem2reg once more with the flattened CFG to catch any remaining loads/stores - .run_pass(Ssa::mem2reg, "After Mem2Reg (2nd):") + .run_pass(Ssa::mem2reg, "Mem2Reg (2nd)") // Run the inlining pass again to handle functions with `InlineType::NoPredicates`. // Before flattening is run, we treat functions marked with the `InlineType::NoPredicates` as an entry point. // This pass must come immediately following `mem2reg` as the succeeding passes // may create an SSA which inlining fails to handle. .run_pass( |ssa| ssa.inline_functions_with_no_predicates(options.inliner_aggressiveness), - "After Inlining (2nd):", + "Inlining (2nd)", ) - .run_pass(Ssa::remove_if_else, "After Remove IfElse:") - .run_pass(Ssa::fold_constants, "After Constant Folding:") - .run_pass(Ssa::remove_enable_side_effects, "After EnableSideEffectsIf removal:") - .run_pass(Ssa::fold_constants_using_constraints, "After Constraint Folding:") - .run_pass(Ssa::dead_instruction_elimination, "After Dead Instruction Elimination:") - .run_pass(Ssa::simplify_cfg, "After Simplifying:") - .run_pass(Ssa::array_set_optimization, "After Array Set Optimizations:") + .run_pass(Ssa::remove_if_else, "Remove IfElse") + .run_pass(Ssa::fold_constants, "Constant Folding") + .run_pass(Ssa::remove_enable_side_effects, "EnableSideEffectsIf removal") + .run_pass(Ssa::fold_constants_using_constraints, "Constraint Folding") + .run_pass(Ssa::dead_instruction_elimination, "Dead Instruction Elimination (1st)") + .run_pass(Ssa::simplify_cfg, "Simplifying:") + .run_pass(Ssa::array_set_optimization, "Array Set Optimizations") .finish(); let ssa_level_warnings = if options.skip_underconstrained_check { @@ -146,14 +161,11 @@ pub(crate) fn optimize_into_acir( let ssa = SsaBuilder { ssa, - print_ssa_passes: options.enable_ssa_logging, + ssa_logging: options.ssa_logging.clone(), print_codegen_timings: options.print_codegen_timings, } - .run_pass( - |ssa| ssa.fold_constants_with_brillig(&brillig), - "After Constant Folding with Brillig:", - ) - .run_pass(Ssa::dead_instruction_elimination, "After Dead Instruction Elimination:") + .run_pass(|ssa| ssa.fold_constants_with_brillig(&brillig), "Inlining Brillig Calls Inlining") + .run_pass(Ssa::dead_instruction_elimination, "Dead Instruction Elimination (2nd)") .finish(); drop(ssa_gen_span_guard); @@ -226,7 +238,7 @@ impl SsaProgramArtifact { } } -/// Compiles the [`Program`] into [`ACIR``][acvm::acir::circuit::Program]. +/// Compiles the [`Program`] into [`ACIR`][acvm::acir::circuit::Program]. /// /// The output ACIR is backend-agnostic and so must go through a transformation pass before usage in proof generation. #[tracing::instrument(level = "trace", skip_all)] @@ -411,14 +423,14 @@ fn split_public_and_private_inputs( // This is just a convenience object to bundle the ssa with `print_ssa_passes` for debug printing. struct SsaBuilder { ssa: Ssa, - print_ssa_passes: bool, + ssa_logging: SsaLogging, print_codegen_timings: bool, } impl SsaBuilder { fn new( program: Program, - print_ssa_passes: bool, + ssa_logging: SsaLogging, force_brillig_runtime: bool, print_codegen_timings: bool, emit_ssa: &Option, @@ -433,7 +445,7 @@ impl SsaBuilder { let ssa_path = emit_ssa.with_extension("ssa.json"); write_to_file(&serde_json::to_vec(&ssa).unwrap(), &ssa_path); } - Ok(SsaBuilder { print_ssa_passes, print_codegen_timings, ssa }.print("Initial SSA:")) + Ok(SsaBuilder { ssa_logging, print_codegen_timings, ssa }.print("Initial SSA:")) } fn finish(self) -> Ssa { @@ -450,19 +462,28 @@ impl SsaBuilder { } /// The same as `run_pass` but for passes that may fail - fn try_run_pass( - mut self, - pass: fn(Ssa) -> Result, - msg: &str, - ) -> Result { + fn try_run_pass(mut self, pass: F, msg: &str) -> Result + where + F: FnOnce(Ssa) -> Result, + { self.ssa = time(msg, self.print_codegen_timings, || pass(self.ssa))?; Ok(self.print(msg)) } fn print(mut self, msg: &str) -> Self { - if self.print_ssa_passes { + let print_ssa_pass = match &self.ssa_logging { + SsaLogging::None => false, + SsaLogging::All => true, + SsaLogging::Contains(string) => { + let string = string.to_lowercase(); + let string = string.strip_prefix("after ").unwrap_or(&string); + let string = string.strip_suffix(':').unwrap_or(string); + msg.to_lowercase().contains(string) + } + }; + if print_ssa_pass { self.ssa.normalize_ids(); - println!("{msg}\n{}", self.ssa); + println!("After {msg}:\n{}", self.ssa); } self } diff --git a/noir/noir-repo/compiler/noirc_evaluator/src/ssa/function_builder/data_bus.rs b/noir/noir-repo/compiler/noirc_evaluator/src/ssa/function_builder/data_bus.rs index e4a2eeb8c22..bd2585a3bfa 100644 --- a/noir/noir-repo/compiler/noirc_evaluator/src/ssa/function_builder/data_bus.rs +++ b/noir/noir-repo/compiler/noirc_evaluator/src/ssa/function_builder/data_bus.rs @@ -160,7 +160,7 @@ impl FunctionBuilder { for value in values { self.add_to_data_bus(*value, &mut databus); } - let len = databus.values.len(); + let len = databus.values.len() as u32; let array = (len > 0 && matches!(self.current_function.runtime(), RuntimeType::Acir(_))) .then(|| { @@ -223,9 +223,11 @@ impl FunctionBuilder { ssa_params: &[ValueId], mut flattened_params_databus_visibility: Vec, ) -> Vec { - let ssa_param_sizes: Vec<_> = ssa_params + let ssa_param_sizes: Vec = ssa_params .iter() - .map(|ssa_param| self.current_function.dfg[*ssa_param].get_type().flattened_size()) + .map(|ssa_param| { + self.current_function.dfg[*ssa_param].get_type().flattened_size() as usize + }) .collect(); let mut is_ssa_params_databus = Vec::with_capacity(ssa_params.len()); diff --git a/noir/noir-repo/compiler/noirc_evaluator/src/ssa/function_builder/mod.rs b/noir/noir-repo/compiler/noirc_evaluator/src/ssa/function_builder/mod.rs index 0479f8da0b7..0ae61404442 100644 --- a/noir/noir-repo/compiler/noirc_evaluator/src/ssa/function_builder/mod.rs +++ b/noir/noir-repo/compiler/noirc_evaluator/src/ssa/function_builder/mod.rs @@ -441,29 +441,38 @@ impl FunctionBuilder { /// Insert instructions to increment the reference count of any array(s) stored /// within the given value. If the given value is not an array and does not contain /// any arrays, this does nothing. - pub(crate) fn increment_array_reference_count(&mut self, value: ValueId) { - self.update_array_reference_count(value, true); + /// + /// Returns whether a reference count instruction was issued. + pub(crate) fn increment_array_reference_count(&mut self, value: ValueId) -> bool { + self.update_array_reference_count(value, true) } /// Insert instructions to decrement the reference count of any array(s) stored /// within the given value. If the given value is not an array and does not contain /// any arrays, this does nothing. - pub(crate) fn decrement_array_reference_count(&mut self, value: ValueId) { - self.update_array_reference_count(value, false); + /// + /// Returns whether a reference count instruction was issued. + pub(crate) fn decrement_array_reference_count(&mut self, value: ValueId) -> bool { + self.update_array_reference_count(value, false) } /// Increment or decrement the given value's reference count if it is an array. /// If it is not an array, this does nothing. Note that inc_rc and dec_rc instructions /// are ignored outside of unconstrained code. - fn update_array_reference_count(&mut self, value: ValueId, increment: bool) { + /// + /// Returns whether a reference count instruction was issued. + fn update_array_reference_count(&mut self, value: ValueId, increment: bool) -> bool { match self.type_of_value(value) { - Type::Numeric(_) => (), - Type::Function => (), + Type::Numeric(_) => false, + Type::Function => false, Type::Reference(element) => { if element.contains_an_array() { let reference = value; let value = self.insert_load(reference, element.as_ref().clone()); self.update_array_reference_count(value, increment); + true + } else { + false } } Type::Array(..) | Type::Slice(..) => { @@ -474,6 +483,7 @@ impl FunctionBuilder { } else { self.insert_dec_rc(value); } + true } } } diff --git a/noir/noir-repo/compiler/noirc_evaluator/src/ssa/ir/dfg.rs b/noir/noir-repo/compiler/noirc_evaluator/src/ssa/ir/dfg.rs index e3f3f33682b..827944e22d1 100644 --- a/noir/noir-repo/compiler/noirc_evaluator/src/ssa/ir/dfg.rs +++ b/noir/noir-repo/compiler/noirc_evaluator/src/ssa/ir/dfg.rs @@ -307,13 +307,13 @@ impl DataFlowGraph { instruction_id: InstructionId, ctrl_typevars: Option>, ) { - self.results.insert(instruction_id, Default::default()); + let result_types = self.instruction_result_types(instruction_id, ctrl_typevars); + let results = vecmap(result_types.into_iter().enumerate(), |(position, typ)| { + let instruction = instruction_id; + self.values.insert(Value::Instruction { typ, position, instruction }) + }); - // Get all of the types that this instruction produces - // and append them as results. - for typ in self.instruction_result_types(instruction_id, ctrl_typevars) { - self.append_result(instruction_id, typ); - } + self.results.insert(instruction_id, results); } /// Return the result types of this instruction. @@ -370,22 +370,6 @@ impl DataFlowGraph { matches!(self.values[value].get_type(), Type::Reference(_)) } - /// Appends a result type to the instruction. - pub(crate) fn append_result(&mut self, instruction_id: InstructionId, typ: Type) -> ValueId { - let results = self.results.get_mut(&instruction_id).unwrap(); - let expected_res_position = results.len(); - - let value_id = self.values.insert(Value::Instruction { - typ, - position: expected_res_position, - instruction: instruction_id, - }); - - // Add value to the list of results for this instruction - results.push(value_id); - value_id - } - /// Replaces an instruction result with a fresh id. pub(crate) fn replace_result( &mut self, @@ -463,7 +447,7 @@ impl DataFlowGraph { /// If this value is an array, return the length of the array as indicated by its type. /// Otherwise, return None. - pub(crate) fn try_get_array_length(&self, value: ValueId) -> Option { + pub(crate) fn try_get_array_length(&self, value: ValueId) -> Option { match self.type_of_value(value) { Type::Array(_, length) => Some(length), _ => None, diff --git a/noir/noir-repo/compiler/noirc_evaluator/src/ssa/ir/function.rs b/noir/noir-repo/compiler/noirc_evaluator/src/ssa/ir/function.rs index b1233e3063e..6413107c04a 100644 --- a/noir/noir-repo/compiler/noirc_evaluator/src/ssa/ir/function.rs +++ b/noir/noir-repo/compiler/noirc_evaluator/src/ssa/ir/function.rs @@ -197,6 +197,12 @@ impl Function { } } +impl Clone for Function { + fn clone(&self) -> Self { + Function::clone_with_id(self.id(), self) + } +} + impl std::fmt::Display for RuntimeType { fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result { match self { diff --git a/noir/noir-repo/compiler/noirc_evaluator/src/ssa/ir/function_inserter.rs b/noir/noir-repo/compiler/noirc_evaluator/src/ssa/ir/function_inserter.rs index a0c23ad70aa..6ebd2aa1105 100644 --- a/noir/noir-repo/compiler/noirc_evaluator/src/ssa/ir/function_inserter.rs +++ b/noir/noir-repo/compiler/noirc_evaluator/src/ssa/ir/function_inserter.rs @@ -129,7 +129,7 @@ impl<'f> FunctionInserter<'f> { // another MakeArray instruction. Note that this assumes the function inserter is inserting // in control-flow order. Otherwise we could refer to ValueIds defined later in the program. let make_array = if let Instruction::MakeArray { elements, typ } = &instruction { - if self.array_is_constant(elements) { + if self.array_is_constant(elements) && self.function.runtime().is_acir() { if let Some(fetched_value) = self.get_cached_array(elements, typ) { assert_eq!(results.len(), 1); self.values.insert(results[0], fetched_value); diff --git a/noir/noir-repo/compiler/noirc_evaluator/src/ssa/ir/instruction.rs b/noir/noir-repo/compiler/noirc_evaluator/src/ssa/ir/instruction.rs index f606fffbf91..76409f6a20a 100644 --- a/noir/noir-repo/compiler/noirc_evaluator/src/ssa/ir/instruction.rs +++ b/noir/noir-repo/compiler/noirc_evaluator/src/ssa/ir/instruction.rs @@ -11,7 +11,7 @@ use fxhash::FxHasher64; use iter_extended::vecmap; use noirc_frontend::hir_def::types::Type as HirType; -use crate::ssa::opt::flatten_cfg::value_merger::ValueMerger; +use crate::ssa::{ir::function::RuntimeType, opt::flatten_cfg::value_merger::ValueMerger}; use super::{ basic_block::BasicBlockId, @@ -315,7 +315,12 @@ pub(crate) enum Instruction { /// else_value /// } /// ``` - IfElse { then_condition: ValueId, then_value: ValueId, else_value: ValueId }, + IfElse { + then_condition: ValueId, + then_value: ValueId, + else_condition: ValueId, + else_value: ValueId, + }, /// Creates a new array or slice. /// @@ -389,9 +394,22 @@ impl Instruction { // This should never be side-effectful MakeArray { .. } => false, + // Some binary math can overflow or underflow + Binary(binary) => match binary.operator { + BinaryOp::Add | BinaryOp::Sub | BinaryOp::Mul | BinaryOp::Div | BinaryOp::Mod => { + true + } + BinaryOp::Eq + | BinaryOp::Lt + | BinaryOp::And + | BinaryOp::Or + | BinaryOp::Xor + | BinaryOp::Shl + | BinaryOp::Shr => false, + }, + // These can have different behavior depending on the EnableSideEffectsIf context. - Binary(_) - | Cast(_, _) + Cast(_, _) | Not(_) | Truncate { .. } | IfElse { .. } @@ -411,7 +429,7 @@ impl Instruction { /// conditional on whether the caller wants the predicate to be taken into account or not. pub(crate) fn can_be_deduplicated( &self, - dfg: &DataFlowGraph, + function: &Function, deduplicate_with_predicate: bool, ) -> bool { use Instruction::*; @@ -425,7 +443,7 @@ impl Instruction { | IncrementRc { .. } | DecrementRc { .. } => false, - Call { func, .. } => match dfg[*func] { + Call { func, .. } => match function.dfg[*func] { Value::Intrinsic(intrinsic) => { intrinsic.can_be_deduplicated(deduplicate_with_predicate) } @@ -435,8 +453,11 @@ impl Instruction { // We can deduplicate these instructions if we know the predicate is also the same. Constrain(..) | RangeCheck { .. } => deduplicate_with_predicate, - // This should never be side-effectful - MakeArray { .. } => true, + // Arrays can be mutated in unconstrained code so code that handles this case must + // take care to track whether the array was possibly mutated or not before + // deduplicating. Since we don't know if the containing pass checks for this, we + // can only assume these are safe to deduplicate in constrained code. + MakeArray { .. } => function.runtime().is_acir(), // These can have different behavior depending on the EnableSideEffectsIf context. // Replacing them with a similar instruction potentially enables replacing an instruction @@ -449,7 +470,7 @@ impl Instruction { | IfElse { .. } | ArrayGet { .. } | ArraySet { .. } => { - deduplicate_with_predicate || !self.requires_acir_gen_predicate(dfg) + deduplicate_with_predicate || !self.requires_acir_gen_predicate(&function.dfg) } } } @@ -478,8 +499,19 @@ impl Instruction { | ArraySet { .. } | MakeArray { .. } => true, + // Store instructions must be removed by DIE in acir code, any load + // instructions should already be unused by that point. + // + // Note that this check assumes that it is being performed after the flattening + // pass and after the last mem2reg pass. This is currently the case for the DIE + // pass where this check is done, but does mean that we cannot perform mem2reg + // after the DIE pass. + Store { .. } => { + matches!(function.runtime(), RuntimeType::Acir(_)) + && function.reachable_blocks().len() == 1 + } + Constrain(..) - | Store { .. } | EnableSideEffectsIf { .. } | IncrementRc { .. } | DecrementRc { .. } @@ -608,11 +640,14 @@ impl Instruction { assert_message: assert_message.clone(), } } - Instruction::IfElse { then_condition, then_value, else_value } => Instruction::IfElse { - then_condition: f(*then_condition), - then_value: f(*then_value), - else_value: f(*else_value), - }, + Instruction::IfElse { then_condition, then_value, else_condition, else_value } => { + Instruction::IfElse { + then_condition: f(*then_condition), + then_value: f(*then_value), + else_condition: f(*else_condition), + else_value: f(*else_value), + } + } Instruction::MakeArray { elements, typ } => Instruction::MakeArray { elements: elements.iter().copied().map(f).collect(), typ: typ.clone(), @@ -671,9 +706,10 @@ impl Instruction { | Instruction::RangeCheck { value, .. } => { f(*value); } - Instruction::IfElse { then_condition, then_value, else_value } => { + Instruction::IfElse { then_condition, then_value, else_condition, else_value } => { f(*then_condition); f(*then_value); + f(*else_condition); f(*else_value); } Instruction::MakeArray { elements, typ: _ } => { @@ -836,7 +872,7 @@ impl Instruction { None } } - Instruction::IfElse { then_condition, then_value, else_value } => { + Instruction::IfElse { then_condition, then_value, else_condition, else_value } => { let typ = dfg.type_of_value(*then_value); if let Some(constant) = dfg.get_numeric_constant(*then_condition) { @@ -855,11 +891,13 @@ impl Instruction { if matches!(&typ, Type::Numeric(_)) { let then_condition = *then_condition; + let else_condition = *else_condition; let result = ValueMerger::merge_numeric_values( dfg, block, then_condition, + else_condition, then_value, else_value, ); diff --git a/noir/noir-repo/compiler/noirc_evaluator/src/ssa/ir/instruction/call.rs b/noir/noir-repo/compiler/noirc_evaluator/src/ssa/ir/instruction/call.rs index 6ebe80128c0..a8db5e2ff94 100644 --- a/noir/noir-repo/compiler/noirc_evaluator/src/ssa/ir/instruction/call.rs +++ b/noir/noir-repo/compiler/noirc_evaluator/src/ssa/ir/instruction/call.rs @@ -56,11 +56,13 @@ pub(super) fn simplify_call( if let (Some(constant_args), Some(return_type)) = (constant_args, return_type.clone()) { let field = constant_args[0]; let limb_count = if let Type::Array(_, array_len) = return_type { - array_len as u32 + array_len } else { unreachable!("ICE: Intrinsic::ToRadix return type must be array") }; - constant_to_radix(endian, field, 2, limb_count, dfg, block, call_stack) + constant_to_radix(endian, field, 2, limb_count, |values| { + make_constant_array(dfg, values.into_iter(), Type::bool(), block, call_stack) + }) } else { SimplifyResult::None } @@ -71,11 +73,19 @@ pub(super) fn simplify_call( let field = constant_args[0]; let radix = constant_args[1].to_u128() as u32; let limb_count = if let Type::Array(_, array_len) = return_type { - array_len as u32 + array_len } else { unreachable!("ICE: Intrinsic::ToRadix return type must be array") }; - constant_to_radix(endian, field, radix, limb_count, dfg, block, call_stack) + constant_to_radix(endian, field, radix, limb_count, |values| { + make_constant_array( + dfg, + values.into_iter(), + Type::unsigned(8), + block, + call_stack, + ) + }) } else { SimplifyResult::None } @@ -351,7 +361,7 @@ pub(super) fn simplify_call( Intrinsic::IsUnconstrained => SimplifyResult::None, Intrinsic::DerivePedersenGenerators => { if let Some(Type::Array(_, len)) = return_type.clone() { - simplify_derive_generators(dfg, arguments, len as u32, block, call_stack) + simplify_derive_generators(dfg, arguments, len, block, call_stack) } else { unreachable!("Derive Pedersen Generators must return an array"); } @@ -432,8 +442,8 @@ fn simplify_slice_push_back( for elem in &arguments[2..] { slice.push_back(*elem); } - let slice_size = slice.len(); - let element_size = element_type.element_size(); + let slice_size = slice.len() as u32; + let element_size = element_type.element_size() as u32; let new_slice = make_array(dfg, slice, element_type, block, &call_stack); let set_last_slice_value_instr = Instruction::ArraySet { @@ -455,8 +465,12 @@ fn simplify_slice_push_back( let mut value_merger = ValueMerger::new(dfg, block, &mut slice_sizes, unknown, None, call_stack); - let new_slice = - value_merger.merge_values(len_not_equals_capacity, set_last_slice_value, new_slice); + let new_slice = value_merger.merge_values( + len_not_equals_capacity, + len_equals_capacity, + set_last_slice_value, + new_slice, + ); SimplifyResult::SimplifiedToMultiple(vec![new_slice_length, new_slice]) } @@ -621,7 +635,7 @@ fn make_constant_array( let result_constants: im::Vector<_> = results.map(|element| dfg.make_constant(element, typ.clone())).collect(); - let typ = Type::Array(Arc::new(vec![typ]), result_constants.len()); + let typ = Type::Array(Arc::new(vec![typ]), result_constants.len() as u32); make_array(dfg, result_constants, typ, block, call_stack) } @@ -660,9 +674,7 @@ fn constant_to_radix( field: FieldElement, radix: u32, limb_count: u32, - dfg: &mut DataFlowGraph, - block: BasicBlockId, - call_stack: &CallStack, + mut make_array: impl FnMut(Vec) -> ValueId, ) -> SimplifyResult { let bit_size = u32::BITS - (radix - 1).leading_zeros(); let radix_big = BigUint::from(radix); @@ -683,13 +695,7 @@ fn constant_to_radix( if endian == Endian::Big { limbs.reverse(); } - let result_array = make_constant_array( - dfg, - limbs.into_iter(), - Type::unsigned(bit_size), - block, - call_stack, - ); + let result_array = make_array(limbs); SimplifyResult::SimplifiedTo(result_array) } } @@ -816,7 +822,7 @@ fn simplify_derive_generators( results.push(dfg.make_constant(y, Type::field())); results.push(is_infinite); } - let len = results.len(); + let len = results.len() as u32; let typ = Type::Array(vec![Type::field(), Type::field(), Type::unsigned(1)].into(), len / 3); let result = make_array(dfg, results.into(), typ, block, call_stack); @@ -835,27 +841,27 @@ mod tests { #[test] fn simplify_derive_generators_has_correct_type() { - let src = " + let src = r#" brillig(inline) fn main f0 { b0(): - v0 = make_array [u8 68, u8 69, u8 70, u8 65, u8 85, u8 76, u8 84, u8 95, u8 68, u8 79, u8 77, u8 65, u8 73, u8 78, u8 95, u8 83, u8 69, u8 80, u8 65, u8 82, u8 65, u8 84, u8 79, u8 82] : [u8; 24] + v0 = make_array b"DEFAULT_DOMAIN_SEPARATOR" // This call was previously incorrectly simplified to something that returned `[Field; 3]` v2 = call derive_pedersen_generators(v0, u32 0) -> [(Field, Field, u1); 1] return v2 } - "; + "#; let ssa = Ssa::from_str(src).unwrap(); - let expected = " + let expected = r#" brillig(inline) fn main f0 { b0(): - v15 = make_array [u8 68, u8 69, u8 70, u8 65, u8 85, u8 76, u8 84, u8 95, u8 68, u8 79, u8 77, u8 65, u8 73, u8 78, u8 95, u8 83, u8 69, u8 80, u8 65, u8 82, u8 65, u8 84, u8 79, u8 82] : [u8; 24] + v15 = make_array b"DEFAULT_DOMAIN_SEPARATOR" v19 = make_array [Field 3728882899078719075161482178784387565366481897740339799480980287259621149274, Field -9903063709032878667290627648209915537972247634463802596148419711785767431332, u1 0] : [(Field, Field, u1); 1] return v19 } - "; + "#; assert_normalized_ssa_equals(ssa, expected); } } diff --git a/noir/noir-repo/compiler/noirc_evaluator/src/ssa/ir/printer.rs b/noir/noir-repo/compiler/noirc_evaluator/src/ssa/ir/printer.rs index 6bebd21fe61..aa2952d5abc 100644 --- a/noir/noir-repo/compiler/noirc_evaluator/src/ssa/ir/printer.rs +++ b/noir/noir-repo/compiler/noirc_evaluator/src/ssa/ir/printer.rs @@ -5,8 +5,11 @@ use std::{ }; use acvm::acir::AcirField; +use im::Vector; use iter_extended::vecmap; +use crate::ssa::ir::types::{NumericType, Type}; + use super::{ basic_block::BasicBlockId, dfg::DataFlowGraph, @@ -209,13 +212,39 @@ fn display_instruction_inner( Instruction::RangeCheck { value, max_bit_size, .. } => { writeln!(f, "range_check {} to {} bits", show(*value), *max_bit_size,) } - Instruction::IfElse { then_condition, then_value, else_value } => { + Instruction::IfElse { then_condition, then_value, else_condition, else_value } => { let then_condition = show(*then_condition); let then_value = show(*then_value); + let else_condition = show(*else_condition); let else_value = show(*else_value); - writeln!(f, "if {then_condition} then {then_value} else {else_value}") + writeln!( + f, + "if {then_condition} then {then_value} else (if {else_condition}) {else_value}" + ) } Instruction::MakeArray { elements, typ } => { + // If the array is a byte array, we check if all the bytes are printable ascii characters + // and, if so, we print the array as a string literal (easier to understand). + // It could happen that the byte array is a random byte sequence that happens to be printable + // (it didn't come from a string literal) but this still reduces the noise in the output + // and actually represents the same value. + let (element_types, is_slice) = match typ { + Type::Array(types, _) => (types, false), + Type::Slice(types) => (types, true), + _ => panic!("Expected array or slice type for MakeArray"), + }; + if element_types.len() == 1 + && element_types[0] == Type::Numeric(NumericType::Unsigned { bit_size: 8 }) + { + if let Some(string) = try_byte_array_to_string(elements, function) { + if is_slice { + return writeln!(f, "make_array &b{:?}", string); + } else { + return writeln!(f, "make_array b{:?}", string); + } + } + } + write!(f, "make_array [")?; for (i, element) in elements.iter().enumerate() { @@ -230,6 +259,25 @@ fn display_instruction_inner( } } +fn try_byte_array_to_string(elements: &Vector, function: &Function) -> Option { + let mut string = String::new(); + for element in elements { + let element = function.dfg.get_numeric_constant(*element)?; + let element = element.try_to_u32()?; + if element > 0xFF { + return None; + } + let byte = element as u8; + if byte.is_ascii_alphanumeric() || byte.is_ascii_punctuation() || byte.is_ascii_whitespace() + { + string.push(byte as char); + } else { + return None; + } + } + Some(string) +} + fn result_types(function: &Function, results: &[ValueId]) -> String { let types = vecmap(results, |result| function.dfg.type_of_value(*result).to_string()); if types.is_empty() { diff --git a/noir/noir-repo/compiler/noirc_evaluator/src/ssa/ir/types.rs b/noir/noir-repo/compiler/noirc_evaluator/src/ssa/ir/types.rs index 130f1d59e46..4e4f7e8aa62 100644 --- a/noir/noir-repo/compiler/noirc_evaluator/src/ssa/ir/types.rs +++ b/noir/noir-repo/compiler/noirc_evaluator/src/ssa/ir/types.rs @@ -75,7 +75,7 @@ pub(crate) enum Type { Reference(Arc), /// An immutable array value with the given element type and length - Array(Arc, usize), + Array(Arc, u32), /// An immutable slice value with a given element type Slice(Arc), @@ -111,7 +111,7 @@ impl Type { } /// Creates the str type, of the given length N - pub(crate) fn str(length: usize) -> Type { + pub(crate) fn str(length: u32) -> Type { Type::Array(Arc::new(vec![Type::char()]), length) } @@ -161,7 +161,7 @@ impl Type { } /// Returns the flattened size of a Type - pub(crate) fn flattened_size(&self) -> usize { + pub(crate) fn flattened_size(&self) -> u32 { match self { Type::Array(elements, len) => { elements.iter().fold(0, |sum, elem| sum + (elem.flattened_size() * len)) @@ -190,6 +190,15 @@ impl Type { } } + /// Retrieves the array or slice type within this type, or panics if there is none. + pub(crate) fn get_contained_array(&self) -> &Type { + match self { + Type::Numeric(_) | Type::Function => panic!("Expected an array type"), + Type::Array(_, _) | Type::Slice(_) => self, + Type::Reference(element) => element.get_contained_array(), + } + } + pub(crate) fn element_types(self) -> Arc> { match self { Type::Array(element_types, _) | Type::Slice(element_types) => element_types, diff --git a/noir/noir-repo/compiler/noirc_evaluator/src/ssa/opt/as_slice_length.rs b/noir/noir-repo/compiler/noirc_evaluator/src/ssa/opt/as_slice_length.rs index 76705dcc9db..75cdea349b7 100644 --- a/noir/noir-repo/compiler/noirc_evaluator/src/ssa/opt/as_slice_length.rs +++ b/noir/noir-repo/compiler/noirc_evaluator/src/ssa/opt/as_slice_length.rs @@ -33,7 +33,7 @@ impl Function { } } -fn known_slice_lengths(func: &Function) -> HashMap { +fn known_slice_lengths(func: &Function) -> HashMap { let mut known_slice_lengths = HashMap::default(); for block_id in func.reachable_blocks() { let block = &func.dfg[block_id]; @@ -61,7 +61,7 @@ fn known_slice_lengths(func: &Function) -> HashMap { fn replace_known_slice_lengths( func: &mut Function, - known_slice_lengths: HashMap, + known_slice_lengths: HashMap, ) { known_slice_lengths.into_iter().for_each(|(instruction_id, known_length)| { let call_returns = func.dfg.instruction_results(instruction_id); diff --git a/noir/noir-repo/compiler/noirc_evaluator/src/ssa/opt/constant_folding.rs b/noir/noir-repo/compiler/noirc_evaluator/src/ssa/opt/constant_folding.rs index ceda0c6272f..93ca428c6d0 100644 --- a/noir/noir-repo/compiler/noirc_evaluator/src/ssa/opt/constant_folding.rs +++ b/noir/noir-repo/compiler/noirc_evaluator/src/ssa/opt/constant_folding.rs @@ -149,7 +149,8 @@ impl Function { use_constraint_info: bool, brillig_info: Option, ) { - let mut context = Context::new(self, use_constraint_info, brillig_info); + let mut context = Context::new(use_constraint_info, brillig_info); + let mut dom = DominatorTree::with_function(self); context.block_queue.push_back(self.entry_block()); while let Some(block) = context.block_queue.pop_front() { @@ -158,7 +159,7 @@ impl Function { } context.visited_blocks.insert(block); - context.fold_constants_in_block(self, block); + context.fold_constants_in_block(self, &mut dom, block); } } } @@ -172,22 +173,15 @@ struct Context<'a> { /// Contains sets of values which are constrained to be equivalent to each other. /// - /// The mapping's structure is `side_effects_enabled_var => (constrained_value => [(block, simplified_value)])`. + /// The mapping's structure is `side_effects_enabled_var => (constrained_value => simplified_value)`. /// /// We partition the maps of constrained values according to the side-effects flag at the point /// at which the values are constrained. This prevents constraints which are only sometimes enforced /// being used to modify the rest of the program. - /// - /// We also keep track of how a value was simplified to other values per block. That is, - /// a same ValueId could have been simplified to one value in one block and to another value - /// in another block. - constraint_simplification_mappings: - HashMap>>, + constraint_simplification_mappings: ConstraintSimplificationCache, // Cache of instructions without any side-effects along with their outputs. cached_instruction_results: InstructionResultCache, - - dom: DominatorTree, } #[derive(Copy, Clone)] @@ -196,9 +190,56 @@ pub(crate) struct BrilligInfo<'a> { brillig_functions: &'a BTreeMap, } +/// Records a simplified equivalents of an [`Instruction`] in the blocks +/// where the constraint that advised the simplification has been encountered. +/// +/// For more information see [`ConstraintSimplificationCache`]. +#[derive(Default)] +struct SimplificationCache { + /// Simplified expressions where we found them. + /// + /// It will always have at least one value because `add` is called + /// after the default is constructed. + simplifications: HashMap, +} + +impl SimplificationCache { + /// Called with a newly encountered simplification. + fn add(&mut self, dfg: &DataFlowGraph, simple: ValueId, block: BasicBlockId) { + self.simplifications + .entry(block) + .and_modify(|existing| { + // `SimplificationCache` may already hold a simplification in this block + // so we check whether `simple` is a better simplification than the current one. + if let Some((_, simpler)) = simplify(dfg, *existing, simple) { + *existing = simpler; + }; + }) + .or_insert(simple); + } + + /// Try to find a simplification in a visible block. + fn get(&self, block: BasicBlockId, dom: &DominatorTree) -> Option { + // Deterministically walk up the dominator chain until we encounter a block that contains a simplification. + dom.find_map_dominator(block, |b| self.simplifications.get(&b).cloned()) + } +} + +/// HashMap from `(side_effects_enabled_var, Instruction)` to a simplified expression that it can +/// be replaced with based on constraints that testify to their equivalence, stored together +/// with the set of blocks at which this constraint has been observed. +/// +/// Only blocks dominated by one in the cache should have access to this information, otherwise +/// we create a sort of time paradox where we replace an instruction with a constant we believe +/// it _should_ equal to, without ever actually producing and asserting the value. +type ConstraintSimplificationCache = HashMap>; + /// HashMap from `(Instruction, side_effects_enabled_var)` to the results of the instruction. /// Stored as a two-level map to avoid cloning Instructions during the `.get` call. /// +/// The `side_effects_enabled_var` is optional because we only use them when `Instruction::requires_acir_gen_predicate` +/// is true _and_ the constraint information is also taken into account. +/// /// In addition to each result, the original BasicBlockId is stored as well. This allows us /// to deduplicate instructions across blocks as long as the new block dominates the original. type InstructionResultCache = HashMap, ResultCache>>; @@ -208,15 +249,11 @@ type InstructionResultCache = HashMap, Resu /// For more information see [`InstructionResultCache`]. #[derive(Default)] struct ResultCache { - results: Vec<(BasicBlockId, Vec)>, + result: Option<(BasicBlockId, Vec)>, } impl<'brillig> Context<'brillig> { - fn new( - function: &Function, - use_constraint_info: bool, - brillig_info: Option>, - ) -> Self { + fn new(use_constraint_info: bool, brillig_info: Option>) -> Self { Self { use_constraint_info, brillig_info, @@ -224,19 +261,25 @@ impl<'brillig> Context<'brillig> { block_queue: Default::default(), constraint_simplification_mappings: Default::default(), cached_instruction_results: Default::default(), - dom: DominatorTree::with_function(function), } } - fn fold_constants_in_block(&mut self, function: &mut Function, block: BasicBlockId) { + fn fold_constants_in_block( + &mut self, + function: &mut Function, + dom: &mut DominatorTree, + block: BasicBlockId, + ) { let instructions = function.dfg[block].take_instructions(); + // Default side effect condition variable with an enabled state. let mut side_effects_enabled_var = function.dfg.make_constant(FieldElement::one(), Type::bool()); for instruction_id in instructions { self.fold_constants_into_instruction( - &mut function.dfg, + function, + dom, block, instruction_id, &mut side_effects_enabled_var, @@ -247,48 +290,59 @@ impl<'brillig> Context<'brillig> { fn fold_constants_into_instruction( &mut self, - dfg: &mut DataFlowGraph, - block: BasicBlockId, + function: &mut Function, + dom: &mut DominatorTree, + mut block: BasicBlockId, id: InstructionId, side_effects_enabled_var: &mut ValueId, ) { - let constraint_simplification_mapping = - self.constraint_simplification_mappings.get(side_effects_enabled_var); - let instruction = Self::resolve_instruction( - id, - block, - dfg, - &mut self.dom, - constraint_simplification_mapping, - ); + let constraint_simplification_mapping = self.get_constraint_map(*side_effects_enabled_var); + let dfg = &mut function.dfg; + + let instruction = + Self::resolve_instruction(id, block, dfg, dom, constraint_simplification_mapping); + let old_results = dfg.instruction_results(id).to_vec(); // If a copy of this instruction exists earlier in the block, then reuse the previous results. - if let Some(cached_results) = - self.get_cached(dfg, &instruction, *side_effects_enabled_var, block) + if let Some(cache_result) = + self.get_cached(dfg, dom, &instruction, *side_effects_enabled_var, block) { - Self::replace_result_ids(dfg, &old_results, cached_results); - return; - } + match cache_result { + CacheResult::Cached(cached) => { + // We track whether we may mutate MakeArray instructions before we deduplicate + // them but we still need to issue an extra inc_rc in case they're mutated afterward. + if matches!(instruction, Instruction::MakeArray { .. }) { + let value = *cached.last().unwrap(); + let inc_rc = Instruction::IncrementRc { value }; + let call_stack = dfg.get_call_stack(id); + dfg.insert_instruction_and_results(inc_rc, block, None, call_stack); + } + + Self::replace_result_ids(dfg, &old_results, cached); + return; + } + CacheResult::NeedToHoistToCommonBlock(dominator) => { + // Just change the block to insert in the common dominator instead. + // This will only move the current instance of the instruction right now. + // When constant folding is run a second time later on, it'll catch + // that the previous instance can be deduplicated to this instance. + block = dominator; + } + } + }; - let new_results = // First try to inline a call to a brillig function with all constant arguments. - Self::try_inline_brillig_call_with_all_constants( + let new_results = Self::try_inline_brillig_call_with_all_constants( &instruction, &old_results, block, dfg, self.brillig_info, ) + // Otherwise, try inserting the instruction again to apply any optimizations using the newly resolved inputs. .unwrap_or_else(|| { - // Otherwise, try inserting the instruction again to apply any optimizations using the newly resolved inputs. - Self::push_instruction( - id, - instruction.clone(), - &old_results, - block, - dfg, - ) + Self::push_instruction(id, instruction.clone(), &old_results, block, dfg) }); Self::replace_result_ids(dfg, &old_results, &new_results); @@ -296,7 +350,7 @@ impl<'brillig> Context<'brillig> { self.cache_instruction( instruction.clone(), new_results, - dfg, + function, *side_effects_enabled_var, block, ); @@ -314,7 +368,7 @@ impl<'brillig> Context<'brillig> { block: BasicBlockId, dfg: &DataFlowGraph, dom: &mut DominatorTree, - constraint_simplification_mapping: Option<&HashMap>>, + constraint_simplification_mapping: &HashMap, ) -> Instruction { let instruction = dfg[instruction_id].clone(); @@ -324,30 +378,28 @@ impl<'brillig> Context<'brillig> { // This allows us to reach a stable final `ValueId` for each instruction input as we add more // constraints to the cache. fn resolve_cache( + block: BasicBlockId, dfg: &DataFlowGraph, dom: &mut DominatorTree, - cache: Option<&HashMap>>, + cache: &HashMap, value_id: ValueId, - block: BasicBlockId, ) -> ValueId { let resolved_id = dfg.resolve(value_id); - let Some(cached_values) = cache.and_then(|cache| cache.get(&resolved_id)) else { - return resolved_id; - }; - - for (cached_block, cached_value) in cached_values { - // We can only use the simplified value if it was simplified in a block that dominates the current one - if dom.dominates(*cached_block, block) { - return resolve_cache(dfg, dom, cache, *cached_value, block); + match cache.get(&resolved_id) { + Some(simplification_cache) => { + if let Some(simplified) = simplification_cache.get(block, dom) { + resolve_cache(block, dfg, dom, cache, simplified) + } else { + resolved_id + } } + None => resolved_id, } - - resolved_id } // Resolve any inputs to ensure that we're comparing like-for-like instructions. instruction.map_values(|value_id| { - resolve_cache(dfg, dom, constraint_simplification_mapping, value_id, block) + resolve_cache(block, dfg, dom, constraint_simplification_mapping, value_id) }) } @@ -385,7 +437,7 @@ impl<'brillig> Context<'brillig> { &mut self, instruction: Instruction, instruction_results: Vec, - dfg: &DataFlowGraph, + function: &Function, side_effects_enabled_var: ValueId, block: BasicBlockId, ) { @@ -394,21 +446,53 @@ impl<'brillig> Context<'brillig> { // to map from the more complex to the simpler value. if let Instruction::Constrain(lhs, rhs, _) = instruction { // These `ValueId`s should be fully resolved now. - if let Some((complex, simple)) = simplify(dfg, lhs, rhs) { + if let Some((complex, simple)) = simplify(&function.dfg, lhs, rhs) { self.get_constraint_map(side_effects_enabled_var) .entry(complex) .or_default() - .push((block, simple)); + .add(&function.dfg, simple, block); } } } + // If we have an array get whose value is from an array set on the same array at the same index, + // we can simplify that array get to the value of the previous array set. + // + // For example: + // v3 = array_set v0, index v1, value v2 + // v4 = array_get v3, index v1 -> Field + // + // We know that `v4` can be simplified to `v2`. + // Thus, even if the index is dynamic (meaning the array get would have side effects), + // we can simplify the operation when we take into account the predicate. + if let Instruction::ArraySet { index, value, .. } = &instruction { + let use_predicate = + self.use_constraint_info && instruction.requires_acir_gen_predicate(&function.dfg); + let predicate = use_predicate.then_some(side_effects_enabled_var); + + let array_get = Instruction::ArrayGet { array: instruction_results[0], index: *index }; + + self.cached_instruction_results + .entry(array_get) + .or_default() + .entry(predicate) + .or_default() + .cache(block, vec![*value]); + } + + self.remove_possibly_mutated_cached_make_arrays(&instruction, function); + // If the instruction doesn't have side-effects and if it won't interact with enable_side_effects during acir_gen, // we cache the results so we can reuse them if the same instruction appears again later in the block. // Others have side effects representing failure, which are implicit in the ACIR code and can also be deduplicated. - if instruction.can_be_deduplicated(dfg, self.use_constraint_info) { + let can_be_deduplicated = + instruction.can_be_deduplicated(function, self.use_constraint_info); + + // We also allow deduplicating MakeArray instructions that we have tracked which haven't + // been mutated. + if can_be_deduplicated || matches!(instruction, Instruction::MakeArray { .. }) { let use_predicate = - self.use_constraint_info && instruction.requires_acir_gen_predicate(dfg); + self.use_constraint_info && instruction.requires_acir_gen_predicate(&function.dfg); let predicate = use_predicate.then_some(side_effects_enabled_var); self.cached_instruction_results @@ -420,10 +504,12 @@ impl<'brillig> Context<'brillig> { } } + /// Get the simplification mapping from complex to simpler instructions, + /// which all depend on the same side effect condition variable. fn get_constraint_map( &mut self, side_effects_enabled_var: ValueId, - ) -> &mut HashMap> { + ) -> &mut HashMap { self.constraint_simplification_mappings.entry(side_effects_enabled_var).or_default() } @@ -438,19 +524,20 @@ impl<'brillig> Context<'brillig> { } } - fn get_cached<'a>( - &'a mut self, + /// Get a cached result if it can be used in this context. + fn get_cached( + &self, dfg: &DataFlowGraph, + dom: &mut DominatorTree, instruction: &Instruction, side_effects_enabled_var: ValueId, block: BasicBlockId, - ) -> Option<&'a [ValueId]> { + ) -> Option { let results_for_instruction = self.cached_instruction_results.get(instruction)?; - let predicate = self.use_constraint_info && instruction.requires_acir_gen_predicate(dfg); let predicate = predicate.then_some(side_effects_enabled_var); - results_for_instruction.get(&predicate)?.get(block, &mut self.dom) + results_for_instruction.get(&predicate)?.get(block, dom, instruction.has_side_effects(dfg)) } /// Checks if the given instruction is a call to a brillig function with all constant arguments. @@ -612,12 +699,34 @@ impl<'brillig> Context<'brillig> { } } } + + fn remove_possibly_mutated_cached_make_arrays( + &mut self, + instruction: &Instruction, + function: &Function, + ) { + use Instruction::{ArraySet, Store}; + + // Should we consider calls to slice_push_back and similar to be mutating operations as well? + if let Store { value: array, .. } | ArraySet { array, .. } = instruction { + let instruction = match &function.dfg[*array] { + Value::Instruction { instruction, .. } => &function.dfg[*instruction], + _ => return, + }; + + if matches!(instruction, Instruction::MakeArray { .. }) { + self.cached_instruction_results.remove(instruction); + } + } + } } impl ResultCache { /// Records that an `Instruction` in block `block` produced the result values `results`. fn cache(&mut self, block: BasicBlockId, results: Vec) { - self.results.push((block, results)); + if self.result.is_none() { + self.result = Some((block, results)); + } } /// Returns a set of [`ValueId`]s produced from a copy of this [`Instruction`] which sits @@ -626,13 +735,23 @@ impl ResultCache { /// We require that the cached instruction's block dominates `block` in order to avoid /// cycles causing issues (e.g. two instructions being replaced with the results of each other /// such that neither instruction exists anymore.) - fn get(&self, block: BasicBlockId, dom: &mut DominatorTree) -> Option<&[ValueId]> { - for (origin_block, results) in &self.results { + fn get( + &self, + block: BasicBlockId, + dom: &mut DominatorTree, + has_side_effects: bool, + ) -> Option { + self.result.as_ref().and_then(|(origin_block, results)| { if dom.dominates(*origin_block, block) { - return Some(results); + Some(CacheResult::Cached(results)) + } else if !has_side_effects { + // Insert a copy of this instruction in the common dominator + let dominator = dom.common_dominator(*origin_block, block); + Some(CacheResult::NeedToHoistToCommonBlock(dominator)) + } else { + None } - } - None + }) } } @@ -663,7 +782,7 @@ pub(crate) fn type_to_brillig_parameter(typ: &Type) -> Option for item_typ in item_type.iter() { parameters.push(type_to_brillig_parameter(item_typ)?); } - Some(BrilligParameter::Array(parameters, *size)) + Some(BrilligParameter::Array(parameters, *size as usize)) } _ => None, } @@ -940,32 +1059,22 @@ mod test { // Regression for #4600 #[test] fn array_get_regression() { - // fn main f0 { - // b0(v0: u1, v1: u64): - // enable_side_effects_if v0 - // v2 = make_array [Field 0, Field 1] - // v3 = array_get v2, index v1 - // v4 = not v0 - // enable_side_effects_if v4 - // v5 = array_get v2, index v1 - // } - // // We want to make sure after constant folding both array_gets remain since they are // under different enable_side_effects_if contexts and thus one may be disabled while // the other is not. If one is removed, it is possible e.g. v4 is replaced with v2 which // is disabled (only gets from index 0) and thus returns the wrong result. let src = " - acir(inline) fn main f0 { - b0(v0: u1, v1: u64): - enable_side_effects v0 - v4 = make_array [Field 0, Field 1] : [Field; 2] - v5 = array_get v4, index v1 -> Field - v6 = not v0 - enable_side_effects v6 - v7 = array_get v4, index v1 -> Field - return - } - "; + acir(inline) fn main f0 { + b0(v0: u1, v1: u64): + enable_side_effects v0 + v4 = make_array [Field 0, Field 1] : [Field; 2] + v5 = array_get v4, index v1 -> Field + v6 = not v0 + enable_side_effects v6 + v7 = array_get v4, index v1 -> Field + return + } + "; let ssa = Ssa::from_str(src).unwrap(); // Expected output is unchanged @@ -1032,7 +1141,6 @@ mod test { // v5 = call keccakf1600(v1) // v6 = call keccakf1600(v2) // } - // // Here we're checking a situation where two identical arrays are being initialized twice and being assigned separate `ValueId`s. // This would result in otherwise identical instructions not being deduplicated. let main_id = Id::test_new(0); @@ -1071,6 +1179,7 @@ mod test { // fn main f0 { // b0(v0: u64): // v1 = make_array [v0, u64 0, u64 0, u64 0, u64 0, u64 0, u64 0, u64 0, u64 0, u64 0, u64 0, u64 0, u64 0, u64 0, u64 0, u64 0, u64 0, u64 0, u64 0, u64 0, u64 0, u64 0, u64 0, u64 0, u64 0] + // inc_rc v1 // v5 = call keccakf1600(v1) // } let ssa = ssa.fold_constants(); @@ -1080,7 +1189,107 @@ mod test { let main = ssa.main(); let instructions = main.dfg[main.entry_block()].instructions(); let ending_instruction_count = instructions.len(); - assert_eq!(ending_instruction_count, 2); + assert_eq!(ending_instruction_count, 3); + } + + #[test] + fn deduplicate_across_blocks() { + // fn main f0 { + // b0(v0: u1): + // v1 = not v0 + // jmp b1() + // b1(): + // v2 = not v0 + // return v2 + // } + let main_id = Id::test_new(0); + + // Compiling main + let mut builder = FunctionBuilder::new("main".into(), main_id); + let b1 = builder.insert_block(); + + let v0 = builder.add_parameter(Type::bool()); + let _v1 = builder.insert_not(v0); + builder.terminate_with_jmp(b1, Vec::new()); + + builder.switch_to_block(b1); + let v2 = builder.insert_not(v0); + builder.terminate_with_return(vec![v2]); + + let ssa = builder.finish(); + let main = ssa.main(); + assert_eq!(main.dfg[main.entry_block()].instructions().len(), 1); + assert_eq!(main.dfg[b1].instructions().len(), 1); + + // Expected output: + // + // fn main f0 { + // b0(v0: u1): + // v1 = not v0 + // jmp b1() + // b1(): + // return v1 + // } + let ssa = ssa.fold_constants_using_constraints(); + let main = ssa.main(); + assert_eq!(main.dfg[main.entry_block()].instructions().len(), 1); + assert_eq!(main.dfg[b1].instructions().len(), 0); + } + + #[test] + fn deduplicate_across_non_dominated_blocks() { + let src = " + brillig(inline) fn main f0 { + b0(v0: u32): + v2 = lt u32 1000, v0 + jmpif v2 then: b1, else: b2 + b1(): + v4 = shl v0, u32 1 + v5 = lt v0, v4 + constrain v5 == u1 1 + jmp b2() + b2(): + v7 = lt u32 1000, v0 + jmpif v7 then: b3, else: b4 + b3(): + v8 = shl v0, u32 1 + v9 = lt v0, v8 + constrain v9 == u1 1 + jmp b4() + b4(): + return + } + "; + let ssa = Ssa::from_str(src).unwrap(); + + // v4 has been hoisted, although: + // - v5 has not yet been removed since it was encountered earlier in the program + // - v8 hasn't been recognized as a duplicate of v6 yet since they still reference v4 and + // v5 respectively + let expected = " + brillig(inline) fn main f0 { + b0(v0: u32): + v2 = lt u32 1000, v0 + v4 = shl v0, u32 1 + jmpif v2 then: b1, else: b2 + b1(): + v5 = shl v0, u32 1 + v6 = lt v0, v5 + constrain v6 == u1 1 + jmp b2() + b2(): + jmpif v2 then: b3, else: b4 + b3(): + v8 = lt v0, v4 + constrain v8 == u1 1 + jmp b4() + b4(): + return + } + "; + + let ssa = ssa.fold_constants_using_constraints(); + assert_normalized_ssa_equals(ssa, expected); } #[test] @@ -1260,46 +1469,158 @@ mod test { } #[test] - fn deduplicate_across_blocks() { - // fn main f0 { - // b0(v0: u1): - // v1 = not v0 - // jmp b1() - // b1(): - // v2 = not v0 - // return v2 - // } - let main_id = Id::test_new(0); + fn does_not_use_cached_constrain_in_block_that_is_not_dominated() { + let src = " + brillig(inline) fn main f0 { + b0(v0: Field, v1: Field): + v3 = eq v0, Field 0 + jmpif v3 then: b1, else: b2 + b1(): + v5 = eq v1, Field 1 + constrain v1 == Field 1 + jmp b2() + b2(): + v6 = eq v1, Field 0 + constrain v1 == Field 0 + return + } + "; + let ssa = Ssa::from_str(src).unwrap(); + let ssa = ssa.fold_constants_using_constraints(); + assert_normalized_ssa_equals(ssa, src); + } - // Compiling main - let mut builder = FunctionBuilder::new("main".into(), main_id); - let b1 = builder.insert_block(); + #[test] + fn does_not_hoist_constrain_to_common_ancestor() { + let src = " + brillig(inline) fn main f0 { + b0(v0: Field, v1: Field): + v3 = eq v0, Field 0 + jmpif v3 then: b1, else: b2 + b1(): + constrain v1 == Field 1 + jmp b2() + b2(): + jmpif v0 then: b3, else: b4 + b3(): + constrain v1 == Field 1 // This was incorrectly hoisted to b0 but this condition is not valid when going b0 -> b2 -> b4 + jmp b4() + b4(): + return + } + "; + let ssa = Ssa::from_str(src).unwrap(); + let ssa = ssa.fold_constants_using_constraints(); + assert_normalized_ssa_equals(ssa, src); + } - let v0 = builder.add_parameter(Type::bool()); - let _v1 = builder.insert_not(v0); - builder.terminate_with_jmp(b1, Vec::new()); + #[test] + fn does_not_hoist_sub_to_common_ancestor() { + let src = " + acir(inline) fn main f0 { + b0(v0: u32): + v2 = eq v0, u32 0 + jmpif v2 then: b4, else: b1 + b4(): + v5 = sub v0, u32 1 + jmp b5() + b5(): + return + b1(): + jmpif v0 then: b3, else: b2 + b3(): + v4 = sub v0, u32 1 // We can't hoist this because v0 is zero here and it will lead to an underflow + jmp b5() + b2(): + jmp b5() + } + "; + let ssa = Ssa::from_str(src).unwrap(); + let ssa = ssa.fold_constants_using_constraints(); + assert_normalized_ssa_equals(ssa, src); + } - builder.switch_to_block(b1); - let v2 = builder.insert_not(v0); - builder.terminate_with_return(vec![v2]); + #[test] + fn deduplicates_side_effecting_intrinsics() { + let src = " + // After EnableSideEffectsIf removal: + acir(inline) fn main f0 { + b0(v0: Field, v1: Field, v2: u1): + v4 = call is_unconstrained() -> u1 + v7 = call to_be_radix(v0, u32 256) -> [u8; 1] // `a.to_be_radix(256)`; + inc_rc v7 + v8 = call to_be_radix(v0, u32 256) -> [u8; 1] // duplicate load of `a` + inc_rc v8 + v9 = cast v2 as Field // `if c { a.to_be_radix(256) }` + v10 = mul v0, v9 // attaching `c` to `a` + v11 = call to_be_radix(v10, u32 256) -> [u8; 1] // calling `to_radix(c * a)` + inc_rc v11 + enable_side_effects v2 // side effect var for `c` shifted down by removal + return + } + "; + let ssa = Ssa::from_str(src).unwrap(); + let expected = " + acir(inline) fn main f0 { + b0(v0: Field, v1: Field, v2: u1): + v4 = call is_unconstrained() -> u1 + v7 = call to_be_radix(v0, u32 256) -> [u8; 1] + inc_rc v7 + inc_rc v7 + v8 = cast v2 as Field + v9 = mul v0, v8 + v10 = call to_be_radix(v9, u32 256) -> [u8; 1] + inc_rc v10 + enable_side_effects v2 + return + } + "; + let ssa = ssa.fold_constants_using_constraints(); + assert_normalized_ssa_equals(ssa, expected); + } - let ssa = builder.finish(); - let main = ssa.main(); - assert_eq!(main.dfg[main.entry_block()].instructions().len(), 1); - assert_eq!(main.dfg[b1].instructions().len(), 1); + #[test] + fn array_get_from_array_set_with_different_predicates() { + let src = " + acir(inline) fn main f0 { + b0(v0: [Field; 3], v1: u32, v2: Field): + enable_side_effects u1 0 + v4 = array_set v0, index v1, value v2 + enable_side_effects u1 1 + v6 = array_get v4, index v1 -> Field + return v6 + } + "; + + let ssa = Ssa::from_str(src).unwrap(); - // Expected output: - // - // fn main f0 { - // b0(v0: u1): - // v1 = not v0 - // jmp b1() - // b1(): - // return v1 - // } let ssa = ssa.fold_constants_using_constraints(); - let main = ssa.main(); - assert_eq!(main.dfg[main.entry_block()].instructions().len(), 1); - assert_eq!(main.dfg[b1].instructions().len(), 0); + // We expect the code to be unchanged + assert_normalized_ssa_equals(ssa, src); + } + + #[test] + fn array_get_from_array_set_same_predicates() { + let src = " + acir(inline) fn main f0 { + b0(v0: [Field; 3], v1: u32, v2: Field): + enable_side_effects u1 1 + v4 = array_set v0, index v1, value v2 + v6 = array_get v4, index v1 -> Field + return v6 + } + "; + let ssa = Ssa::from_str(src).unwrap(); + + let expected = " + acir(inline) fn main f0 { + b0(v0: [Field; 3], v1: u32, v2: Field): + enable_side_effects u1 1 + v4 = array_set v0, index v1, value v2 + return v2 + } + "; + let ssa = ssa.fold_constants_using_constraints(); + assert_normalized_ssa_equals(ssa, expected); } } diff --git a/noir/noir-repo/compiler/noirc_evaluator/src/ssa/opt/die.rs b/noir/noir-repo/compiler/noirc_evaluator/src/ssa/opt/die.rs index dbdb6e3b4ef..f7ac6f7b313 100644 --- a/noir/noir-repo/compiler/noirc_evaluator/src/ssa/opt/die.rs +++ b/noir/noir-repo/compiler/noirc_evaluator/src/ssa/opt/die.rs @@ -18,8 +18,6 @@ use crate::ssa::{ ssa_gen::{Ssa, SSA_WORD_SIZE}, }; -use super::rc::{pop_rc_for, RcInstruction}; - impl Ssa { /// Performs Dead Instruction Elimination (DIE) to remove any instructions with /// unused results. @@ -108,8 +106,6 @@ impl Context { let instructions_len = block.instructions().len(); - let mut rc_tracker = RcTracker::default(); - // Indexes of instructions that might be out of bounds. // We'll remove those, but before that we'll insert bounds checks for them. let mut possible_index_out_of_bounds_indexes = Vec::new(); @@ -127,8 +123,9 @@ impl Context { .push(instructions_len - instruction_index - 1); } } else { - use Instruction::*; - if matches!(instruction, IncrementRc { .. } | DecrementRc { .. }) { + // We can't remove rc instructions if they're loaded from a reference + // since we'd have no way of knowing whether the reference is still used. + if Self::is_inc_dec_instruction_on_known_array(instruction, &function.dfg) { self.rc_instructions.push((*instruction_id, block_id)); } else { instruction.for_each_value(|value| { @@ -136,13 +133,8 @@ impl Context { }); } } - - rc_tracker.track_inc_rcs_to_remove(*instruction_id, function); } - self.instructions_to_remove.extend(rc_tracker.get_non_mutated_arrays()); - self.instructions_to_remove.extend(rc_tracker.rc_pairs_to_remove); - // If there are some instructions that might trigger an out of bounds error, // first add constrain checks. Then run the DIE pass again, which will remove those // but leave the constrains (any any value needed by those constrains) @@ -346,6 +338,28 @@ impl Context { inserted_check } + + /// True if this is a `Instruction::IncrementRc` or `Instruction::DecrementRc` + /// operating on an array directly from a `Instruction::MakeArray` or an + /// intrinsic known to return a fresh array. + fn is_inc_dec_instruction_on_known_array( + instruction: &Instruction, + dfg: &DataFlowGraph, + ) -> bool { + use Instruction::*; + if let IncrementRc { value } | DecrementRc { value } = instruction { + if let Value::Instruction { instruction, .. } = &dfg[*value] { + return match &dfg[*instruction] { + MakeArray { .. } => true, + Call { func, .. } => { + matches!(&dfg[*func], Value::Intrinsic(_) | Value::ForeignFunction(_)) + } + _ => false, + }; + } + } + false + } } fn instruction_might_result_in_out_of_bounds( @@ -508,103 +522,6 @@ fn apply_side_effects( (lhs, rhs) } -#[derive(Default)] -struct RcTracker { - // We can track IncrementRc instructions per block to determine whether they are useless. - // IncrementRc and DecrementRc instructions are normally side effectual instructions, but we remove - // them if their value is not used anywhere in the function. However, even when their value is used, their existence - // is pointless logic if there is no array set between the increment and the decrement of the reference counter. - // We track per block whether an IncrementRc instruction has a paired DecrementRc instruction - // with the same value but no array set in between. - // If we see an inc/dec RC pair within a block we can safely remove both instructions. - rcs_with_possible_pairs: HashMap>, - rc_pairs_to_remove: HashSet, - // We also separately track all IncrementRc instructions and all arrays which have been mutably borrowed. - // If an array has not been mutably borrowed we can then safely remove all IncrementRc instructions on that array. - inc_rcs: HashMap>, - mut_borrowed_arrays: HashSet, - // The SSA often creates patterns where after simplifications we end up with repeat - // IncrementRc instructions on the same value. We track whether the previous instruction was an IncrementRc, - // and if the current instruction is also an IncrementRc on the same value we remove the current instruction. - // `None` if the previous instruction was anything other than an IncrementRc - previous_inc_rc: Option, -} - -impl RcTracker { - fn track_inc_rcs_to_remove(&mut self, instruction_id: InstructionId, function: &Function) { - let instruction = &function.dfg[instruction_id]; - - if let Instruction::IncrementRc { value } = instruction { - if let Some(previous_value) = self.previous_inc_rc { - if previous_value == *value { - self.rc_pairs_to_remove.insert(instruction_id); - } - } - self.previous_inc_rc = Some(*value); - } else { - self.previous_inc_rc = None; - } - - // DIE loops over a block in reverse order, so we insert an RC instruction for possible removal - // when we see a DecrementRc and check whether it was possibly mutated when we see an IncrementRc. - match instruction { - Instruction::IncrementRc { value } => { - if let Some(inc_rc) = - pop_rc_for(*value, function, &mut self.rcs_with_possible_pairs) - { - if !inc_rc.possibly_mutated { - self.rc_pairs_to_remove.insert(inc_rc.id); - self.rc_pairs_to_remove.insert(instruction_id); - } - } - - self.inc_rcs.entry(*value).or_default().insert(instruction_id); - } - Instruction::DecrementRc { value } => { - let typ = function.dfg.type_of_value(*value); - - // We assume arrays aren't mutated until we find an array_set - let dec_rc = - RcInstruction { id: instruction_id, array: *value, possibly_mutated: false }; - self.rcs_with_possible_pairs.entry(typ).or_default().push(dec_rc); - } - Instruction::ArraySet { array, .. } => { - let typ = function.dfg.type_of_value(*array); - if let Some(dec_rcs) = self.rcs_with_possible_pairs.get_mut(&typ) { - for dec_rc in dec_rcs { - dec_rc.possibly_mutated = true; - } - } - - self.mut_borrowed_arrays.insert(*array); - } - Instruction::Store { value, .. } => { - // We are very conservative and say that any store of an array value means it has the potential - // to be mutated. This is done due to the tracking of mutable borrows still being per block. - let typ = function.dfg.type_of_value(*value); - if matches!(&typ, Type::Array(..) | Type::Slice(..)) { - self.mut_borrowed_arrays.insert(*value); - } - } - _ => {} - } - } - - fn get_non_mutated_arrays(&self) -> HashSet { - self.inc_rcs - .keys() - .filter_map(|value| { - if !self.mut_borrowed_arrays.contains(value) { - Some(&self.inc_rcs[value]) - } else { - None - } - }) - .flatten() - .copied() - .collect() - } -} #[cfg(test)] mod test { use std::sync::Arc; @@ -613,7 +530,7 @@ mod test { use crate::ssa::{ function_builder::FunctionBuilder, - ir::{instruction::Instruction, map::Id, types::Type}, + ir::{map::Id, types::Type}, opt::assert_normalized_ssa_equals, Ssa, }; @@ -685,30 +602,6 @@ mod test { assert_normalized_ssa_equals(ssa, expected); } - #[test] - fn remove_useless_paired_rcs_even_when_used() { - let src = " - acir(inline) fn main f0 { - b0(v0: [Field; 2]): - inc_rc v0 - v2 = array_get v0, index u32 0 -> Field - dec_rc v0 - return v2 - } - "; - let ssa = Ssa::from_str(src).unwrap(); - - let expected = " - acir(inline) fn main f0 { - b0(v0: [Field; 2]): - v2 = array_get v0, index u32 0 -> Field - return v2 - } - "; - let ssa = ssa.dead_instruction_elimination(); - assert_normalized_ssa_equals(ssa, expected); - } - #[test] fn keep_paired_rcs_with_array_set() { let src = " @@ -779,92 +672,23 @@ mod test { } #[test] - fn keep_inc_rc_on_borrowed_array_set() { - // acir(inline) fn main f0 { - // b0(v0: [u32; 2]): - // inc_rc v0 - // v3 = array_set v0, index u32 0, value u32 1 - // inc_rc v0 - // inc_rc v0 - // inc_rc v0 - // v4 = array_get v3, index u32 1 - // return v4 - // } - let main_id = Id::test_new(0); - - // Compiling main - let mut builder = FunctionBuilder::new("main".into(), main_id); - let array_type = Type::Array(Arc::new(vec![Type::unsigned(32)]), 2); - let v0 = builder.add_parameter(array_type.clone()); - builder.increment_array_reference_count(v0); - let zero = builder.numeric_constant(0u128, Type::unsigned(32)); - let one = builder.numeric_constant(1u128, Type::unsigned(32)); - let v3 = builder.insert_array_set(v0, zero, one); - builder.increment_array_reference_count(v0); - builder.increment_array_reference_count(v0); - builder.increment_array_reference_count(v0); - - let v4 = builder.insert_array_get(v3, one, Type::unsigned(32)); - - builder.terminate_with_return(vec![v4]); - - let ssa = builder.finish(); - let main = ssa.main(); - - // The instruction count never includes the terminator instruction - assert_eq!(main.dfg[main.entry_block()].instructions().len(), 6); - - // We expect the output to be unchanged - // Expected output: - // - // acir(inline) fn main f0 { - // b0(v0: [u32; 2]): - // inc_rc v0 - // v3 = array_set v0, index u32 0, value u32 1 - // inc_rc v0 - // v4 = array_get v3, index u32 1 - // return v4 - // } - let ssa = ssa.dead_instruction_elimination(); - let main = ssa.main(); - - let instructions = main.dfg[main.entry_block()].instructions(); - // We expect only the repeated inc_rc instructions to be collapsed into a single inc_rc. - assert_eq!(instructions.len(), 4); - - assert!(matches!(&main.dfg[instructions[0]], Instruction::IncrementRc { .. })); - assert!(matches!(&main.dfg[instructions[1]], Instruction::ArraySet { .. })); - assert!(matches!(&main.dfg[instructions[2]], Instruction::IncrementRc { .. })); - assert!(matches!(&main.dfg[instructions[3]], Instruction::ArrayGet { .. })); - } - - #[test] - fn remove_inc_rcs_that_are_never_mutably_borrowed() { + fn does_not_remove_inc_or_dec_rc_of_if_they_are_loaded_from_a_reference() { let src = " - acir(inline) fn main f0 { - b0(v0: [Field; 2]): - inc_rc v0 - inc_rc v0 - inc_rc v0 - v2 = array_get v0, index u32 0 -> Field - inc_rc v0 - return v2 + brillig(inline) fn borrow_mut f0 { + b0(v0: &mut [Field; 3]): + v1 = load v0 -> [Field; 3] + inc_rc v1 // this one shouldn't be removed + v2 = load v0 -> [Field; 3] + inc_rc v2 // this one shouldn't be removed + v3 = load v0 -> [Field; 3] + v6 = array_set v3, index u32 0, value Field 5 + store v6 at v0 + dec_rc v6 + return } "; let ssa = Ssa::from_str(src).unwrap(); - let main = ssa.main(); - - // The instruction count never includes the terminator instruction - assert_eq!(main.dfg[main.entry_block()].instructions().len(), 5); - - let expected = " - acir(inline) fn main f0 { - b0(v0: [Field; 2]): - v2 = array_get v0, index u32 0 -> Field - return v2 - } - "; let ssa = ssa.dead_instruction_elimination(); - assert_normalized_ssa_equals(ssa, expected); + assert_normalized_ssa_equals(ssa, src); } } diff --git a/noir/noir-repo/compiler/noirc_evaluator/src/ssa/opt/flatten_cfg.rs b/noir/noir-repo/compiler/noirc_evaluator/src/ssa/opt/flatten_cfg.rs index 5d114672a55..3fbccf93ec9 100644 --- a/noir/noir-repo/compiler/noirc_evaluator/src/ssa/opt/flatten_cfg.rs +++ b/noir/noir-repo/compiler/noirc_evaluator/src/ssa/opt/flatten_cfg.rs @@ -131,8 +131,7 @@ //! v11 = mul v4, Field 12 //! v12 = add v10, v11 //! store v12 at v5 (new store) -use fxhash::FxHashMap as HashMap; -use std::collections::{BTreeMap, HashSet}; +use fxhash::{FxHashMap as HashMap, FxHashSet as HashSet}; use acvm::{acir::AcirField, acir::BlackBoxFunc, FieldElement}; use iter_extended::vecmap; @@ -186,18 +185,6 @@ struct Context<'f> { /// Maps start of branch -> end of branch branch_ends: HashMap, - /// Maps an address to the old and new value of the element at that address - /// These only hold stores for one block at a time and is cleared - /// between inlining of branches. - store_values: HashMap, - - /// Stores all allocations local to the current branch. - /// Since these branches are local to the current branch (ie. only defined within one branch of - /// an if expression), they should not be merged with their previous value or stored value in - /// the other branch since there is no such value. The ValueId here is that which is returned - /// by the allocate instruction. - local_allocations: HashSet, - /// A stack of each jmpif condition that was taken to reach a particular point in the program. /// When two branches are merged back into one, this constitutes a join point, and is analogous /// to the rest of the program after an if statement. When such a join point / end block is @@ -214,13 +201,15 @@ struct Context<'f> { /// When processing a block, we pop this stack to get its arguments /// and at the end we push the arguments for his successor arguments_stack: Vec>, -} -#[derive(Clone)] -pub(crate) struct Store { - old_value: ValueId, - new_value: ValueId, - call_stack: CallStack, + /// Stores all allocations local to the current branch. + /// + /// Since these branches are local to the current branch (i.e. only defined within one branch of + /// an if expression), they should not be merged with their previous value or stored value in + /// the other branch since there is no such value. + /// + /// The `ValueId` here is that which is returned by the allocate instruction. + local_allocations: HashSet, } #[derive(Clone)] @@ -231,8 +220,6 @@ struct ConditionalBranch { old_condition: ValueId, // The condition of the branch condition: ValueId, - // The store values accumulated when processing the branch - store_values: HashMap, // The allocations accumulated when processing the branch local_allocations: HashSet, } @@ -263,12 +250,11 @@ fn flatten_function_cfg(function: &mut Function, no_predicates: &HashMap Context<'f> { let old_condition = *condition; let then_condition = self.inserter.resolve(old_condition); - let old_stores = std::mem::take(&mut self.store_values); let old_allocations = std::mem::take(&mut self.local_allocations); let branch = ConditionalBranch { old_condition, condition: self.link_condition(then_condition), - store_values: old_stores, - local_allocations: old_allocations, last_block: *then_destination, + local_allocations: old_allocations, }; let cond_context = ConditionalContext { condition: then_condition, @@ -473,19 +457,12 @@ impl<'f> Context<'f> { ); let else_condition = self.link_condition(else_condition); - // Make sure the else branch sees the previous values of each store - // rather than any values created in the 'then' branch. - let old_stores = std::mem::take(&mut cond_context.then_branch.store_values); - cond_context.then_branch.store_values = std::mem::take(&mut self.store_values); - self.undo_stores_in_then_branch(&cond_context.then_branch.store_values); - let old_allocations = std::mem::take(&mut self.local_allocations); let else_branch = ConditionalBranch { old_condition: cond_context.then_branch.old_condition, condition: else_condition, - store_values: old_stores, - local_allocations: old_allocations, last_block: *block, + local_allocations: old_allocations, }; cond_context.then_branch.local_allocations.clear(); cond_context.else_branch = Some(else_branch); @@ -509,10 +486,8 @@ impl<'f> Context<'f> { } let mut else_branch = cond_context.else_branch.unwrap(); - let stores_in_branch = std::mem::replace(&mut self.store_values, else_branch.store_values); self.local_allocations = std::mem::take(&mut else_branch.local_allocations); else_branch.last_block = *block; - else_branch.store_values = stores_in_branch; cond_context.else_branch = Some(else_branch); // We must remember to reset whether side effects are enabled when both branches @@ -562,7 +537,11 @@ impl<'f> Context<'f> { let args = vecmap(then_args.iter().zip(else_args), |(then_arg, else_arg)| { (self.inserter.resolve(*then_arg), self.inserter.resolve(else_arg)) }); - + let else_condition = if let Some(branch) = cond_context.else_branch { + branch.condition + } else { + self.inserter.function.dfg.make_constant(FieldElement::zero(), Type::bool()) + }; let block = self.inserter.function.entry_block(); // Cannot include this in the previous vecmap since it requires exclusive access to self @@ -570,6 +549,7 @@ impl<'f> Context<'f> { let instruction = Instruction::IfElse { then_condition: cond_context.then_branch.condition, then_value: then_arg, + else_condition, else_value: else_arg, }; let call_stack = cond_context.call_stack.clone(); @@ -580,8 +560,6 @@ impl<'f> Context<'f> { .first() }); - let call_stack = cond_context.call_stack; - self.merge_stores(cond_context.then_branch, cond_context.else_branch, call_stack); self.arguments_stack.pop(); self.arguments_stack.pop(); self.arguments_stack.push(args); @@ -636,116 +614,29 @@ impl<'f> Context<'f> { self.insert_instruction_with_typevars(enable_side_effects, None, call_stack); } - /// Merge any store instructions found in each branch. - /// - /// This function relies on the 'then' branch being merged before the 'else' branch of a jmpif - /// instruction. If this ordering is changed, the ordering that store values are merged within - /// this function also needs to be changed to reflect that. - fn merge_stores( - &mut self, - then_branch: ConditionalBranch, - else_branch: Option, - call_stack: CallStack, - ) { - // Address -> (then_value, else_value, value_before_the_if) - let mut new_map = BTreeMap::new(); - - for (address, store) in then_branch.store_values { - new_map.insert(address, (store.new_value, store.old_value, store.old_value)); - } - - if else_branch.is_some() { - for (address, store) in else_branch.clone().unwrap().store_values { - if let Some(entry) = new_map.get_mut(&address) { - entry.1 = store.new_value; - } else { - new_map.insert(address, (store.old_value, store.new_value, store.old_value)); - } - } - } - - let then_condition = then_branch.condition; - let block = self.inserter.function.entry_block(); - - // Merging must occur in a separate loop as we cannot borrow `self` as mutable while `value_merger` does - let mut new_values = HashMap::default(); - for (address, (then_case, else_case, _)) in &new_map { - let instruction = Instruction::IfElse { - then_condition, - then_value: *then_case, - else_value: *else_case, - }; - let dfg = &mut self.inserter.function.dfg; - let value = dfg - .insert_instruction_and_results(instruction, block, None, call_stack.clone()) - .first(); - - new_values.insert(address, value); - } - - // Replace stores with new merged values - for (address, (_, _, old_value)) in &new_map { - let value = new_values[address]; - let address = *address; - self.insert_instruction_with_typevars( - Instruction::Store { address, value }, - None, - call_stack.clone(), - ); - - if let Some(store) = self.store_values.get_mut(&address) { - store.new_value = value; - } else { - self.store_values.insert( - address, - Store { - old_value: *old_value, - new_value: value, - call_stack: call_stack.clone(), - }, - ); - } - } - } - - fn remember_store(&mut self, address: ValueId, new_value: ValueId, call_stack: CallStack) { - if !self.local_allocations.contains(&address) { - if let Some(store_value) = self.store_values.get_mut(&address) { - store_value.new_value = new_value; - } else { - let load = Instruction::Load { address }; - - let load_type = Some(vec![self.inserter.function.dfg.type_of_value(new_value)]); - let old_value = self - .insert_instruction_with_typevars(load.clone(), load_type, call_stack.clone()) - .first(); - - self.store_values.insert(address, Store { old_value, new_value, call_stack }); - } - } - } - /// Push the given instruction to the end of the entry block of the current function. /// /// Note that each ValueId of the instruction will be mapped via self.inserter.resolve. /// As a result, the instruction that will be pushed will actually be a new instruction /// with a different InstructionId from the original. The results of the given instruction /// will also be mapped to the results of the new instruction. - fn push_instruction(&mut self, id: InstructionId) -> Vec { + /// + /// `previous_allocate_result` should only be set to the result of an allocate instruction + /// if that instruction was the instruction immediately previous to this one - if there are + /// any instructions in between it should be None. + fn push_instruction(&mut self, id: InstructionId) { let (instruction, call_stack) = self.inserter.map_instruction(id); let instruction = self.handle_instruction_side_effects(instruction, call_stack.clone()); - let is_allocate = matches!(instruction, Instruction::Allocate); + let instruction_is_allocate = matches!(&instruction, Instruction::Allocate); let entry = self.inserter.function.entry_block(); let results = self.inserter.push_instruction_value(instruction, id, entry, call_stack); // Remember an allocate was created local to this branch so that we do not try to merge store // values across branches for it later. - if is_allocate { + if instruction_is_allocate { self.local_allocations.insert(results.first()); } - - results.results().into_owned() } /// If we are currently in a branch, we need to modify constrain instructions @@ -782,8 +673,35 @@ impl<'f> Context<'f> { Instruction::Constrain(lhs, rhs, message) } Instruction::Store { address, value } => { - self.remember_store(address, value, call_stack); - Instruction::Store { address, value } + // If this instruction immediately follows an allocate, and stores to that + // address there is no previous value to load and we don't need a merge anyway. + if self.local_allocations.contains(&address) { + Instruction::Store { address, value } + } else { + // Instead of storing `value`, store `if condition { value } else { previous_value }` + let typ = self.inserter.function.dfg.type_of_value(value); + let load = Instruction::Load { address }; + let previous_value = self + .insert_instruction_with_typevars( + load, + Some(vec![typ]), + call_stack.clone(), + ) + .first(); + + let else_condition = self + .insert_instruction(Instruction::Not(condition), call_stack.clone()); + + let instruction = Instruction::IfElse { + then_condition: condition, + then_value: value, + else_condition, + else_value: previous_value, + }; + + let updated_value = self.insert_instruction(instruction, call_stack); + Instruction::Store { address, value: updated_value } + } } Instruction::RangeCheck { value, max_bit_size, assert_message } => { // Replace value with `value * predicate` to zero out value when predicate is inactive. @@ -905,23 +823,11 @@ impl<'f> Context<'f> { call_stack, ) } - - fn undo_stores_in_then_branch(&mut self, store_values: &HashMap) { - for (address, store) in store_values { - let address = *address; - let value = store.old_value; - let instruction = Instruction::Store { address, value }; - // Considering the location of undoing a store to be the same as the original store. - self.insert_instruction_with_typevars(instruction, None, store.call_stack.clone()); - } - } } #[cfg(test)] mod test { - use std::sync::Arc; - - use acvm::{acir::AcirField, FieldElement}; + use acvm::acir::AcirField; use crate::ssa::{ function_builder::FunctionBuilder, @@ -961,9 +867,11 @@ mod test { v1 = not v0 enable_side_effects u1 1 v3 = cast v0 as Field - v5 = mul v3, Field -1 - v7 = add Field 4, v5 - return v7 + v4 = cast v1 as Field + v6 = mul v3, Field 3 + v8 = mul v4, Field 4 + v9 = add v6, v8 + return v9 } "; @@ -1023,15 +931,15 @@ mod test { b0(v0: u1, v1: &mut Field): enable_side_effects v0 v2 = load v1 -> Field - store Field 5 at v1 - v4 = not v0 - store v2 at v1 - enable_side_effects u1 1 - v6 = cast v0 as Field - v7 = sub Field 5, v2 - v8 = mul v6, v7 - v9 = add v2, v8 + v3 = not v0 + v4 = cast v0 as Field + v5 = cast v3 as Field + v7 = mul v4, Field 5 + v8 = mul v5, v2 + v9 = add v7, v8 store v9 at v1 + v10 = not v0 + enable_side_effects u1 1 return } "; @@ -1062,17 +970,23 @@ mod test { b0(v0: u1, v1: &mut Field): enable_side_effects v0 v2 = load v1 -> Field - store Field 5 at v1 - v4 = not v0 - store v2 at v1 - enable_side_effects v4 - v5 = load v1 -> Field - store Field 6 at v1 + v3 = not v0 + v4 = cast v0 as Field + v5 = cast v3 as Field + v7 = mul v4, Field 5 + v8 = mul v5, v2 + v9 = add v7, v8 + store v9 at v1 + v10 = not v0 + enable_side_effects v10 + v11 = load v1 -> Field + v12 = cast v10 as Field + v13 = cast v0 as Field + v15 = mul v12, Field 6 + v16 = mul v13, v11 + v17 = add v15, v16 + store v17 at v1 enable_side_effects u1 1 - v8 = cast v0 as Field - v10 = mul v8, Field -1 - v11 = add Field 6, v10 - store v11 at v1 return } "; @@ -1115,6 +1029,7 @@ mod test { // b7 b8 // ↘ ↙ // b9 + let src = " acir(inline) fn main f0 { b0(v0: u1, v1: u1): @@ -1165,33 +1080,49 @@ mod test { v20 = load v2 -> Field // call v1(Field 8, v20) jmp b9() - } - "; + }"; let ssa = Ssa::from_str(src).unwrap(); + let ssa = ssa.flatten_cfg().mem2reg(); - // Expected results after mem2reg removes the allocation and each load and store: let expected = " acir(inline) fn main f0 { b0(v0: u1, v1: u1): v2 = allocate -> &mut Field enable_side_effects v0 - v3 = mul v0, v1 - enable_side_effects v3 - v4 = not v1 - v5 = mul v0, v4 + v3 = not v0 + v4 = cast v0 as Field + v5 = cast v3 as Field + v7 = mul v4, Field 2 + v8 = add v7, v5 + v9 = mul v0, v1 + enable_side_effects v9 + v10 = not v9 + v11 = cast v9 as Field + v12 = cast v10 as Field + v14 = mul v11, Field 5 + v15 = mul v12, v8 + v16 = add v14, v15 + v17 = not v1 + v18 = mul v0, v17 + enable_side_effects v18 + v19 = not v18 + v20 = cast v18 as Field + v21 = cast v19 as Field + v23 = mul v20, Field 6 + v24 = mul v21, v16 + v25 = add v23, v24 enable_side_effects v0 - v6 = cast v3 as Field - v8 = mul v6, Field -1 - v10 = add Field 6, v8 - v11 = not v0 + v26 = not v0 + enable_side_effects v26 + v27 = cast v26 as Field + v28 = cast v0 as Field + v30 = mul v27, Field 3 + v31 = mul v28, v25 + v32 = add v30, v31 enable_side_effects u1 1 - v13 = cast v0 as Field - v15 = sub v10, Field 3 - v16 = mul v13, v15 - v17 = add Field 3, v16 - return v17 + return v32 }"; let main = ssa.main(); @@ -1201,10 +1132,7 @@ mod test { }; let merged_values = get_all_constants_reachable_from_instruction(&main.dfg, ret); - assert_eq!( - merged_values, - vec![FieldElement::from(3u128), FieldElement::from(6u128), -FieldElement::from(1u128)] - ); + assert_eq!(merged_values, vec![2, 3, 5, 6]); assert_normalized_ssa_equals(ssa, expected); } @@ -1287,7 +1215,7 @@ mod test { fn get_all_constants_reachable_from_instruction( dfg: &DataFlowGraph, value: ValueId, - ) -> Vec { + ) -> Vec { match dfg[value] { Value::Instruction { instruction, .. } => { let mut values = vec![]; @@ -1305,7 +1233,7 @@ mod test { values.dedup(); values } - Value::NumericConstant { constant, .. } => vec![constant], + Value::NumericConstant { constant, .. } => vec![constant.to_u128()], _ => Vec::new(), } } @@ -1344,63 +1272,71 @@ mod test { fn should_not_merge_incorrectly_to_false() { // Regression test for #1792 // Tests that it does not simplify a true constraint an always-false constraint - // acir(inline) fn main f1 { - // b0(v0: [u8; 2]): - // v5 = array_get v0, index u8 0 - // v6 = cast v5 as u32 - // v8 = truncate v6 to 1 bits, max_bit_size: 32 - // v9 = cast v8 as u1 - // v10 = allocate - // store u8 0 at v10 - // jmpif v9 then: b2, else: b3 - // b2(): - // v12 = cast v5 as Field - // v13 = add v12, Field 1 - // store v13 at v10 - // jmp b4() - // b4(): - // constrain v9 == u1 1 - // return - // b3(): - // store u8 0 at v10 - // jmp b4() - // } - let main_id = Id::test_new(1); - let mut builder = FunctionBuilder::new("main".into(), main_id); - builder.insert_block(); // b0 - let b1 = builder.insert_block(); - let b2 = builder.insert_block(); - let b3 = builder.insert_block(); - let element_type = Arc::new(vec![Type::unsigned(8)]); - let array_type = Type::Array(element_type.clone(), 2); - let array = builder.add_parameter(array_type); - let zero = builder.numeric_constant(0_u128, Type::unsigned(8)); - let v5 = builder.insert_array_get(array, zero, Type::unsigned(8)); - let v6 = builder.insert_cast(v5, Type::unsigned(32)); - let i_two = builder.numeric_constant(2_u128, Type::unsigned(32)); - let v8 = builder.insert_binary(v6, BinaryOp::Mod, i_two); - let v9 = builder.insert_cast(v8, Type::bool()); - let v10 = builder.insert_allocate(Type::field()); - builder.insert_store(v10, zero); - builder.terminate_with_jmpif(v9, b1, b2); - builder.switch_to_block(b1); - let one = builder.field_constant(1_u128); - let v5b = builder.insert_cast(v5, Type::field()); - let v13: Id = builder.insert_binary(v5b, BinaryOp::Add, one); - let v14 = builder.insert_cast(v13, Type::unsigned(8)); - builder.insert_store(v10, v14); - builder.terminate_with_jmp(b3, vec![]); - builder.switch_to_block(b2); - builder.insert_store(v10, zero); - builder.terminate_with_jmp(b3, vec![]); - builder.switch_to_block(b3); - let v_true = builder.numeric_constant(true, Type::bool()); - let v12 = builder.insert_binary(v9, BinaryOp::Eq, v_true); - builder.insert_constrain(v12, v_true, None); - builder.terminate_with_return(vec![]); - let ssa = builder.finish(); + + let src = " + acir(inline) fn main f0 { + b0(v0: [u8; 2]): + v2 = array_get v0, index u8 0 -> u8 + v3 = cast v2 as u32 + v4 = truncate v3 to 1 bits, max_bit_size: 32 + v5 = cast v4 as u1 + v6 = allocate -> &mut Field + store u8 0 at v6 + jmpif v5 then: b2, else: b1 + b2(): + v7 = cast v2 as Field + v9 = add v7, Field 1 + v10 = cast v9 as u8 + store v10 at v6 + jmp b3() + b3(): + constrain v5 == u1 1 + return + b1(): + store u8 0 at v6 + jmp b3() + } + "; + + let ssa = Ssa::from_str(src).unwrap(); + + let expected = " + acir(inline) fn main f0 { + b0(v0: [u8; 2]): + v2 = array_get v0, index u8 0 -> u8 + v3 = cast v2 as u32 + v4 = truncate v3 to 1 bits, max_bit_size: 32 + v5 = cast v4 as u1 + v6 = allocate -> &mut Field + store u8 0 at v6 + enable_side_effects v5 + v7 = cast v2 as Field + v9 = add v7, Field 1 + v10 = cast v9 as u8 + v11 = load v6 -> u8 + v12 = not v5 + v13 = cast v4 as u8 + v14 = cast v12 as u8 + v15 = mul v13, v10 + v16 = mul v14, v11 + v17 = add v15, v16 + store v17 at v6 + v18 = not v5 + enable_side_effects v18 + v19 = load v6 -> u8 + v20 = cast v18 as u8 + v21 = cast v4 as u8 + v22 = mul v21, v19 + store v22 at v6 + enable_side_effects u1 1 + constrain v5 == u1 1 + return + } + "; + let flattened_ssa = ssa.flatten_cfg(); let main = flattened_ssa.main(); + // Now assert that there is not an always-false constraint after flattening: let mut constrain_count = 0; for instruction in main.dfg[main.entry_block()].instructions() { @@ -1414,6 +1350,8 @@ mod test { } } assert_eq!(constrain_count, 1); + + assert_normalized_ssa_equals(flattened_ssa, expected); } #[test] @@ -1549,7 +1487,7 @@ mod test { b2(): return b1(): - jmp b2() + jmp b2() } "; let merged_ssa = Ssa::from_str(src).unwrap(); diff --git a/noir/noir-repo/compiler/noirc_evaluator/src/ssa/opt/flatten_cfg/capacity_tracker.rs b/noir/noir-repo/compiler/noirc_evaluator/src/ssa/opt/flatten_cfg/capacity_tracker.rs index ddc8b0bfe6b..a01be691778 100644 --- a/noir/noir-repo/compiler/noirc_evaluator/src/ssa/opt/flatten_cfg/capacity_tracker.rs +++ b/noir/noir-repo/compiler/noirc_evaluator/src/ssa/opt/flatten_cfg/capacity_tracker.rs @@ -21,7 +21,7 @@ impl<'a> SliceCapacityTracker<'a> { pub(crate) fn collect_slice_information( &self, instruction: &Instruction, - slice_sizes: &mut HashMap, + slice_sizes: &mut HashMap, results: &[ValueId], ) { match instruction { @@ -106,13 +106,12 @@ impl<'a> SliceCapacityTracker<'a> { Intrinsic::ToBits(_) => { // Compiler sanity check assert!(matches!(self.dfg.type_of_value(result_slice), Type::Slice(_))); - slice_sizes.insert(result_slice, FieldElement::max_num_bits() as usize); + slice_sizes.insert(result_slice, FieldElement::max_num_bits()); } Intrinsic::ToRadix(_) => { // Compiler sanity check assert!(matches!(self.dfg.type_of_value(result_slice), Type::Slice(_))); - slice_sizes - .insert(result_slice, FieldElement::max_num_bytes() as usize); + slice_sizes.insert(result_slice, FieldElement::max_num_bytes()); } Intrinsic::AsSlice => { let array_size = self @@ -157,7 +156,7 @@ impl<'a> SliceCapacityTracker<'a> { pub(crate) fn compute_slice_capacity( &self, array_id: ValueId, - slice_sizes: &mut HashMap, + slice_sizes: &mut HashMap, ) { if let Some((array, typ)) = self.dfg.get_array_constant(array_id) { // Compiler sanity check @@ -165,7 +164,7 @@ impl<'a> SliceCapacityTracker<'a> { if let Type::Slice(_) = typ { let element_size = typ.element_size(); let len = array.len() / element_size; - slice_sizes.insert(array_id, len); + slice_sizes.insert(array_id, len as u32); } } } diff --git a/noir/noir-repo/compiler/noirc_evaluator/src/ssa/opt/flatten_cfg/value_merger.rs b/noir/noir-repo/compiler/noirc_evaluator/src/ssa/opt/flatten_cfg/value_merger.rs index 8ea26d4e96d..6ea235b9414 100644 --- a/noir/noir-repo/compiler/noirc_evaluator/src/ssa/opt/flatten_cfg/value_merger.rs +++ b/noir/noir-repo/compiler/noirc_evaluator/src/ssa/opt/flatten_cfg/value_merger.rs @@ -17,7 +17,7 @@ pub(crate) struct ValueMerger<'a> { // Maps SSA array values with a slice type to their size. // This must be computed before merging values. - slice_sizes: &'a mut HashMap, + slice_sizes: &'a mut HashMap, array_set_conditionals: &'a mut HashMap, @@ -28,7 +28,7 @@ impl<'a> ValueMerger<'a> { pub(crate) fn new( dfg: &'a mut DataFlowGraph, block: BasicBlockId, - slice_sizes: &'a mut HashMap, + slice_sizes: &'a mut HashMap, array_set_conditionals: &'a mut HashMap, current_condition: Option, call_stack: CallStack, @@ -54,6 +54,7 @@ impl<'a> ValueMerger<'a> { pub(crate) fn merge_values( &mut self, then_condition: ValueId, + else_condition: ValueId, then_value: ValueId, else_value: ValueId, ) -> ValueId { @@ -69,14 +70,15 @@ impl<'a> ValueMerger<'a> { self.dfg, self.block, then_condition, + else_condition, then_value, else_value, ), typ @ Type::Array(_, _) => { - self.merge_array_values(typ, then_condition, then_value, else_value) + self.merge_array_values(typ, then_condition, else_condition, then_value, else_value) } typ @ Type::Slice(_) => { - self.merge_slice_values(typ, then_condition, then_value, else_value) + self.merge_slice_values(typ, then_condition, else_condition, then_value, else_value) } Type::Reference(_) => panic!("Cannot return references from an if expression"), Type::Function => panic!("Cannot return functions from an if expression"), @@ -84,11 +86,12 @@ impl<'a> ValueMerger<'a> { } /// Merge two numeric values a and b from separate basic blocks to a single value. This - /// function would return the result of `if c { a } else { b }` as `c * (a-b) + b`. + /// function would return the result of `if c { a } else { b }` as `c*a + (!c)*b`. pub(crate) fn merge_numeric_values( dfg: &mut DataFlowGraph, block: BasicBlockId, then_condition: ValueId, + else_condition: ValueId, then_value: ValueId, else_value: ValueId, ) -> ValueId { @@ -111,38 +114,31 @@ impl<'a> ValueMerger<'a> { // We must cast the bool conditions to the actual numeric type used by each value. let then_condition = dfg .insert_instruction_and_results( - Instruction::Cast(then_condition, Type::field()), + Instruction::Cast(then_condition, then_type), block, None, call_stack.clone(), ) .first(); - - let then_field = Instruction::Cast(then_value, Type::field()); - let then_field_value = - dfg.insert_instruction_and_results(then_field, block, None, call_stack.clone()).first(); - - let else_field = Instruction::Cast(else_value, Type::field()); - let else_field_value = - dfg.insert_instruction_and_results(else_field, block, None, call_stack.clone()).first(); - - let diff = Instruction::binary(BinaryOp::Sub, then_field_value, else_field_value); - let diff_value = - dfg.insert_instruction_and_results(diff, block, None, call_stack.clone()).first(); - - let conditional_diff = Instruction::binary(BinaryOp::Mul, then_condition, diff_value); - let conditional_diff_value = dfg - .insert_instruction_and_results(conditional_diff, block, None, call_stack.clone()) + let else_condition = dfg + .insert_instruction_and_results( + Instruction::Cast(else_condition, else_type), + block, + None, + call_stack.clone(), + ) .first(); - let merged_field = - Instruction::binary(BinaryOp::Add, else_field_value, conditional_diff_value); - let merged_field_value = dfg - .insert_instruction_and_results(merged_field, block, None, call_stack.clone()) - .first(); + let mul = Instruction::binary(BinaryOp::Mul, then_condition, then_value); + let then_value = + dfg.insert_instruction_and_results(mul, block, None, call_stack.clone()).first(); + + let mul = Instruction::binary(BinaryOp::Mul, else_condition, else_value); + let else_value = + dfg.insert_instruction_and_results(mul, block, None, call_stack.clone()).first(); - let merged = Instruction::Cast(merged_field_value, then_type); - dfg.insert_instruction_and_results(merged, block, None, call_stack).first() + let add = Instruction::binary(BinaryOp::Add, then_value, else_value); + dfg.insert_instruction_and_results(add, block, None, call_stack).first() } /// Given an if expression that returns an array: `if c { array1 } else { array2 }`, @@ -152,6 +148,7 @@ impl<'a> ValueMerger<'a> { &mut self, typ: Type, then_condition: ValueId, + else_condition: ValueId, then_value: ValueId, else_value: ValueId, ) -> ValueId { @@ -162,10 +159,11 @@ impl<'a> ValueMerger<'a> { _ => panic!("Expected array type"), }; - let actual_length = len * element_types.len(); + let actual_length = len * element_types.len() as u32; if let Some(result) = self.try_merge_only_changed_indices( then_condition, + else_condition, then_value, else_value, actual_length, @@ -175,7 +173,8 @@ impl<'a> ValueMerger<'a> { for i in 0..len { for (element_index, element_type) in element_types.iter().enumerate() { - let index = ((i * element_types.len() + element_index) as u128).into(); + let index = + ((i * element_types.len() as u32 + element_index as u32) as u128).into(); let index = self.dfg.make_constant(index, Type::field()); let typevars = Some(vec![element_type.clone()]); @@ -195,7 +194,12 @@ impl<'a> ValueMerger<'a> { let then_element = get_element(then_value, typevars.clone()); let else_element = get_element(else_value, typevars); - merged.push_back(self.merge_values(then_condition, then_element, else_element)); + merged.push_back(self.merge_values( + then_condition, + else_condition, + then_element, + else_element, + )); } } @@ -208,6 +212,7 @@ impl<'a> ValueMerger<'a> { &mut self, typ: Type, then_condition: ValueId, + else_condition: ValueId, then_value_id: ValueId, else_value_id: ValueId, ) -> ValueId { @@ -222,22 +227,22 @@ impl<'a> ValueMerger<'a> { let (slice, typ) = self.dfg.get_array_constant(then_value_id).unwrap_or_else(|| { panic!("ICE: Merging values during flattening encountered slice {then_value_id} without a preset size"); }); - slice.len() / typ.element_types().len() + (slice.len() / typ.element_types().len()) as u32 }); let else_len = self.slice_sizes.get(&else_value_id).copied().unwrap_or_else(|| { let (slice, typ) = self.dfg.get_array_constant(else_value_id).unwrap_or_else(|| { panic!("ICE: Merging values during flattening encountered slice {else_value_id} without a preset size"); }); - slice.len() / typ.element_types().len() + (slice.len() / typ.element_types().len()) as u32 }); let len = then_len.max(else_len); for i in 0..len { for (element_index, element_type) in element_types.iter().enumerate() { - let index_usize = i * element_types.len() + element_index; - let index_value = (index_usize as u128).into(); + let index_u32 = i * element_types.len() as u32 + element_index as u32; + let index_value = (index_u32 as u128).into(); let index = self.dfg.make_constant(index_value, Type::field()); let typevars = Some(vec![element_type.clone()]); @@ -245,7 +250,7 @@ impl<'a> ValueMerger<'a> { let mut get_element = |array, typevars, len| { // The smaller slice is filled with placeholder data. Codegen for slice accesses must // include checks against the dynamic slice length so that this placeholder data is not incorrectly accessed. - if len <= index_usize { + if len <= index_u32 { self.make_slice_dummy_data(element_type) } else { let get = Instruction::ArrayGet { array, index }; @@ -260,12 +265,20 @@ impl<'a> ValueMerger<'a> { } }; - let then_element = - get_element(then_value_id, typevars.clone(), then_len * element_types.len()); + let then_element = get_element( + then_value_id, + typevars.clone(), + then_len * element_types.len() as u32, + ); let else_element = - get_element(else_value_id, typevars, else_len * element_types.len()); - - merged.push_back(self.merge_values(then_condition, then_element, else_element)); + get_element(else_value_id, typevars, else_len * element_types.len() as u32); + + merged.push_back(self.merge_values( + then_condition, + else_condition, + then_element, + else_element, + )); } } @@ -314,9 +327,10 @@ impl<'a> ValueMerger<'a> { fn try_merge_only_changed_indices( &mut self, then_condition: ValueId, + else_condition: ValueId, then_value: ValueId, else_value: ValueId, - array_length: usize, + array_length: u32, ) -> Option { let mut found = false; let current_condition = self.current_condition?; @@ -370,7 +384,7 @@ impl<'a> ValueMerger<'a> { .chain(seen_else.into_iter().map(|(_, index, typ, condition)| (index, typ, condition))) .collect(); - if !found || changed_indices.len() >= array_length { + if !found || changed_indices.len() as u32 >= array_length { return None; } @@ -397,7 +411,8 @@ impl<'a> ValueMerger<'a> { let then_element = get_element(then_value, typevars.clone()); let else_element = get_element(else_value, typevars); - let value = self.merge_values(then_condition, then_element, else_element); + let value = + self.merge_values(then_condition, else_condition, then_element, else_element); array = self.insert_array_set(array, index, value, Some(condition)).first(); } diff --git a/noir/noir-repo/compiler/noirc_evaluator/src/ssa/opt/inlining.rs b/noir/noir-repo/compiler/noirc_evaluator/src/ssa/opt/inlining.rs index 6cf7070e65e..f91487fd73e 100644 --- a/noir/noir-repo/compiler/noirc_evaluator/src/ssa/opt/inlining.rs +++ b/noir/noir-repo/compiler/noirc_evaluator/src/ssa/opt/inlining.rs @@ -1089,7 +1089,6 @@ mod test { } #[test] - #[ignore] #[should_panic( expected = "Attempted to recur more than 1000 times during inlining function 'main': acir(inline) fn main f0 {" )] diff --git a/noir/noir-repo/compiler/noirc_evaluator/src/ssa/opt/loop_invariant.rs b/noir/noir-repo/compiler/noirc_evaluator/src/ssa/opt/loop_invariant.rs index 14233ca73e5..290d2a33846 100644 --- a/noir/noir-repo/compiler/noirc_evaluator/src/ssa/opt/loop_invariant.rs +++ b/noir/noir-repo/compiler/noirc_evaluator/src/ssa/opt/loop_invariant.rs @@ -7,14 +7,16 @@ //! - Already marked as loop invariants //! //! We also check that we are not hoisting instructions with side effects. -use fxhash::FxHashSet as HashSet; +use acvm::{acir::AcirField, FieldElement}; +use fxhash::{FxHashMap as HashMap, FxHashSet as HashSet}; use crate::ssa::{ ir::{ basic_block::BasicBlockId, function::{Function, RuntimeType}, function_inserter::FunctionInserter, - instruction::InstructionId, + instruction::{Instruction, InstructionId}, + types::Type, value::ValueId, }, Ssa, @@ -45,25 +47,51 @@ impl Function { } impl Loops { - fn hoist_loop_invariants(self, function: &mut Function) { + fn hoist_loop_invariants(mut self, function: &mut Function) { let mut context = LoopInvariantContext::new(function); - for loop_ in self.yet_to_unroll.iter() { + // The loops should be sorted by the number of blocks. + // We want to access outer nested loops first, which we do by popping + // from the top of the list. + while let Some(loop_) = self.yet_to_unroll.pop() { let Ok(pre_header) = loop_.get_pre_header(context.inserter.function, &self.cfg) else { // If the loop does not have a preheader we skip hoisting loop invariants for this loop continue; }; - context.hoist_loop_invariants(loop_, pre_header); + + context.hoist_loop_invariants(&loop_, pre_header); } context.map_dependent_instructions(); } } +impl Loop { + /// Find the value that controls whether to perform a loop iteration. + /// This is going to be the block parameter of the loop header. + /// + /// Consider the following example of a `for i in 0..4` loop: + /// ```text + /// brillig(inline) fn main f0 { + /// b0(v0: u32): + /// ... + /// jmp b1(u32 0) + /// b1(v1: u32): // Loop header + /// v5 = lt v1, u32 4 // Upper bound + /// jmpif v5 then: b3, else: b2 + /// ``` + /// In the example above, `v1` is the induction variable + fn get_induction_variable(&self, function: &Function) -> ValueId { + function.dfg.block_parameters(self.header)[0] + } +} + struct LoopInvariantContext<'f> { inserter: FunctionInserter<'f>, defined_in_loop: HashSet, loop_invariants: HashSet, + // Maps induction variable -> fixed upper loop bound + outer_induction_variables: HashMap, } impl<'f> LoopInvariantContext<'f> { @@ -72,6 +100,7 @@ impl<'f> LoopInvariantContext<'f> { inserter: FunctionInserter::new(function), defined_in_loop: HashSet::default(), loop_invariants: HashSet::default(), + outer_induction_variables: HashMap::default(), } } @@ -88,13 +117,29 @@ impl<'f> LoopInvariantContext<'f> { self.inserter.push_instruction(instruction_id, *block); } - self.update_values_defined_in_loop_and_invariants(instruction_id, hoist_invariant); + self.extend_values_defined_in_loop_and_invariants(instruction_id, hoist_invariant); } } + + // Keep track of a loop induction variable and respective upper bound. + // This will be used by later loops to determine whether they have operations + // reliant upon the maximum induction variable. + let upper_bound = loop_.get_const_upper_bound(self.inserter.function); + if let Some(upper_bound) = upper_bound { + let induction_variable = loop_.get_induction_variable(self.inserter.function); + let induction_variable = self.inserter.resolve(induction_variable); + self.outer_induction_variables.insert(induction_variable, upper_bound); + } } /// Gather the variables declared within the loop fn set_values_defined_in_loop(&mut self, loop_: &Loop) { + // Clear any values that may be defined in previous loops, as the context is per function. + self.defined_in_loop.clear(); + // These are safe to keep per function, but we want to be clear that these values + // are used per loop. + self.loop_invariants.clear(); + for block in loop_.blocks.iter() { let params = self.inserter.function.dfg.block_parameters(*block); self.defined_in_loop.extend(params); @@ -107,7 +152,7 @@ impl<'f> LoopInvariantContext<'f> { /// Update any values defined in the loop and loop invariants after a /// analyzing and re-inserting a loop's instruction. - fn update_values_defined_in_loop_and_invariants( + fn extend_values_defined_in_loop_and_invariants( &mut self, instruction_id: InstructionId, hoist_invariant: bool, @@ -143,9 +188,44 @@ impl<'f> LoopInvariantContext<'f> { is_loop_invariant &= !self.defined_in_loop.contains(&value) || self.loop_invariants.contains(&value); }); - is_loop_invariant && instruction.can_be_deduplicated(&self.inserter.function.dfg, false) + + let can_be_deduplicated = instruction.can_be_deduplicated(self.inserter.function, false) + || self.can_be_deduplicated_from_upper_bound(&instruction); + + is_loop_invariant && can_be_deduplicated + } + + /// Certain instructions can take advantage of that our induction variable has a fixed maximum. + /// + /// For example, an array access can usually only be safely deduplicated when we have a constant + /// index that is below the length of the array. + /// Checking an array get where the index is the loop's induction variable on its own + /// would determine that the instruction is not safe for hoisting. + /// However, if we know that the induction variable's upper bound will always be in bounds of the array + /// we can safely hoist the array access. + fn can_be_deduplicated_from_upper_bound(&self, instruction: &Instruction) -> bool { + match instruction { + Instruction::ArrayGet { array, index } => { + let array_typ = self.inserter.function.dfg.type_of_value(*array); + let upper_bound = self.outer_induction_variables.get(index); + if let (Type::Array(_, len), Some(upper_bound)) = (array_typ, upper_bound) { + upper_bound.to_u128() <= len.into() + } else { + false + } + } + _ => false, + } } + /// Loop invariant hoisting only operates over loop instructions. + /// The `FunctionInserter` is used for mapping old values to new values after + /// re-inserting loop invariant instructions. + /// However, there may be instructions which are not within loops that are + /// still reliant upon the instruction results altered during the pass. + /// This method re-inserts all instructions so that all instructions have + /// correct new value IDs based upon the `FunctionInserter` internal map. + /// Leaving out this mapping could lead to instructions with values that do not exist. fn map_dependent_instructions(&mut self) { let blocks = self.inserter.function.reachable_blocks(); for block in blocks { @@ -375,4 +455,108 @@ mod test { // The code should be unchanged assert_normalized_ssa_equals(ssa, src); } + + #[test] + fn hoist_array_gets_using_induction_variable_with_const_bound() { + // SSA for the following program: + // + // fn triple_loop(x: u32) { + // let arr = [2; 5]; + // for i in 0..4 { + // for j in 0..4 { + // for _ in 0..4 { + // assert_eq(arr[i], x); + // assert_eq(arr[j], x); + // } + // } + // } + // } + // + // `arr[i]` and `arr[j]` are safe to hoist as we know the maximum possible index + // to be used for both array accesses. + // We want to make sure `arr[i]` is hoisted to the outermost loop body and that + // `arr[j]` is hoisted to the second outermost loop body. + let src = " + brillig(inline) fn main f0 { + b0(v0: u32, v1: u32): + v6 = make_array [u32 2, u32 2, u32 2, u32 2, u32 2] : [u32; 5] + inc_rc v6 + jmp b1(u32 0) + b1(v2: u32): + v9 = lt v2, u32 4 + jmpif v9 then: b3, else: b2 + b3(): + jmp b4(u32 0) + b4(v3: u32): + v10 = lt v3, u32 4 + jmpif v10 then: b6, else: b5 + b6(): + jmp b7(u32 0) + b7(v4: u32): + v13 = lt v4, u32 4 + jmpif v13 then: b9, else: b8 + b9(): + v15 = array_get v6, index v2 -> u32 + v16 = eq v15, v0 + constrain v15 == v0 + v17 = array_get v6, index v3 -> u32 + v18 = eq v17, v0 + constrain v17 == v0 + v19 = add v4, u32 1 + jmp b7(v19) + b8(): + v14 = add v3, u32 1 + jmp b4(v14) + b5(): + v12 = add v2, u32 1 + jmp b1(v12) + b2(): + return + } + "; + + let ssa = Ssa::from_str(src).unwrap(); + + let expected = " + brillig(inline) fn main f0 { + b0(v0: u32, v1: u32): + v6 = make_array [u32 2, u32 2, u32 2, u32 2, u32 2] : [u32; 5] + inc_rc v6 + jmp b1(u32 0) + b1(v2: u32): + v9 = lt v2, u32 4 + jmpif v9 then: b3, else: b2 + b3(): + v10 = array_get v6, index v2 -> u32 + v11 = eq v10, v0 + jmp b4(u32 0) + b4(v3: u32): + v12 = lt v3, u32 4 + jmpif v12 then: b6, else: b5 + b6(): + v15 = array_get v6, index v3 -> u32 + v16 = eq v15, v0 + jmp b7(u32 0) + b7(v4: u32): + v17 = lt v4, u32 4 + jmpif v17 then: b9, else: b8 + b9(): + constrain v10 == v0 + constrain v15 == v0 + v19 = add v4, u32 1 + jmp b7(v19) + b8(): + v18 = add v3, u32 1 + jmp b4(v18) + b5(): + v14 = add v2, u32 1 + jmp b1(v14) + b2(): + return + } + "; + + let ssa = ssa.loop_invariant_code_motion(); + assert_normalized_ssa_equals(ssa, expected); + } } diff --git a/noir/noir-repo/compiler/noirc_evaluator/src/ssa/opt/mem2reg.rs b/noir/noir-repo/compiler/noirc_evaluator/src/ssa/opt/mem2reg.rs index 0690dbbf204..53a31ae57c1 100644 --- a/noir/noir-repo/compiler/noirc_evaluator/src/ssa/opt/mem2reg.rs +++ b/noir/noir-repo/compiler/noirc_evaluator/src/ssa/opt/mem2reg.rs @@ -18,6 +18,7 @@ //! - A reference with 0 aliases means we were unable to find which reference this reference //! refers to. If such a reference is stored to, we must conservatively invalidate every //! reference in the current block. +//! - We also track the last load instruction to each address per block. //! //! From there, to figure out the value of each reference at the end of block, iterate each instruction: //! - On `Instruction::Allocate`: @@ -28,6 +29,13 @@ //! - Furthermore, if the result of the load is a reference, mark the result as an alias //! of the reference it dereferences to (if known). //! - If which reference it dereferences to is not known, this load result has no aliases. +//! - We also track the last instance of a load instruction to each address in a block. +//! If we see that the last load instruction was from the same address as the current load instruction, +//! we move to replace the result of the current load with the result of the previous load. +//! This removal requires a couple conditions: +//! - No store occurs to that address before the next load, +//! - The address is not used as an argument to a call +//! This optimization helps us remove repeated loads for which there are not known values. //! - On `Instruction::Store { address, value }`: //! - If the address of the store is known: //! - If the address has exactly 1 alias: @@ -40,11 +48,13 @@ //! - Conservatively mark every alias in the block to `Unknown`. //! - Additionally, if there were no Loads to any alias of the address between this Store and //! the previous Store to the same address, the previous store can be removed. +//! - Remove the instance of the last load instruction to the address and its aliases //! - On `Instruction::Call { arguments }`: //! - If any argument of the call is a reference, set the value of each alias of that //! reference to `Unknown` //! - Any builtin functions that may return aliases if their input also contains a //! reference should be tracked. Examples: `slice_push_back`, `slice_insert`, `slice_remove`, etc. +//! - Remove the instance of the last load instruction for any reference arguments and their aliases //! //! On a terminator instruction: //! - If the terminator is a `Jmp`: @@ -274,6 +284,9 @@ impl<'f> PerFunctionContext<'f> { if let Some(first_predecessor) = predecessors.next() { let mut first = self.blocks.get(&first_predecessor).cloned().unwrap_or_default(); first.last_stores.clear(); + // Last loads are tracked per block. During unification we are creating a new block from the current one, + // so we must clear the last loads of the current block before we return the new block. + first.last_loads.clear(); // Note that we have to start folding with the first block as the accumulator. // If we started with an empty block, an empty block union'd with any other block @@ -410,6 +423,28 @@ impl<'f> PerFunctionContext<'f> { self.last_loads.insert(address, (instruction, block_id)); } + + // Check whether the block has a repeat load from the same address (w/ no calls or stores in between the loads). + // If we do have a repeat load, we can remove the current load and map its result to the previous load's result. + if let Some(last_load) = references.last_loads.get(&address) { + let Instruction::Load { address: previous_address } = + &self.inserter.function.dfg[*last_load] + else { + panic!("Expected a Load instruction here"); + }; + let result = self.inserter.function.dfg.instruction_results(instruction)[0]; + let previous_result = + self.inserter.function.dfg.instruction_results(*last_load)[0]; + if *previous_address == address { + self.inserter.map_value(result, previous_result); + self.instructions_to_remove.insert(instruction); + } + } + // We want to set the load for every load even if the address has a known value + // and the previous load instruction was removed. + // We are safe to still remove a repeat load in this case as we are mapping from the current load's + // result to the previous load, which if it was removed should already have a mapping to the known value. + references.set_last_load(address, instruction); } Instruction::Store { address, value } => { let address = self.inserter.function.dfg.resolve(*address); @@ -435,6 +470,8 @@ impl<'f> PerFunctionContext<'f> { } references.set_known_value(address, value); + // If we see a store to an address, the last load to that address needs to remain. + references.keep_last_load_for(address, self.inserter.function); references.last_stores.insert(address, instruction); } Instruction::Allocate => { @@ -542,6 +579,9 @@ impl<'f> PerFunctionContext<'f> { let value = self.inserter.function.dfg.resolve(*value); references.set_unknown(value); references.mark_value_used(value, self.inserter.function); + + // If a reference is an argument to a call, the last load to that address and its aliases needs to remain. + references.keep_last_load_for(value, self.inserter.function); } } } @@ -572,6 +612,12 @@ impl<'f> PerFunctionContext<'f> { let destination_parameters = self.inserter.function.dfg[*destination].parameters(); assert_eq!(destination_parameters.len(), arguments.len()); + // If we have multiple parameters that alias that same argument value, + // then those parameters also alias each other. + // We save parameters with repeat arguments to later mark those + // parameters as aliasing one another. + let mut arg_set: HashMap> = HashMap::default(); + // Add an alias for each reference parameter for (parameter, argument) in destination_parameters.iter().zip(arguments) { if self.inserter.function.dfg.value_is_reference(*parameter) { @@ -581,10 +627,27 @@ impl<'f> PerFunctionContext<'f> { if let Some(aliases) = references.aliases.get_mut(expression) { // The argument reference is possibly aliased by this block parameter aliases.insert(*parameter); + + // Check if we have seen the same argument + let seen_parameters = arg_set.entry(argument).or_default(); + // Add the current parameter to the parameters we have seen for this argument. + // The previous parameters and the current one alias one another. + seen_parameters.insert(*parameter); } } } } + + // Set the aliases of the parameters + for (_, aliased_params) in arg_set { + for param in aliased_params.iter() { + self.set_aliases( + references, + *param, + AliasSet::known_multiple(aliased_params.clone()), + ); + } + } } TerminatorInstruction::Return { return_values, .. } => { // Removing all `last_stores` for each returned reference is more important here @@ -612,6 +675,8 @@ mod tests { map::Id, types::Type, }, + opt::assert_normalized_ssa_equals, + Ssa, }; #[test] @@ -822,88 +887,53 @@ mod tests { // is later stored in a successor block #[test] fn load_aliases_in_predecessor_block() { - // fn main { - // b0(): - // v0 = allocate - // store Field 0 at v0 - // v2 = allocate - // store v0 at v2 - // v3 = load v2 - // v4 = load v2 - // jmp b1() - // b1(): - // store Field 1 at v3 - // store Field 2 at v4 - // v7 = load v3 - // v8 = eq v7, Field 2 - // return - // } - let main_id = Id::test_new(0); - let mut builder = FunctionBuilder::new("main".into(), main_id); - - let v0 = builder.insert_allocate(Type::field()); - - let zero = builder.field_constant(0u128); - builder.insert_store(v0, zero); - - let v2 = builder.insert_allocate(Type::Reference(Arc::new(Type::field()))); - builder.insert_store(v2, v0); - - let v3 = builder.insert_load(v2, Type::field()); - let v4 = builder.insert_load(v2, Type::field()); - let b1 = builder.insert_block(); - builder.terminate_with_jmp(b1, vec![]); - - builder.switch_to_block(b1); - - let one = builder.field_constant(1u128); - builder.insert_store(v3, one); - - let two = builder.field_constant(2u128); - builder.insert_store(v4, two); - - let v8 = builder.insert_load(v3, Type::field()); - let _ = builder.insert_binary(v8, BinaryOp::Eq, two); - - builder.terminate_with_return(vec![]); - - let ssa = builder.finish(); - assert_eq!(ssa.main().reachable_blocks().len(), 2); + let src = " + acir(inline) fn main f0 { + b0(): + v0 = allocate -> &mut Field + store Field 0 at v0 + v2 = allocate -> &mut &mut Field + store v0 at v2 + v3 = load v2 -> &mut Field + v4 = load v2 -> &mut Field + jmp b1() + b1(): + store Field 1 at v3 + store Field 2 at v4 + v7 = load v3 -> Field + v8 = eq v7, Field 2 + return + } + "; - // Expected result: - // acir fn main f0 { - // b0(): - // v9 = allocate - // store Field 0 at v9 - // v10 = allocate - // jmp b1() - // b1(): - // return - // } - let ssa = ssa.mem2reg(); - println!("{}", ssa); + let mut ssa = Ssa::from_str(src).unwrap(); + let main = ssa.main_mut(); - let main = ssa.main(); - assert_eq!(main.reachable_blocks().len(), 2); + let instructions = main.dfg[main.entry_block()].instructions(); + assert_eq!(instructions.len(), 6); // The final return is not counted // All loads should be removed - assert_eq!(count_loads(main.entry_block(), &main.dfg), 0); - assert_eq!(count_loads(b1, &main.dfg), 0); - // The first store is not removed as it is used as a nested reference in another store. - // We would need to track whether the store where `v9` is the store value gets removed to know whether + // We would need to track whether the store where `v0` is the store value gets removed to know whether // to remove it. - assert_eq!(count_stores(main.entry_block(), &main.dfg), 1); // The first store in b1 is removed since there is another store to the same reference // in the same block, and the store is not needed before the later store. // The rest of the stores are also removed as no loads are done within any blocks // to the stored values. - assert_eq!(count_stores(b1, &main.dfg), 0); - - let b1_instructions = main.dfg[b1].instructions(); + let expected = " + acir(inline) fn main f0 { + b0(): + v0 = allocate -> &mut Field + store Field 0 at v0 + v2 = allocate -> &mut &mut Field + jmp b1() + b1(): + return + } + "; - // We expect the last eq to be optimized out - assert_eq!(b1_instructions.len(), 0); + let ssa = ssa.mem2reg(); + assert_normalized_ssa_equals(ssa, expected); } #[test] @@ -933,7 +963,7 @@ mod tests { // v10 = eq v9, Field 2 // constrain v9 == Field 2 // v11 = load v2 - // v12 = load v10 + // v12 = load v11 // v13 = eq v12, Field 2 // constrain v11 == Field 2 // return @@ -992,7 +1022,7 @@ mod tests { let main = ssa.main(); assert_eq!(main.reachable_blocks().len(), 4); - // The store from the original SSA should remain + // The stores from the original SSA should remain assert_eq!(count_stores(main.entry_block(), &main.dfg), 2); assert_eq!(count_stores(b2, &main.dfg), 1); @@ -1039,4 +1069,160 @@ mod tests { let main = ssa.main(); assert_eq!(count_loads(main.entry_block(), &main.dfg), 1); } + + #[test] + fn remove_repeat_loads() { + // This tests starts with two loads from the same unknown load. + // Specifically you should look for `load v2` in `b3`. + // We should be able to remove the second repeated load. + let src = " + acir(inline) fn main f0 { + b0(): + v0 = allocate -> &mut Field + store Field 0 at v0 + v2 = allocate -> &mut &mut Field + store v0 at v2 + jmp b1(Field 0) + b1(v3: Field): + v4 = eq v3, Field 0 + jmpif v4 then: b2, else: b3 + b2(): + v5 = load v2 -> &mut Field + store Field 2 at v5 + v8 = add v3, Field 1 + jmp b1(v8) + b3(): + v9 = load v0 -> Field + v10 = eq v9, Field 2 + constrain v9 == Field 2 + v11 = load v2 -> &mut Field + v12 = load v2 -> &mut Field + v13 = load v12 -> Field + v14 = eq v13, Field 2 + constrain v13 == Field 2 + return + } + "; + + let ssa = Ssa::from_str(src).unwrap(); + + // The repeated load from v3 should be removed + // b3 should only have three loads now rather than four previously + // + // All stores are expected to remain. + let expected = " + acir(inline) fn main f0 { + b0(): + v1 = allocate -> &mut Field + store Field 0 at v1 + v3 = allocate -> &mut &mut Field + store v1 at v3 + jmp b1(Field 0) + b1(v0: Field): + v4 = eq v0, Field 0 + jmpif v4 then: b3, else: b2 + b3(): + v11 = load v3 -> &mut Field + store Field 2 at v11 + v13 = add v0, Field 1 + jmp b1(v13) + b2(): + v5 = load v1 -> Field + v7 = eq v5, Field 2 + constrain v5 == Field 2 + v8 = load v3 -> &mut Field + v9 = load v8 -> Field + v10 = eq v9, Field 2 + constrain v9 == Field 2 + return + } + "; + + let ssa = ssa.mem2reg(); + assert_normalized_ssa_equals(ssa, expected); + } + + #[test] + fn keep_repeat_loads_passed_to_a_call() { + // The test is the exact same as `remove_repeat_loads` above except with the call + // to `f1` between the repeated loads. + let src = " + acir(inline) fn main f0 { + b0(): + v1 = allocate -> &mut Field + store Field 0 at v1 + v3 = allocate -> &mut &mut Field + store v1 at v3 + jmp b1(Field 0) + b1(v0: Field): + v4 = eq v0, Field 0 + jmpif v4 then: b3, else: b2 + b3(): + v13 = load v3 -> &mut Field + store Field 2 at v13 + v15 = add v0, Field 1 + jmp b1(v15) + b2(): + v5 = load v1 -> Field + v7 = eq v5, Field 2 + constrain v5 == Field 2 + v8 = load v3 -> &mut Field + call f1(v3) + v10 = load v3 -> &mut Field + v11 = load v10 -> Field + v12 = eq v11, Field 2 + constrain v11 == Field 2 + return + } + acir(inline) fn foo f1 { + b0(v0: &mut Field): + return + } + "; + + let ssa = Ssa::from_str(src).unwrap(); + + let ssa = ssa.mem2reg(); + // We expect the program to be unchanged + assert_normalized_ssa_equals(ssa, src); + } + + #[test] + fn keep_repeat_loads_with_alias_store() { + // v7, v8, and v9 alias one another. We want to make sure that a repeat load to v7 with a store + // to its aliases in between the repeat loads does not remove those loads. + let src = " + acir(inline) fn main f0 { + b0(v0: u1): + jmpif v0 then: b2, else: b1 + b2(): + v6 = allocate -> &mut Field + store Field 0 at v6 + jmp b3(v6, v6, v6) + b3(v1: &mut Field, v2: &mut Field, v3: &mut Field): + v8 = load v1 -> Field + store Field 2 at v2 + v10 = load v1 -> Field + store Field 1 at v3 + v11 = load v1 -> Field + store Field 3 at v3 + v13 = load v1 -> Field + constrain v8 == Field 0 + constrain v10 == Field 2 + constrain v11 == Field 1 + constrain v13 == Field 3 + return + b1(): + v4 = allocate -> &mut Field + store Field 1 at v4 + jmp b3(v4, v4, v4) + } + "; + + let ssa = Ssa::from_str(src).unwrap(); + + let ssa = ssa.mem2reg(); + // We expect the program to be unchanged + assert_normalized_ssa_equals(ssa, src); + } } diff --git a/noir/noir-repo/compiler/noirc_evaluator/src/ssa/opt/remove_bit_shifts.rs b/noir/noir-repo/compiler/noirc_evaluator/src/ssa/opt/remove_bit_shifts.rs index cdbb1043232..ccf5bd9d9f8 100644 --- a/noir/noir-repo/compiler/noirc_evaluator/src/ssa/opt/remove_bit_shifts.rs +++ b/noir/noir-repo/compiler/noirc_evaluator/src/ssa/opt/remove_bit_shifts.rs @@ -191,7 +191,7 @@ impl Context<'_> { let typ = self.function.dfg.type_of_value(rhs); if let Type::Numeric(NumericType::Unsigned { bit_size }) = typ { let to_bits = self.function.dfg.import_intrinsic(Intrinsic::ToBits(Endian::Little)); - let result_types = vec![Type::Array(Arc::new(vec![Type::bool()]), bit_size as usize)]; + let result_types = vec![Type::Array(Arc::new(vec![Type::bool()]), bit_size)]; let rhs_bits = self.insert_call(to_bits, vec![rhs], result_types); let rhs_bits = rhs_bits[0]; diff --git a/noir/noir-repo/compiler/noirc_evaluator/src/ssa/opt/remove_if_else.rs b/noir/noir-repo/compiler/noirc_evaluator/src/ssa/opt/remove_if_else.rs index 8e25c3f0a35..02191801fcd 100644 --- a/noir/noir-repo/compiler/noirc_evaluator/src/ssa/opt/remove_if_else.rs +++ b/noir/noir-repo/compiler/noirc_evaluator/src/ssa/opt/remove_if_else.rs @@ -48,7 +48,7 @@ impl Function { #[derive(Default)] struct Context { - slice_sizes: HashMap, + slice_sizes: HashMap, // Maps array_set result -> element that was overwritten by that instruction. // Used to undo array_sets while merging values @@ -66,8 +66,9 @@ impl Context { for instruction in instructions { match &function.dfg[instruction] { - Instruction::IfElse { then_condition, then_value, else_value } => { + Instruction::IfElse { then_condition, then_value, else_condition, else_value } => { let then_condition = *then_condition; + let else_condition = *else_condition; let then_value = *then_value; let else_value = *else_value; @@ -84,7 +85,12 @@ impl Context { call_stack, ); - let value = value_merger.merge_values(then_condition, then_value, else_value); + let value = value_merger.merge_values( + then_condition, + else_condition, + then_value, + else_value, + ); let _typ = function.dfg.type_of_value(value); let results = function.dfg.instruction_results(instruction); @@ -142,13 +148,13 @@ impl Context { } } - fn get_or_find_capacity(&mut self, dfg: &DataFlowGraph, value: ValueId) -> usize { + fn get_or_find_capacity(&mut self, dfg: &DataFlowGraph, value: ValueId) -> u32 { match self.slice_sizes.entry(value) { Entry::Occupied(entry) => return *entry.get(), Entry::Vacant(entry) => { if let Some((array, typ)) = dfg.get_array_constant(value) { let length = array.len() / typ.element_types().len(); - return *entry.insert(length); + return *entry.insert(length as u32); } if let Type::Array(_, length) = dfg.type_of_value(value) { @@ -164,7 +170,7 @@ impl Context { enum SizeChange { None, - SetTo(ValueId, usize), + SetTo(ValueId, u32), // These two variants store the old and new slice ids // not their lengths which should be old_len = new_len +/- 1 diff --git a/noir/noir-repo/compiler/noirc_evaluator/src/ssa/opt/unrolling.rs b/noir/noir-repo/compiler/noirc_evaluator/src/ssa/opt/unrolling.rs index 777c16dacd1..1a13acc5435 100644 --- a/noir/noir-repo/compiler/noirc_evaluator/src/ssa/opt/unrolling.rs +++ b/noir/noir-repo/compiler/noirc_evaluator/src/ssa/opt/unrolling.rs @@ -19,8 +19,10 @@ //! When unrolling ACIR code, we remove reference count instructions because they are //! only used by Brillig bytecode. use acvm::{acir::AcirField, FieldElement}; +use im::HashSet; use crate::{ + brillig::brillig_gen::convert_ssa_function, errors::RuntimeError, ssa::{ ir::{ @@ -37,38 +39,60 @@ use crate::{ ssa_gen::Ssa, }, }; -use fxhash::{FxHashMap as HashMap, FxHashSet as HashSet}; +use fxhash::FxHashMap as HashMap; impl Ssa { /// Loop unrolling can return errors, since ACIR functions need to be fully unrolled. /// This meta-pass will keep trying to unroll loops and simplifying the SSA until no more errors are found. - #[tracing::instrument(level = "trace", skip(ssa))] - pub(crate) fn unroll_loops_iteratively(mut ssa: Ssa) -> Result { - for (_, function) in ssa.functions.iter_mut() { + /// + /// The `max_bytecode_incr_pct`, when given, is used to limit the growth of the Brillig bytecode size + /// after unrolling small loops to some percentage of the original loop. For example a value of 150 would + /// mean the new loop can be 150% (ie. 2.5 times) larger than the original loop. It will still contain + /// fewer SSA instructions, but that can still result in more Brillig opcodes. + #[tracing::instrument(level = "trace", skip(self))] + pub(crate) fn unroll_loops_iteratively( + mut self: Ssa, + max_bytecode_increase_percent: Option, + ) -> Result { + for (_, function) in self.functions.iter_mut() { + // Take a snapshot of the function to compare byte size increase, + // but only if the setting indicates we have to, otherwise skip it. + let orig_func_and_max_incr_pct = max_bytecode_increase_percent + .filter(|_| function.runtime().is_brillig()) + .map(|max_incr_pct| (function.clone(), max_incr_pct)); + // Try to unroll loops first: - let mut unroll_errors = function.try_unroll_loops(); + let (mut has_unrolled, mut unroll_errors) = function.try_unroll_loops(); // Keep unrolling until no more errors are found while !unroll_errors.is_empty() { let prev_unroll_err_count = unroll_errors.len(); // Simplify the SSA before retrying - - // Do a mem2reg after the last unroll to aid simplify_cfg - function.mem2reg(); - function.simplify_function(); - // Do another mem2reg after simplify_cfg to aid the next unroll - function.mem2reg(); + simplify_between_unrolls(function); // Unroll again - unroll_errors = function.try_unroll_loops(); + let (new_unrolled, new_errors) = function.try_unroll_loops(); + unroll_errors = new_errors; + has_unrolled |= new_unrolled; + // If we didn't manage to unroll any more loops, exit if unroll_errors.len() >= prev_unroll_err_count { return Err(unroll_errors.swap_remove(0)); } } + + if has_unrolled { + if let Some((orig_function, max_incr_pct)) = orig_func_and_max_incr_pct { + let new_size = brillig_bytecode_size(function); + let orig_size = brillig_bytecode_size(&orig_function); + if !is_new_size_ok(orig_size, new_size, max_incr_pct) { + *function = orig_function; + } + } + } } - Ok(ssa) + Ok(self) } } @@ -77,7 +101,7 @@ impl Function { // This can also be true for ACIR, but we have no alternative to unrolling in ACIR. // Brillig also generally prefers smaller code rather than faster code, // so we only attempt to unroll small loops, which we decide on a case-by-case basis. - fn try_unroll_loops(&mut self) -> Vec { + fn try_unroll_loops(&mut self) -> (bool, Vec) { Loops::find_all(self).unroll_each(self) } } @@ -85,7 +109,7 @@ impl Function { pub(super) struct Loop { /// The header block of a loop is the block which dominates all the /// other blocks in the loop. - header: BasicBlockId, + pub(super) header: BasicBlockId, /// The start of the back_edge n -> d is the block n at the end of /// the loop that jumps back to the header block d which restarts the loop. @@ -170,8 +194,10 @@ impl Loops { /// Unroll all loops within a given function. /// Any loops which fail to be unrolled (due to using non-constant indices) will be unmodified. - fn unroll_each(mut self, function: &mut Function) -> Vec { + /// Returns whether any blocks have been modified + fn unroll_each(mut self, function: &mut Function) -> (bool, Vec) { let mut unroll_errors = vec![]; + let mut has_unrolled = false; while let Some(next_loop) = self.yet_to_unroll.pop() { if function.runtime().is_brillig() && !next_loop.is_small_loop(function, &self.cfg) { continue; @@ -181,13 +207,17 @@ impl Loops { if next_loop.blocks.iter().any(|block| self.modified_blocks.contains(block)) { let mut new_loops = Self::find_all(function); new_loops.failed_to_unroll = self.failed_to_unroll; - return unroll_errors.into_iter().chain(new_loops.unroll_each(function)).collect(); + let (new_unrolled, new_errors) = new_loops.unroll_each(function); + return (has_unrolled || new_unrolled, [unroll_errors, new_errors].concat()); } // Don't try to unroll the loop again if it is known to fail if !self.failed_to_unroll.contains(&next_loop.header) { match next_loop.unroll(function, &self.cfg) { - Ok(_) => self.modified_blocks.extend(next_loop.blocks), + Ok(_) => { + has_unrolled = true; + self.modified_blocks.extend(next_loop.blocks); + } Err(call_stack) => { self.failed_to_unroll.insert(next_loop.header); unroll_errors.push(RuntimeError::UnknownLoopBound { call_stack }); @@ -195,7 +225,7 @@ impl Loops { } } } - unroll_errors + (has_unrolled, unroll_errors) } } @@ -269,7 +299,7 @@ impl Loop { /// v5 = lt v1, u32 4 // Upper bound /// jmpif v5 then: b3, else: b2 /// ``` - fn get_const_upper_bound(&self, function: &Function) -> Option { + pub(super) fn get_const_upper_bound(&self, function: &Function) -> Option { let block = &function.dfg[self.header]; let instructions = block.instructions(); assert_eq!( @@ -947,21 +977,59 @@ impl<'f> LoopIteration<'f> { } } +/// Unrolling leaves some duplicate instructions which can potentially be removed. +fn simplify_between_unrolls(function: &mut Function) { + // Do a mem2reg after the last unroll to aid simplify_cfg + function.mem2reg(); + function.simplify_function(); + // Do another mem2reg after simplify_cfg to aid the next unroll + function.mem2reg(); +} + +/// Convert the function to Brillig bytecode and return the resulting size. +fn brillig_bytecode_size(function: &Function) -> usize { + // We need to do some SSA passes in order for the conversion to be able to go ahead, + // otherwise we can hit `unreachable!()` instructions in `convert_ssa_instruction`. + // Creating a clone so as not to modify the originals. + let mut temp = function.clone(); + + // Might as well give it the best chance. + simplify_between_unrolls(&mut temp); + + // This is to try to prevent hitting ICE. + temp.dead_instruction_elimination(false); + + convert_ssa_function(&temp, false).byte_code.len() +} + +/// Decide if the new bytecode size is acceptable, compared to the original. +/// +/// The maximum increase can be expressed as a negative value if we demand a decrease. +/// (Values -100 and under mean the new size should be 0). +fn is_new_size_ok(orig_size: usize, new_size: usize, max_incr_pct: i32) -> bool { + let max_size_pct = 100i32.saturating_add(max_incr_pct).max(0) as usize; + let max_size = orig_size.saturating_mul(max_size_pct); + new_size.saturating_mul(100) <= max_size +} + #[cfg(test)] mod tests { use acvm::FieldElement; + use test_case::test_case; use crate::errors::RuntimeError; use crate::ssa::{ir::value::ValueId, opt::assert_normalized_ssa_equals, Ssa}; - use super::{BoilerplateStats, Loops}; + use super::{is_new_size_ok, BoilerplateStats, Loops}; - /// Tries to unroll all loops in each SSA function. + /// Tries to unroll all loops in each SSA function once, calling the `Function` directly, + /// bypassing the iterative loop done by the SSA which does further optimisations. + /// /// If any loop cannot be unrolled, it is left as-is or in a partially unrolled state. fn try_unroll_loops(mut ssa: Ssa) -> (Ssa, Vec) { let mut errors = vec![]; for function in ssa.functions.values_mut() { - errors.extend(function.try_unroll_loops()); + errors.extend(function.try_unroll_loops().1); } (ssa, errors) } @@ -1221,9 +1289,26 @@ mod tests { let (ssa, errors) = try_unroll_loops(ssa); assert_eq!(errors.len(), 0, "Unroll should have no errors"); + // Check that it's still the original assert_normalized_ssa_equals(ssa, parse_ssa().to_string().as_str()); } + #[test] + fn test_brillig_unroll_iteratively_respects_max_increase() { + let ssa = brillig_unroll_test_case(); + let ssa = ssa.unroll_loops_iteratively(Some(-90)).unwrap(); + // Check that it's still the original + assert_normalized_ssa_equals(ssa, brillig_unroll_test_case().to_string().as_str()); + } + + #[test] + fn test_brillig_unroll_iteratively_with_large_max_increase() { + let ssa = brillig_unroll_test_case(); + let ssa = ssa.unroll_loops_iteratively(Some(50)).unwrap(); + // Check that it did the unroll + assert_eq!(ssa.main().reachable_blocks().len(), 2, "The loop should be unrolled"); + } + /// Test that `break` and `continue` stop unrolling without any panic. #[test] fn test_brillig_unroll_break_and_continue() { @@ -1377,4 +1462,14 @@ mod tests { let loop0 = loops.yet_to_unroll.pop().expect("there should be a loop"); loop0.boilerplate_stats(function, &loops.cfg).expect("there should be stats") } + + #[test_case(1000, 700, 50, true; "size decreased")] + #[test_case(1000, 1500, 50, true; "size increased just by the max")] + #[test_case(1000, 1501, 50, false; "size increased over the max")] + #[test_case(1000, 700, -50, false; "size decreased but not enough")] + #[test_case(1000, 250, -50, true; "size decreased over expectations")] + #[test_case(1000, 250, -1250, false; "demanding more than minus 100 is handled")] + fn test_is_new_size_ok(old: usize, new: usize, max: i32, ok: bool) { + assert_eq!(is_new_size_ok(old, new, max), ok); + } } diff --git a/noir/noir-repo/compiler/noirc_evaluator/src/ssa/parser/lexer.rs b/noir/noir-repo/compiler/noirc_evaluator/src/ssa/parser/lexer.rs index d89bc1e9e28..5b66810c641 100644 --- a/noir/noir-repo/compiler/noirc_evaluator/src/ssa/parser/lexer.rs +++ b/noir/noir-repo/compiler/noirc_evaluator/src/ssa/parser/lexer.rs @@ -62,6 +62,7 @@ impl<'a> Lexer<'a> { Some('-') if self.peek_char() == Some('>') => self.double_char_token(Token::Arrow), Some('-') => self.single_char_token(Token::Dash), Some('"') => self.eat_string_literal(), + Some('b') if self.peek_char() == Some('"') => self.eat_byte_string_literal(), Some(ch) if ch.is_ascii_alphanumeric() || ch == '_' => self.eat_alpha_numeric(ch), Some(char) => Err(LexerError::UnexpectedCharacter { char, @@ -180,8 +181,23 @@ impl<'a> Lexer<'a> { fn eat_string_literal(&mut self) -> SpannedTokenResult { let start = self.position; - let mut string = String::new(); + let string = self.eat_string(start)?; + let str_literal_token = Token::Str(string); + let end = self.position; + Ok(str_literal_token.into_span(start, end)) + } + + fn eat_byte_string_literal(&mut self) -> SpannedTokenResult { + let start = self.position; + self.next_char(); // skip the b + let string = self.eat_string(start)?; + let str_literal_token = Token::ByteStr(string); + let end = self.position; + Ok(str_literal_token.into_span(start, end)) + } + fn eat_string(&mut self, start: u32) -> Result { + let mut string = String::new(); while let Some(next) = self.next_char() { let char = match next { '"' => break, @@ -206,11 +222,7 @@ impl<'a> Lexer<'a> { string.push(char); } - - let str_literal_token = Token::Str(string); - - let end = self.position; - Ok(str_literal_token.into_span(start, end)) + Ok(string) } fn eat_while bool>( diff --git a/noir/noir-repo/compiler/noirc_evaluator/src/ssa/parser/mod.rs b/noir/noir-repo/compiler/noirc_evaluator/src/ssa/parser/mod.rs index 3d8bd37dead..24a5ff43071 100644 --- a/noir/noir-repo/compiler/noirc_evaluator/src/ssa/parser/mod.rs +++ b/noir/noir-repo/compiler/noirc_evaluator/src/ssa/parser/mod.rs @@ -4,7 +4,10 @@ use std::{ }; use super::{ - ir::{instruction::BinaryOp, types::Type}, + ir::{ + instruction::BinaryOp, + types::{NumericType, Type}, + }, Ssa, }; @@ -448,12 +451,39 @@ impl<'a> Parser<'a> { } if self.eat_keyword(Keyword::MakeArray)? { - self.eat_or_error(Token::LeftBracket)?; - let elements = self.parse_comma_separated_values()?; - self.eat_or_error(Token::RightBracket)?; - self.eat_or_error(Token::Colon)?; - let typ = self.parse_type()?; - return Ok(ParsedInstruction::MakeArray { target, elements, typ }); + if self.eat(Token::Ampersand)? { + let Some(string) = self.eat_byte_str()? else { + return self.expected_byte_string(); + }; + let u8 = Type::Numeric(NumericType::Unsigned { bit_size: 8 }); + let typ = Type::Slice(Arc::new(vec![u8.clone()])); + let elements = string + .bytes() + .map(|byte| ParsedValue::NumericConstant { + constant: FieldElement::from(byte as u128), + typ: u8.clone(), + }) + .collect(); + return Ok(ParsedInstruction::MakeArray { target, elements, typ }); + } else if let Some(string) = self.eat_byte_str()? { + let u8 = Type::Numeric(NumericType::Unsigned { bit_size: 8 }); + let typ = Type::Array(Arc::new(vec![u8.clone()]), string.len() as u32); + let elements = string + .bytes() + .map(|byte| ParsedValue::NumericConstant { + constant: FieldElement::from(byte as u128), + typ: u8.clone(), + }) + .collect(); + return Ok(ParsedInstruction::MakeArray { target, elements, typ }); + } else { + self.eat_or_error(Token::LeftBracket)?; + let elements = self.parse_comma_separated_values()?; + self.eat_or_error(Token::RightBracket)?; + self.eat_or_error(Token::Colon)?; + let typ = self.parse_type()?; + return Ok(ParsedInstruction::MakeArray { target, elements, typ }); + } } if self.eat_keyword(Keyword::Not)? { @@ -656,7 +686,7 @@ impl<'a> Parser<'a> { if self.eat(Token::Semicolon)? { let length = self.eat_int_or_error()?; self.eat_or_error(Token::RightBracket)?; - return Ok(Type::Array(Arc::new(element_types), length.to_u128() as usize)); + return Ok(Type::Array(Arc::new(element_types), length.to_u128() as u32)); } else { self.eat_or_error(Token::RightBracket)?; return Ok(Type::Slice(Arc::new(element_types))); @@ -796,6 +826,18 @@ impl<'a> Parser<'a> { } } + fn eat_byte_str(&mut self) -> ParseResult> { + if matches!(self.token.token(), Token::ByteStr(..)) { + let token = self.bump()?; + match token.into_token() { + Token::ByteStr(string) => Ok(Some(string)), + _ => unreachable!(), + } + } else { + Ok(None) + } + } + fn eat(&mut self, token: Token) -> ParseResult { if self.token.token() == &token { self.bump()?; @@ -848,6 +890,13 @@ impl<'a> Parser<'a> { }) } + fn expected_byte_string(&mut self) -> ParseResult { + Err(ParserError::ExpectedByteString { + found: self.token.token().clone(), + span: self.token.to_span(), + }) + } + fn expected_identifier(&mut self) -> ParseResult { Err(ParserError::ExpectedIdentifier { found: self.token.token().clone(), @@ -911,6 +960,8 @@ pub(crate) enum ParserError { ExpectedInstructionOrTerminator { found: Token, span: Span }, #[error("Expected a string literal or 'data', found '{found}'")] ExpectedStringOrData { found: Token, span: Span }, + #[error("Expected a byte string literal, found '{found}'")] + ExpectedByteString { found: Token, span: Span }, #[error("Expected a value, found '{found}'")] ExpectedValue { found: Token, span: Span }, #[error("Multiple return values only allowed for call")] @@ -928,6 +979,7 @@ impl ParserError { | ParserError::ExpectedType { span, .. } | ParserError::ExpectedInstructionOrTerminator { span, .. } | ParserError::ExpectedStringOrData { span, .. } + | ParserError::ExpectedByteString { span, .. } | ParserError::ExpectedValue { span, .. } => *span, ParserError::MultipleReturnValuesOnlyAllowedForCall { second_target, .. } => { second_target.span diff --git a/noir/noir-repo/compiler/noirc_evaluator/src/ssa/parser/tests.rs b/noir/noir-repo/compiler/noirc_evaluator/src/ssa/parser/tests.rs index 593b66d0c98..6318f9dc56e 100644 --- a/noir/noir-repo/compiler/noirc_evaluator/src/ssa/parser/tests.rs +++ b/noir/noir-repo/compiler/noirc_evaluator/src/ssa/parser/tests.rs @@ -89,6 +89,30 @@ fn test_make_composite_array() { assert_ssa_roundtrip(src); } +#[test] +fn test_make_byte_array_with_string_literal() { + let src = " + acir(inline) fn main f0 { + b0(): + v9 = make_array b\"Hello world!\" + return v9 + } + "; + assert_ssa_roundtrip(src); +} + +#[test] +fn test_make_byte_slice_with_string_literal() { + let src = " + acir(inline) fn main f0 { + b0(): + v9 = make_array &b\"Hello world!\" + return v9 + } + "; + assert_ssa_roundtrip(src); +} + #[test] fn test_block_parameters() { let src = " @@ -228,14 +252,14 @@ fn test_constrain_with_static_message() { #[test] fn test_constrain_with_dynamic_message() { - let src = " + let src = r#" acir(inline) fn main f0 { b0(v0: Field, v1: Field): - v7 = make_array [u8 123, u8 120, u8 125, u8 32, u8 123, u8 121, u8 125] : [u8; 7] + v7 = make_array b"{x} {y}" constrain v0 == Field 1, data v7, u32 2, v0, v1 return } - "; + "#; assert_ssa_roundtrip(src); } diff --git a/noir/noir-repo/compiler/noirc_evaluator/src/ssa/parser/token.rs b/noir/noir-repo/compiler/noirc_evaluator/src/ssa/parser/token.rs index d8dd4ec011e..83a2a1d1ed2 100644 --- a/noir/noir-repo/compiler/noirc_evaluator/src/ssa/parser/token.rs +++ b/noir/noir-repo/compiler/noirc_evaluator/src/ssa/parser/token.rs @@ -30,6 +30,7 @@ pub(crate) enum Token { Ident(String), Int(FieldElement), Str(String), + ByteStr(String), Keyword(Keyword), IntType(IntType), /// = @@ -79,6 +80,7 @@ impl Display for Token { Token::Ident(ident) => write!(f, "{}", ident), Token::Int(int) => write!(f, "{}", int), Token::Str(string) => write!(f, "{string:?}"), + Token::ByteStr(string) => write!(f, "{string:?}"), Token::Keyword(keyword) => write!(f, "{}", keyword), Token::IntType(int_type) => write!(f, "{}", int_type), Token::Assign => write!(f, "="), diff --git a/noir/noir-repo/compiler/noirc_evaluator/src/ssa/ssa_gen/context.rs b/noir/noir-repo/compiler/noirc_evaluator/src/ssa/ssa_gen/context.rs index 0c6041029da..116e0de4ecd 100644 --- a/noir/noir-repo/compiler/noirc_evaluator/src/ssa/ssa_gen/context.rs +++ b/noir/noir-repo/compiler/noirc_evaluator/src/ssa/ssa_gen/context.rs @@ -20,7 +20,7 @@ use crate::ssa::ir::value::ValueId; use super::value::{Tree, Value, Values}; use super::SSA_WORD_SIZE; -use fxhash::FxHashMap as HashMap; +use fxhash::{FxHashMap as HashMap, FxHashSet as HashSet}; /// The FunctionContext is the main context object for translating a /// function into SSA form during the SSA-gen pass. @@ -159,7 +159,8 @@ impl<'a> FunctionContext<'a> { let parameter_value = Self::map_type(parameter_type, |typ| { let value = self.builder.add_parameter(typ); if mutable { - self.new_mutable_variable(value) + // This will wrap any `mut var: T` in a reference and increase the rc of an array if needed + self.new_mutable_variable(value, true) } else { value.into() } @@ -170,8 +171,17 @@ impl<'a> FunctionContext<'a> { /// Allocate a single slot of memory and store into it the given initial value of the variable. /// Always returns a Value::Mutable wrapping the allocate instruction. - pub(super) fn new_mutable_variable(&mut self, value_to_store: ValueId) -> Value { + pub(super) fn new_mutable_variable( + &mut self, + value_to_store: ValueId, + increment_array_rc: bool, + ) -> Value { let element_type = self.builder.current_function.dfg.type_of_value(value_to_store); + + if increment_array_rc { + self.builder.increment_array_reference_count(value_to_store); + } + let alloc = self.builder.insert_allocate(element_type); self.builder.insert_store(alloc, value_to_store); let typ = self.builder.type_of_value(value_to_store); @@ -236,12 +246,12 @@ impl<'a> FunctionContext<'a> { ast::Type::Field => Type::field(), ast::Type::Array(len, element) => { let element_types = Self::convert_type(element).flatten(); - Type::Array(Arc::new(element_types), *len as usize) + Type::Array(Arc::new(element_types), *len) } ast::Type::Integer(Signedness::Signed, bits) => Type::signed((*bits).into()), ast::Type::Integer(Signedness::Unsigned, bits) => Type::unsigned((*bits).into()), ast::Type::Bool => Type::unsigned(1), - ast::Type::String(len) => Type::str(*len as usize), + ast::Type::String(len) => Type::str(*len), ast::Type::FmtString(_, _) => { panic!("convert_non_tuple_type called on a fmt string: {typ}") } @@ -732,10 +742,6 @@ impl<'a> FunctionContext<'a> { let element_types = Self::convert_type(element_type); values.map_both(element_types, |value, element_type| { let reference = value.eval_reference(); - // Reference counting in brillig relies on us incrementing reference - // counts when arrays/slices are constructed or indexed. - // Thus, if we dereference an lvalue which happens to be array/slice we should increment its reference counter. - self.builder.increment_array_reference_count(reference); self.builder.insert_load(reference, element_type).into() }) } @@ -907,33 +913,55 @@ impl<'a> FunctionContext<'a> { } } - /// Increments the reference count of all parameters. Returns the entry block of the function. + /// Increments the reference count of mutable reference array parameters. + /// Any mutable-value (`mut a: [T; N]` versus `a: &mut [T; N]`) are already incremented + /// by `FunctionBuilder::add_parameter_to_scope`. + /// Returns each array id that was incremented. /// /// This is done on parameters rather than call arguments so that we can optimize out /// paired inc/dec instructions within brillig functions more easily. - pub(crate) fn increment_parameter_rcs(&mut self) -> BasicBlockId { + pub(crate) fn increment_parameter_rcs(&mut self) -> HashSet { let entry = self.builder.current_function.entry_block(); let parameters = self.builder.current_function.dfg.block_parameters(entry).to_vec(); + let mut incremented = HashSet::default(); + let mut seen_array_types = HashSet::default(); + for parameter in parameters { - self.builder.increment_array_reference_count(parameter); + // Avoid reference counts for immutable arrays that aren't behind references. + let typ = self.builder.current_function.dfg.type_of_value(parameter); + + if let Type::Reference(element) = typ { + if element.contains_an_array() { + // If we haven't already seen this array type, the value may be possibly + // aliased, so issue an inc_rc for it. + if !seen_array_types.insert(element.get_contained_array().clone()) + && self.builder.increment_array_reference_count(parameter) + { + incremented.insert(parameter); + } + } + } } - entry + incremented } /// Ends a local scope of a function. /// This will issue DecrementRc instructions for any arrays in the given starting scope /// block's parameters. Arrays that are also used in terminator instructions for the scope are /// ignored. - pub(crate) fn end_scope(&mut self, scope: BasicBlockId, terminator_args: &[ValueId]) { - let mut dropped_parameters = - self.builder.current_function.dfg.block_parameters(scope).to_vec(); - - dropped_parameters.retain(|parameter| !terminator_args.contains(parameter)); + pub(crate) fn end_scope( + &mut self, + mut incremented_params: HashSet, + terminator_args: &[ValueId], + ) { + incremented_params.retain(|parameter| !terminator_args.contains(parameter)); - for parameter in dropped_parameters { - self.builder.decrement_array_reference_count(parameter); + for parameter in incremented_params { + if self.builder.current_function.dfg.value_is_reference(parameter) { + self.builder.decrement_array_reference_count(parameter); + } } } diff --git a/noir/noir-repo/compiler/noirc_evaluator/src/ssa/ssa_gen/mod.rs b/noir/noir-repo/compiler/noirc_evaluator/src/ssa/ssa_gen/mod.rs index c50f0a7f45c..2fe0a38af00 100644 --- a/noir/noir-repo/compiler/noirc_evaluator/src/ssa/ssa_gen/mod.rs +++ b/noir/noir-repo/compiler/noirc_evaluator/src/ssa/ssa_gen/mod.rs @@ -125,10 +125,10 @@ impl<'a> FunctionContext<'a> { /// Codegen a function's body and set its return value to that of its last parameter. /// For functions returning nothing, this will be an empty list. fn codegen_function_body(&mut self, body: &Expression) -> Result<(), RuntimeError> { - let entry_block = self.increment_parameter_rcs(); + let incremented_params = self.increment_parameter_rcs(); let return_value = self.codegen_expression(body)?; let results = return_value.into_value_list(self); - self.end_scope(entry_block, &results); + self.end_scope(incremented_params, &results); self.builder.terminate_with_return(results); Ok(()) @@ -195,8 +195,7 @@ impl<'a> FunctionContext<'a> { fn codegen_literal(&mut self, literal: &ast::Literal) -> Result { match literal { ast::Literal::Array(array) => { - let elements = - try_vecmap(&array.contents, |element| self.codegen_expression(element))?; + let elements = self.codegen_array_elements(&array.contents)?; let typ = Self::convert_type(&array.typ).flatten(); Ok(match array.typ { @@ -207,8 +206,7 @@ impl<'a> FunctionContext<'a> { }) } ast::Literal::Slice(array) => { - let elements = - try_vecmap(&array.contents, |element| self.codegen_expression(element))?; + let elements = self.codegen_array_elements(&array.contents)?; let typ = Self::convert_type(&array.typ).flatten(); Ok(match array.typ { @@ -245,18 +243,33 @@ impl<'a> FunctionContext<'a> { } } + fn codegen_array_elements( + &mut self, + elements: &[Expression], + ) -> Result, RuntimeError> { + try_vecmap(elements, |element| { + let value = self.codegen_expression(element)?; + Ok((value, element.is_array_or_slice_literal())) + }) + } + fn codegen_string(&mut self, string: &str) -> Values { let elements = vecmap(string.as_bytes(), |byte| { - self.builder.numeric_constant(*byte as u128, Type::unsigned(8)).into() + let char = self.builder.numeric_constant(*byte as u128, Type::unsigned(8)); + (char.into(), false) }); let typ = Self::convert_non_tuple_type(&ast::Type::String(elements.len() as u32)); self.codegen_array(elements, typ) } // Codegen an array but make sure that we do not have a nested slice + /// + /// The bool aspect of each array element indicates whether the element is an array constant + /// or not. If it is, we avoid incrementing the reference count because we consider the + /// constant to be moved into this larger array constant. fn codegen_array_checked( &mut self, - elements: Vec, + elements: Vec<(Values, bool)>, typ: Type, ) -> Result { if typ.is_nested_slice() { @@ -273,11 +286,15 @@ impl<'a> FunctionContext<'a> { /// stored next to the other fields in memory. So an array such as [(1, 2), (3, 4)] is /// stored the same as the array [1, 2, 3, 4]. /// + /// The bool aspect of each array element indicates whether the element is an array constant + /// or not. If it is, we avoid incrementing the reference count because we consider the + /// constant to be moved into this larger array constant. + /// /// The value returned from this function is always that of the allocate instruction. - fn codegen_array(&mut self, elements: Vec, typ: Type) -> Values { + fn codegen_array(&mut self, elements: Vec<(Values, bool)>, typ: Type) -> Values { let mut array = im::Vector::new(); - for element in elements { + for (element, is_array_constant) in elements { element.for_each(|element| { let element = element.eval(self); @@ -286,7 +303,10 @@ impl<'a> FunctionContext<'a> { // pessimistic reference count (since some are likely moved rather than shared) // which is important for Brillig's copy on write optimization. This has no // effect in ACIR code. - self.builder.increment_array_reference_count(element); + if !is_array_constant { + self.builder.increment_array_reference_count(element); + } + array.push_back(element); }); } @@ -662,15 +682,22 @@ impl<'a> FunctionContext<'a> { fn codegen_let(&mut self, let_expr: &ast::Let) -> Result { let mut values = self.codegen_expression(&let_expr.expression)?; + // Don't mutate the reference count if we're assigning an array literal to a Let: + // `let mut foo = [1, 2, 3];` + // we consider the array to be moved, so we should have an initial rc of just 1. + let should_inc_rc = !let_expr.expression.is_array_or_slice_literal(); + values = values.map(|value| { let value = value.eval(self); - // Make sure to increment array reference counts on each let binding - self.builder.increment_array_reference_count(value); - Tree::Leaf(if let_expr.mutable { - self.new_mutable_variable(value) + self.new_mutable_variable(value, should_inc_rc) } else { + // `new_mutable_variable` increments rcs internally so we have to + // handle it separately for the immutable case + if should_inc_rc { + self.builder.increment_array_reference_count(value); + } value::Value::Normal(value) }) }); @@ -729,10 +756,14 @@ impl<'a> FunctionContext<'a> { fn codegen_assign(&mut self, assign: &ast::Assign) -> Result { let lhs = self.extract_current_value(&assign.lvalue)?; let rhs = self.codegen_expression(&assign.expression)?; + let should_inc_rc = !assign.expression.is_array_or_slice_literal(); rhs.clone().for_each(|value| { let value = value.eval(self); - self.builder.increment_array_reference_count(value); + + if should_inc_rc { + self.builder.increment_array_reference_count(value); + } }); self.assign_new_value(lhs, rhs); diff --git a/noir/noir-repo/compiler/noirc_frontend/src/debug/mod.rs b/noir/noir-repo/compiler/noirc_frontend/src/debug/mod.rs index fed3149118b..f05fc721581 100644 --- a/noir/noir-repo/compiler/noirc_frontend/src/debug/mod.rs +++ b/noir/noir-repo/compiler/noirc_frontend/src/debug/mod.rs @@ -67,12 +67,16 @@ impl DebugInstrumenter { self.insert_state_set_oracle(module, 8); } - fn insert_var(&mut self, var_name: &str) -> SourceVarId { + fn insert_var(&mut self, var_name: &str) -> Option { + if var_name == "_" { + return None; + } + let var_id = SourceVarId(self.next_var_id); self.next_var_id += 1; self.variables.insert(var_id, var_name.to_string()); self.scope.last_mut().unwrap().insert(var_name.to_string(), var_id); - var_id + Some(var_id) } fn lookup_var(&self, var_name: &str) -> Option { @@ -107,9 +111,9 @@ impl DebugInstrumenter { .flat_map(|param| { pattern_vars(¶m.pattern) .iter() - .map(|(id, _is_mut)| { - let var_id = self.insert_var(&id.0.contents); - build_assign_var_stmt(var_id, id_expr(id)) + .filter_map(|(id, _is_mut)| { + let var_id = self.insert_var(&id.0.contents)?; + Some(build_assign_var_stmt(var_id, id_expr(id))) }) .collect::>() }) @@ -225,13 +229,28 @@ impl DebugInstrumenter { } }) .collect(); - let vars_exprs: Vec = vars.iter().map(|(id, _)| id_expr(id)).collect(); + let vars_exprs: Vec = vars + .iter() + .map(|(id, _)| { + // We don't want to generate an expression to read from "_". + // And since this expression is going to be assigned to "_" so it doesn't matter + // what it is, we can use `()` for it. + if id.0.contents == "_" { + ast::Expression { + kind: ast::ExpressionKind::Literal(ast::Literal::Unit), + span: id.span(), + } + } else { + id_expr(id) + } + }) + .collect(); let mut block_stmts = vec![ast::Statement { kind: ast::StatementKind::Let(let_stmt.clone()), span: *span }]; - block_stmts.extend(vars.iter().map(|(id, _)| { - let var_id = self.insert_var(&id.0.contents); - build_assign_var_stmt(var_id, id_expr(id)) + block_stmts.extend(vars.iter().filter_map(|(id, _)| { + let var_id = self.insert_var(&id.0.contents)?; + Some(build_assign_var_stmt(var_id, id_expr(id))) })); block_stmts.push(ast::Statement { kind: ast::StatementKind::Expression(ast::Expression { @@ -422,21 +441,31 @@ impl DebugInstrumenter { let var_name = &for_stmt.identifier.0.contents; let var_id = self.insert_var(var_name); - let set_stmt = build_assign_var_stmt(var_id, id_expr(&for_stmt.identifier)); - let drop_stmt = build_drop_var_stmt(var_id, Span::empty(for_stmt.span.end())); + let set_and_drop_stmt = var_id.map(|var_id| { + ( + build_assign_var_stmt(var_id, id_expr(&for_stmt.identifier)), + build_drop_var_stmt(var_id, Span::empty(for_stmt.span.end())), + ) + }); self.walk_expr(&mut for_stmt.block); + + let mut statements = Vec::new(); + let block_statement = ast::Statement { + kind: ast::StatementKind::Semi(for_stmt.block.clone()), + span: for_stmt.block.span, + }; + + if let Some((set_stmt, drop_stmt)) = set_and_drop_stmt { + statements.push(set_stmt); + statements.push(block_statement); + statements.push(drop_stmt); + } else { + statements.push(block_statement); + } + for_stmt.block = ast::Expression { - kind: ast::ExpressionKind::Block(ast::BlockExpression { - statements: vec![ - set_stmt, - ast::Statement { - kind: ast::StatementKind::Semi(for_stmt.block.clone()), - span: for_stmt.block.span, - }, - drop_stmt, - ], - }), + kind: ast::ExpressionKind::Block(ast::BlockExpression { statements }), span: for_stmt.span, }; } diff --git a/noir/noir-repo/compiler/noirc_frontend/src/elaborator/comptime.rs b/noir/noir-repo/compiler/noirc_frontend/src/elaborator/comptime.rs index a27e2bf0163..962356d6dd9 100644 --- a/noir/noir-repo/compiler/noirc_frontend/src/elaborator/comptime.rs +++ b/noir/noir-repo/compiler/noirc_frontend/src/elaborator/comptime.rs @@ -329,8 +329,6 @@ impl<'context> Elaborator<'context> { push_arg(Value::TraitDefinition(trait_id)); } else { let (expr_id, expr_type) = interpreter.elaborator.elaborate_expression(arg); - push_arg(interpreter.evaluate(expr_id)?); - if let Err(UnificationError) = expr_type.unify(param_type) { return Err(InterpreterError::TypeMismatch { expected: param_type.clone(), @@ -338,6 +336,7 @@ impl<'context> Elaborator<'context> { location: arg_location, }); } + push_arg(interpreter.evaluate(expr_id)?); }; } diff --git a/noir/noir-repo/compiler/noirc_frontend/src/elaborator/mod.rs b/noir/noir-repo/compiler/noirc_frontend/src/elaborator/mod.rs index 20d27fbc9ac..478504a79be 100644 --- a/noir/noir-repo/compiler/noirc_frontend/src/elaborator/mod.rs +++ b/noir/noir-repo/compiler/noirc_frontend/src/elaborator/mod.rs @@ -440,6 +440,9 @@ impl<'context> Elaborator<'context> { // so we need to reintroduce the same IDs into scope here. for parameter in &func_meta.parameter_idents { let name = self.interner.definition_name(parameter.id).to_owned(); + if name == "_" { + continue; + } let warn_if_unused = !(func_meta.trait_impl.is_some() && name == "self"); self.add_existing_variable_to_scope(name, parameter.clone(), warn_if_unused); } diff --git a/noir/noir-repo/compiler/noirc_frontend/src/elaborator/patterns.rs b/noir/noir-repo/compiler/noirc_frontend/src/elaborator/patterns.rs index 3928362db11..3fbdadbbee8 100644 --- a/noir/noir-repo/compiler/noirc_frontend/src/elaborator/patterns.rs +++ b/noir/noir-repo/compiler/noirc_frontend/src/elaborator/patterns.rs @@ -331,16 +331,18 @@ impl<'context> Elaborator<'context> { let resolver_meta = ResolverMeta { num_times_used: 0, ident: ident.clone(), warn_if_unused }; - let scope = self.scopes.get_mut_scope(); - let old_value = scope.add_key_value(name.clone(), resolver_meta); - - if !allow_shadowing { - if let Some(old_value) = old_value { - self.push_err(ResolverError::DuplicateDefinition { - name, - first_span: old_value.ident.location.span, - second_span: location.span, - }); + if name != "_" { + let scope = self.scopes.get_mut_scope(); + let old_value = scope.add_key_value(name.clone(), resolver_meta); + + if !allow_shadowing { + if let Some(old_value) = old_value { + self.push_err(ResolverError::DuplicateDefinition { + name, + first_span: old_value.ident.location.span, + second_span: location.span, + }); + } } } diff --git a/noir/noir-repo/compiler/noirc_frontend/src/elaborator/types.rs b/noir/noir-repo/compiler/noirc_frontend/src/elaborator/types.rs index 7e06964b563..0404ae3c2c0 100644 --- a/noir/noir-repo/compiler/noirc_frontend/src/elaborator/types.rs +++ b/noir/noir-repo/compiler/noirc_frontend/src/elaborator/types.rs @@ -1321,11 +1321,23 @@ impl<'context> Elaborator<'context> { { Some(method_id) => Some(HirMethodReference::FuncId(method_id)), None => { - self.push_err(TypeCheckError::UnresolvedMethodCall { - method_name: method_name.to_string(), - object_type: object_type.clone(), - span, - }); + let has_field_with_function_type = + typ.borrow().get_fields_as_written().into_iter().any(|field| { + field.name.0.contents == method_name && field.typ.is_function() + }); + if has_field_with_function_type { + self.push_err(TypeCheckError::CannotInvokeStructFieldFunctionType { + method_name: method_name.to_string(), + object_type: object_type.clone(), + span, + }); + } else { + self.push_err(TypeCheckError::UnresolvedMethodCall { + method_name: method_name.to_string(), + object_type: object_type.clone(), + span, + }); + } None } } diff --git a/noir/noir-repo/compiler/noirc_frontend/src/hir/resolution/errors.rs b/noir/noir-repo/compiler/noirc_frontend/src/hir/resolution/errors.rs index 80bd5247ee6..5c8e0a1b53e 100644 --- a/noir/noir-repo/compiler/noirc_frontend/src/hir/resolution/errors.rs +++ b/noir/noir-repo/compiler/noirc_frontend/src/hir/resolution/errors.rs @@ -223,11 +223,21 @@ impl<'a> From<&'a ResolverError> for Diagnostic { *span, ) } - ResolverError::VariableNotDeclared { name, span } => Diagnostic::simple_error( - format!("cannot find `{name}` in this scope "), - "not found in this scope".to_string(), - *span, - ), + ResolverError::VariableNotDeclared { name, span } => { + if name == "_" { + Diagnostic::simple_error( + "in expressions, `_` can only be used on the left-hand side of an assignment".to_string(), + "`_` not allowed here".to_string(), + *span, + ) + } else { + Diagnostic::simple_error( + format!("cannot find `{name}` in this scope"), + "not found in this scope".to_string(), + *span, + ) + } + }, ResolverError::PathIsNotIdent { span } => Diagnostic::simple_error( "cannot use path as an identifier".to_string(), String::new(), diff --git a/noir/noir-repo/compiler/noirc_frontend/src/hir/type_check/errors.rs b/noir/noir-repo/compiler/noirc_frontend/src/hir/type_check/errors.rs index a6b6120986e..dfa431157e3 100644 --- a/noir/noir-repo/compiler/noirc_frontend/src/hir/type_check/errors.rs +++ b/noir/noir-repo/compiler/noirc_frontend/src/hir/type_check/errors.rs @@ -99,6 +99,8 @@ pub enum TypeCheckError { CannotMutateImmutableVariable { name: String, span: Span }, #[error("No method named '{method_name}' found for type '{object_type}'")] UnresolvedMethodCall { method_name: String, object_type: Type, span: Span }, + #[error("Cannot invoke function field '{method_name}' on type '{object_type}' as a method")] + CannotInvokeStructFieldFunctionType { method_name: String, object_type: Type, span: Span }, #[error("Integers must have the same signedness LHS is {sign_x:?}, RHS is {sign_y:?}")] IntegerSignedness { sign_x: Signedness, sign_y: Signedness, span: Span }, #[error("Integers must have the same bit width LHS is {bit_width_x}, RHS is {bit_width_y}")] @@ -511,6 +513,13 @@ impl<'a> From<&'a TypeCheckError> for Diagnostic { TypeCheckError::CyclicType { typ: _, span } => { Diagnostic::simple_error(error.to_string(), "Cyclic types have unlimited size and are prohibited in Noir".into(), *span) } + TypeCheckError::CannotInvokeStructFieldFunctionType { method_name, object_type, span } => { + Diagnostic::simple_error( + format!("Cannot invoke function field '{method_name}' on type '{object_type}' as a method"), + format!("to call the function stored in '{method_name}', surround the field access with parentheses: '(', ')'"), + *span, + ) + }, } } } diff --git a/noir/noir-repo/compiler/noirc_frontend/src/hir_def/types.rs b/noir/noir-repo/compiler/noirc_frontend/src/hir_def/types.rs index 659fafbbcbb..2c9a44c079d 100644 --- a/noir/noir-repo/compiler/noirc_frontend/src/hir_def/types.rs +++ b/noir/noir-repo/compiler/noirc_frontend/src/hir_def/types.rs @@ -1087,6 +1087,14 @@ impl Type { } } + pub fn is_function(&self) -> bool { + match self.follow_bindings_shallow().as_ref() { + Type::Function(..) => true, + Type::Alias(alias_type, _) => alias_type.borrow().typ.is_function(), + _ => false, + } + } + /// True if this type can be used as a parameter to `main` or a contract function. /// This is only false for unsized types like slices or slices that do not make sense /// as a program input such as named generics or mutable references. diff --git a/noir/noir-repo/compiler/noirc_frontend/src/monomorphization/ast.rs b/noir/noir-repo/compiler/noirc_frontend/src/monomorphization/ast.rs index 8f6817dc15d..5d9b66f4f96 100644 --- a/noir/noir-repo/compiler/noirc_frontend/src/monomorphization/ast.rs +++ b/noir/noir-repo/compiler/noirc_frontend/src/monomorphization/ast.rs @@ -48,6 +48,12 @@ pub enum Expression { Continue, } +impl Expression { + pub fn is_array_or_slice_literal(&self) -> bool { + matches!(self, Expression::Literal(Literal::Array(_) | Literal::Slice(_))) + } +} + /// A definition is either a local (variable), function, or is a built-in /// function that will be generated or referenced by the compiler later. #[derive(Debug, Clone, PartialEq, Eq, Hash)] diff --git a/noir/noir-repo/compiler/noirc_frontend/src/parser/parser/expression.rs b/noir/noir-repo/compiler/noirc_frontend/src/parser/parser/expression.rs index 06f51b16842..e1ecc972eeb 100644 --- a/noir/noir-repo/compiler/noirc_frontend/src/parser/parser/expression.rs +++ b/noir/noir-repo/compiler/noirc_frontend/src/parser/parser/expression.rs @@ -428,8 +428,8 @@ impl<'a> Parser<'a> { Some(if self.eat_colon() { let expression = self.parse_expression_or_error(); (ident, expression) - } else if self.at(Token::Assign) { - // If we find '=' instead of ':', assume the user meant ':`, error and continue + } else if self.at(Token::DoubleColon) || self.at(Token::Assign) { + // If we find '=' or '::' instead of ':', assume the user meant ':`, error and continue self.expected_token(Token::Colon); self.bump(); let expression = self.parse_expression_or_error(); @@ -1369,6 +1369,34 @@ mod tests { assert_eq!(expr.to_string(), "y"); } + #[test] + fn parses_constructor_recovers_if_double_colon_instead_of_colon() { + let src = " + Foo { x: 1, y:: z } + ^^ + "; + let (src, span) = get_source_with_error_span(src); + let mut parser = Parser::for_str(&src); + let expr = parser.parse_expression_or_error(); + + let error = get_single_error(&parser.errors, span); + assert_eq!(error.to_string(), "Expected a ':' but found '::'"); + + let ExpressionKind::Constructor(mut constructor) = expr.kind else { + panic!("Expected constructor"); + }; + assert_eq!(constructor.typ.to_string(), "Foo"); + assert_eq!(constructor.fields.len(), 2); + + let (name, expr) = constructor.fields.remove(0); + assert_eq!(name.to_string(), "x"); + assert_eq!(expr.to_string(), "1"); + + let (name, expr) = constructor.fields.remove(0); + assert_eq!(name.to_string(), "y"); + assert_eq!(expr.to_string(), "z"); + } + #[test] fn parses_parses_if_true() { let src = "if true { 1 }"; diff --git a/noir/noir-repo/compiler/noirc_frontend/src/tests.rs b/noir/noir-repo/compiler/noirc_frontend/src/tests.rs index 605236c8dda..cba29d58ea3 100644 --- a/noir/noir-repo/compiler/noirc_frontend/src/tests.rs +++ b/noir/noir-repo/compiler/noirc_frontend/src/tests.rs @@ -3752,6 +3752,35 @@ fn allows_struct_with_generic_infix_type_as_main_input_3() { assert_no_errors(src); } +#[test] +fn errors_with_better_message_when_trying_to_invoke_struct_field_that_is_a_function() { + let src = r#" + pub struct Foo { + wrapped: fn(Field) -> bool, + } + + impl Foo { + fn call(self) -> bool { + self.wrapped(1) + } + } + + fn main() {} + "#; + let errors = get_program_errors(src); + assert_eq!(errors.len(), 1); + + let CompilationError::TypeError(TypeCheckError::CannotInvokeStructFieldFunctionType { + method_name, + .. + }) = &errors[0].0 + else { + panic!("Expected a 'CannotInvokeStructFieldFunctionType' error, got {:?}", errors[0].0); + }; + + assert_eq!(method_name, "wrapped"); +} + fn test_disallows_attribute_on_impl_method( attr: &str, check_error: impl FnOnce(&CompilationError), @@ -3845,3 +3874,33 @@ fn disallows_export_attribute_on_trait_impl_method() { )); }); } + +#[test] +fn allows_multiple_underscore_parameters() { + let src = r#" + pub fn foo(_: i32, _: i64) {} + + fn main() {} + "#; + assert_no_errors(src); +} + +#[test] +fn disallows_underscore_on_right_hand_side() { + let src = r#" + fn main() { + let _ = 1; + let _x = _; + } + "#; + let errors = get_program_errors(src); + assert_eq!(errors.len(), 1); + + let CompilationError::ResolverError(ResolverError::VariableNotDeclared { name, .. }) = + &errors[0].0 + else { + panic!("Expected a VariableNotDeclared error, got {:?}", errors[0].0); + }; + + assert_eq!(name, "_"); +} diff --git a/noir/noir-repo/compiler/noirc_frontend/src/tests/metaprogramming.rs b/noir/noir-repo/compiler/noirc_frontend/src/tests/metaprogramming.rs index 82c40203244..89a049ebc9d 100644 --- a/noir/noir-repo/compiler/noirc_frontend/src/tests/metaprogramming.rs +++ b/noir/noir-repo/compiler/noirc_frontend/src/tests/metaprogramming.rs @@ -141,3 +141,23 @@ fn errors_if_macros_inject_functions_with_name_collisions() { ) if contents == "foo" )); } + +#[test] +fn uses_correct_type_for_attribute_arguments() { + let src = r#" + #[foo(32)] + comptime fn foo(_f: FunctionDefinition, i: u32) { + let y: u32 = 1; + let _ = y == i; + } + + #[bar([0; 2])] + comptime fn bar(_f: FunctionDefinition, i: [u32; 2]) { + let y: u32 = 1; + let _ = y == i[0]; + } + + fn main() {} + "#; + assert_no_errors(src); +} diff --git a/noir/noir-repo/compiler/noirc_printable_type/src/lib.rs b/noir/noir-repo/compiler/noirc_printable_type/src/lib.rs index 5ab04c6f576..838a2472125 100644 --- a/noir/noir-repo/compiler/noirc_printable_type/src/lib.rs +++ b/noir/noir-repo/compiler/noirc_printable_type/src/lib.rs @@ -69,6 +69,9 @@ pub enum PrintableValueDisplay { #[derive(Debug, Error)] pub enum ForeignCallError { + #[error("No handler could be found for foreign call `{0}`")] + NoHandler(String), + #[error("Foreign call inputs needed for execution are missing")] MissingForeignCallInputs, diff --git a/noir/noir-repo/compiler/wasm/Cargo.toml b/noir/noir-repo/compiler/wasm/Cargo.toml index c8b8c3bb06e..9951b23f609 100644 --- a/noir/noir-repo/compiler/wasm/Cargo.toml +++ b/noir/noir-repo/compiler/wasm/Cargo.toml @@ -1,10 +1,12 @@ [package] name = "noir_wasm" +description = "A JS interface to the Noir compiler" version.workspace = true authors.workspace = true edition.workspace = true rust-version.workspace = true license.workspace = true +repository.workspace = true [lints] workspace = true @@ -42,4 +44,4 @@ getrandom = { workspace = true, features = ["js"] } rust-embed = { workspace = true, features = ["debug-embed"] } [build-dependencies] -build-data.workspace = true \ No newline at end of file +build-data.workspace = true diff --git a/noir/noir-repo/compiler/wasm/LICENSE-APACHE b/noir/noir-repo/compiler/wasm/LICENSE-APACHE new file mode 100644 index 00000000000..261eeb9e9f8 --- /dev/null +++ b/noir/noir-repo/compiler/wasm/LICENSE-APACHE @@ -0,0 +1,201 @@ + Apache License + Version 2.0, January 2004 + http://www.apache.org/licenses/ + + TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION + + 1. Definitions. + + "License" shall mean the terms and conditions for use, reproduction, + and distribution as defined by Sections 1 through 9 of this document. + + "Licensor" shall mean the copyright owner or entity authorized by + the copyright owner that is granting the License. + + "Legal Entity" shall mean the union of the acting entity and all + other entities that control, are controlled by, or are under common + control with that entity. For the purposes of this definition, + "control" means (i) the power, direct or indirect, to cause the + direction or management of such entity, whether by contract or + otherwise, or (ii) ownership of fifty percent (50%) or more of the + outstanding shares, or (iii) beneficial ownership of such entity. + + "You" (or "Your") shall mean an individual or Legal Entity + exercising permissions granted by this License. + + "Source" form shall mean the preferred form for making modifications, + including but not limited to software source code, documentation + source, and configuration files. + + "Object" form shall mean any form resulting from mechanical + transformation or translation of a Source form, including but + not limited to compiled object code, generated documentation, + and conversions to other media types. + + "Work" shall mean the work of authorship, whether in Source or + Object form, made available under the License, as indicated by a + copyright notice that is included in or attached to the work + (an example is provided in the Appendix below). + + "Derivative Works" shall mean any work, whether in Source or Object + form, that is based on (or derived from) the Work and for which the + editorial revisions, annotations, elaborations, or other modifications + represent, as a whole, an original work of authorship. For the purposes + of this License, Derivative Works shall not include works that remain + separable from, or merely link (or bind by name) to the interfaces of, + the Work and Derivative Works thereof. + + "Contribution" shall mean any work of authorship, including + the original version of the Work and any modifications or additions + to that Work or Derivative Works thereof, that is intentionally + submitted to Licensor for inclusion in the Work by the copyright owner + or by an individual or Legal Entity authorized to submit on behalf of + the copyright owner. For the purposes of this definition, "submitted" + means any form of electronic, verbal, or written communication sent + to the Licensor or its representatives, including but not limited to + communication on electronic mailing lists, source code control systems, + and issue tracking systems that are managed by, or on behalf of, the + Licensor for the purpose of discussing and improving the Work, but + excluding communication that is conspicuously marked or otherwise + designated in writing by the copyright owner as "Not a Contribution." + + "Contributor" shall mean Licensor and any individual or Legal Entity + on behalf of whom a Contribution has been received by Licensor and + subsequently incorporated within the Work. + + 2. Grant of Copyright License. Subject to the terms and conditions of + this License, each Contributor hereby grants to You a perpetual, + worldwide, non-exclusive, no-charge, royalty-free, irrevocable + copyright license to reproduce, prepare Derivative Works of, + publicly display, publicly perform, sublicense, and distribute the + Work and such Derivative Works in Source or Object form. + + 3. Grant of Patent License. Subject to the terms and conditions of + this License, each Contributor hereby grants to You a perpetual, + worldwide, non-exclusive, no-charge, royalty-free, irrevocable + (except as stated in this section) patent license to make, have made, + use, offer to sell, sell, import, and otherwise transfer the Work, + where such license applies only to those patent claims licensable + by such Contributor that are necessarily infringed by their + Contribution(s) alone or by combination of their Contribution(s) + with the Work to which such Contribution(s) was submitted. If You + institute patent litigation against any entity (including a + cross-claim or counterclaim in a lawsuit) alleging that the Work + or a Contribution incorporated within the Work constitutes direct + or contributory patent infringement, then any patent licenses + granted to You under this License for that Work shall terminate + as of the date such litigation is filed. + + 4. Redistribution. You may reproduce and distribute copies of the + Work or Derivative Works thereof in any medium, with or without + modifications, and in Source or Object form, provided that You + meet the following conditions: + + (a) You must give any other recipients of the Work or + Derivative Works a copy of this License; and + + (b) You must cause any modified files to carry prominent notices + stating that You changed the files; and + + (c) You must retain, in the Source form of any Derivative Works + that You distribute, all copyright, patent, trademark, and + attribution notices from the Source form of the Work, + excluding those notices that do not pertain to any part of + the Derivative Works; and + + (d) If the Work includes a "NOTICE" text file as part of its + distribution, then any Derivative Works that You distribute must + include a readable copy of the attribution notices contained + within such NOTICE file, excluding those notices that do not + pertain to any part of the Derivative Works, in at least one + of the following places: within a NOTICE text file distributed + as part of the Derivative Works; within the Source form or + documentation, if provided along with the Derivative Works; or, + within a display generated by the Derivative Works, if and + wherever such third-party notices normally appear. The contents + of the NOTICE file are for informational purposes only and + do not modify the License. You may add Your own attribution + notices within Derivative Works that You distribute, alongside + or as an addendum to the NOTICE text from the Work, provided + that such additional attribution notices cannot be construed + as modifying the License. + + You may add Your own copyright statement to Your modifications and + may provide additional or different license terms and conditions + for use, reproduction, or distribution of Your modifications, or + for any such Derivative Works as a whole, provided Your use, + reproduction, and distribution of the Work otherwise complies with + the conditions stated in this License. + + 5. Submission of Contributions. Unless You explicitly state otherwise, + any Contribution intentionally submitted for inclusion in the Work + by You to the Licensor shall be under the terms and conditions of + this License, without any additional terms or conditions. + Notwithstanding the above, nothing herein shall supersede or modify + the terms of any separate license agreement you may have executed + with Licensor regarding such Contributions. + + 6. Trademarks. This License does not grant permission to use the trade + names, trademarks, service marks, or product names of the Licensor, + except as required for reasonable and customary use in describing the + origin of the Work and reproducing the content of the NOTICE file. + + 7. Disclaimer of Warranty. Unless required by applicable law or + agreed to in writing, Licensor provides the Work (and each + Contributor provides its Contributions) on an "AS IS" BASIS, + WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or + implied, including, without limitation, any warranties or conditions + of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A + PARTICULAR PURPOSE. You are solely responsible for determining the + appropriateness of using or redistributing the Work and assume any + risks associated with Your exercise of permissions under this License. + + 8. Limitation of Liability. In no event and under no legal theory, + whether in tort (including negligence), contract, or otherwise, + unless required by applicable law (such as deliberate and grossly + negligent acts) or agreed to in writing, shall any Contributor be + liable to You for damages, including any direct, indirect, special, + incidental, or consequential damages of any character arising as a + result of this License or out of the use or inability to use the + Work (including but not limited to damages for loss of goodwill, + work stoppage, computer failure or malfunction, or any and all + other commercial damages or losses), even if such Contributor + has been advised of the possibility of such damages. + + 9. Accepting Warranty or Additional Liability. While redistributing + the Work or Derivative Works thereof, You may choose to offer, + and charge a fee for, acceptance of support, warranty, indemnity, + or other liability obligations and/or rights consistent with this + License. However, in accepting such obligations, You may act only + on Your own behalf and on Your sole responsibility, not on behalf + of any other Contributor, and only if You agree to indemnify, + defend, and hold each Contributor harmless for any liability + incurred by, or claims asserted against, such Contributor by reason + of your accepting any such warranty or additional liability. + + END OF TERMS AND CONDITIONS + + APPENDIX: How to apply the Apache License to your work. + + To apply the Apache License to your work, attach the following + boilerplate notice, with the fields enclosed by brackets "[]" + replaced with your own identifying information. (Don't include + the brackets!) The text should be enclosed in the appropriate + comment syntax for the file format. We also recommend that a + file or class name and description of purpose be included on the + same "printed page" as the copyright notice for easier + identification within third-party archives. + + Copyright [yyyy] [name of copyright owner] + + Licensed under the Apache License, Version 2.0 (the "License"); + you may not use this file except in compliance with the License. + You may obtain a copy of the License at + + http://www.apache.org/licenses/LICENSE-2.0 + + Unless required by applicable law or agreed to in writing, software + distributed under the License is distributed on an "AS IS" BASIS, + WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + See the License for the specific language governing permissions and + limitations under the License. diff --git a/noir/noir-repo/compiler/wasm/LICENSE-MIT b/noir/noir-repo/compiler/wasm/LICENSE-MIT new file mode 100644 index 00000000000..a93d7f55c8e --- /dev/null +++ b/noir/noir-repo/compiler/wasm/LICENSE-MIT @@ -0,0 +1,21 @@ +MIT License + + Copyright (c) 2021-2023 noir-lang + + Permission is hereby granted, free of charge, to any person obtaining a copy + of this software and associated documentation files (the "Software"), to deal + in the Software without restriction, including without limitation the rights + to use, copy, modify, merge, publish, distribute, sublicense, and/or sell + copies of the Software, and to permit persons to whom the Software is + furnished to do so, subject to the following conditions: + + The above copyright notice and this permission notice shall be included in all + copies or substantial portions of the Software. + + THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE + AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER + LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, + OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE + SOFTWARE. diff --git a/noir/noir-repo/compiler/wasm/tsconfig.json b/noir/noir-repo/compiler/wasm/tsconfig.json index d2ae58b8fc9..42c7396aa83 100644 --- a/noir/noir-repo/compiler/wasm/tsconfig.json +++ b/noir/noir-repo/compiler/wasm/tsconfig.json @@ -18,4 +18,4 @@ "allowJs": true, }, "exclude": ["node_modules"] -} \ No newline at end of file +} diff --git a/noir/noir-repo/cspell.json b/noir/noir-repo/cspell.json index 36bba737cd7..5c707e92e21 100644 --- a/noir/noir-repo/cspell.json +++ b/noir/noir-repo/cspell.json @@ -106,6 +106,7 @@ "Guillaume", "gzipped", "hasher", + "heaptrack", "hexdigit", "higher-kinded", "Hindley-Milner", @@ -154,6 +155,7 @@ "nargo", "neovim", "newtype", + "nextest", "nightlies", "nixpkgs", "noirc", diff --git a/noir/noir-repo/docs/docs/noir/concepts/data_types/integers.md b/noir/noir-repo/docs/docs/noir/concepts/data_types/integers.md index f3badde62be..41a823646dd 100644 --- a/noir/noir-repo/docs/docs/noir/concepts/data_types/integers.md +++ b/noir/noir-repo/docs/docs/noir/concepts/data_types/integers.md @@ -79,7 +79,7 @@ fn main() { You can construct a U128 from its limbs: ```rust fn main(x: u64, y: u64) { - let x = U128::from_u64s_be(x,y); + let z = U128::from_u64s_be(x,y); assert(z.hi == x as Field); assert(z.lo == y as Field); } diff --git a/noir/noir-repo/docs/docs/noir/standard_library/containers/boundedvec.md b/noir/noir-repo/docs/docs/noir/standard_library/containers/boundedvec.md index 98b7d584033..4efb1e4ea0f 100644 --- a/noir/noir-repo/docs/docs/noir/standard_library/containers/boundedvec.md +++ b/noir/noir-repo/docs/docs/noir/standard_library/containers/boundedvec.md @@ -246,6 +246,42 @@ Example: let bounded_vec: BoundedVec = BoundedVec::from_array([1, 2, 3]) ``` +### from_parts + +```rust +pub fn from_parts(mut array: [T; MaxLen], len: u32) -> Self +``` + +Creates a new BoundedVec from the given array and length. +The given length must be less than or equal to the length of the array. + +This function will zero out any elements at or past index `len` of `array`. +This incurs an extra runtime cost of O(MaxLen). If you are sure your array is +zeroed after that index, you can use `from_parts_unchecked` to remove the extra loop. + +Example: + +#include_code from-parts noir_stdlib/src/collections/bounded_vec.nr rust + +### from_parts_unchecked + +```rust +pub fn from_parts_unchecked(array: [T; MaxLen], len: u32) -> Self +``` + +Creates a new BoundedVec from the given array and length. +The given length must be less than or equal to the length of the array. + +This function is unsafe because it expects all elements past the `len` index +of `array` to be zeroed, but does not check for this internally. Use `from_parts` +for a safe version of this function which does zero out any indices past the +given length. Invalidating this assumption can notably cause `BoundedVec::eq` +to give incorrect results since it will check even elements past `len`. + +Example: + +#include_code from-parts-unchecked noir_stdlib/src/collections/bounded_vec.nr rust + ### map ```rust diff --git a/noir/noir-repo/noir_stdlib/src/collections/bounded_vec.nr b/noir/noir-repo/noir_stdlib/src/collections/bounded_vec.nr index f33890f197e..0ad39c518c4 100644 --- a/noir/noir-repo/noir_stdlib/src/collections/bounded_vec.nr +++ b/noir/noir-repo/noir_stdlib/src/collections/bounded_vec.nr @@ -420,6 +420,58 @@ impl BoundedVec { } ret } + + /// Creates a new BoundedVec from the given array and length. + /// The given length must be less than or equal to the length of the array. + /// + /// This function will zero out any elements at or past index `len` of `array`. + /// This incurs an extra runtime cost of O(MaxLen). If you are sure your array is + /// zeroed after that index, you can use `from_parts_unchecked` to remove the extra loop. + /// + /// Example: + /// + /// ```noir + /// let vec: BoundedVec = BoundedVec::from_parts([1, 2, 3, 0], 3); + /// assert_eq(vec.len(), 3); + /// ``` + pub fn from_parts(mut array: [T; MaxLen], len: u32) -> Self { + assert(len <= MaxLen); + let zeroed = crate::mem::zeroed(); + for i in 0..MaxLen { + if i >= len { + array[i] = zeroed; + } + } + BoundedVec { storage: array, len } + } + + /// Creates a new BoundedVec from the given array and length. + /// The given length must be less than or equal to the length of the array. + /// + /// This function is unsafe because it expects all elements past the `len` index + /// of `array` to be zeroed, but does not check for this internally. Use `from_parts` + /// for a safe version of this function which does zero out any indices past the + /// given length. Invalidating this assumption can notably cause `BoundedVec::eq` + /// to give incorrect results since it will check even elements past `len`. + /// + /// Example: + /// + /// ```noir + /// let vec: BoundedVec = BoundedVec::from_parts_unchecked([1, 2, 3, 0], 3); + /// assert_eq(vec.len(), 3); + /// + /// // invalid use! + /// let vec1: BoundedVec = BoundedVec::from_parts_unchecked([1, 2, 3, 1], 3); + /// let vec2: BoundedVec = BoundedVec::from_parts_unchecked([1, 2, 3, 2], 3); + /// + /// // both vecs have length 3 so we'd expect them to be equal, but this + /// // fails because elements past the length are still checked in eq + /// assert_eq(vec1, vec2); // fails + /// ``` + pub fn from_parts_unchecked(array: [T; MaxLen], len: u32) -> Self { + assert(len <= MaxLen); + BoundedVec { storage: array, len } + } } impl Eq for BoundedVec @@ -431,7 +483,11 @@ where // // We make the assumption that the user has used the proper interface for working with `BoundedVec`s // rather than directly manipulating the internal fields as this can result in an inconsistent internal state. - (self.len == other.len) & (self.storage == other.storage) + if self.len == other.len { + self.storage == other.storage + } else { + false + } } } @@ -598,4 +654,38 @@ mod bounded_vec_tests { assert(bounded_vec1 != bounded_vec2); } } + + mod from_parts { + use crate::collections::bounded_vec::BoundedVec; + + #[test] + fn from_parts() { + // docs:start:from-parts + let vec: BoundedVec = BoundedVec::from_parts([1, 2, 3, 0], 3); + assert_eq(vec.len(), 3); + + // Any elements past the given length are zeroed out, so these + // two BoundedVecs will be completely equal + let vec1: BoundedVec = BoundedVec::from_parts([1, 2, 3, 1], 3); + let vec2: BoundedVec = BoundedVec::from_parts([1, 2, 3, 2], 3); + assert_eq(vec1, vec2); + // docs:end:from-parts + } + + #[test] + fn from_parts_unchecked() { + // docs:start:from-parts-unchecked + let vec: BoundedVec = BoundedVec::from_parts_unchecked([1, 2, 3, 0], 3); + assert_eq(vec.len(), 3); + + // invalid use! + let vec1: BoundedVec = BoundedVec::from_parts_unchecked([1, 2, 3, 1], 3); + let vec2: BoundedVec = BoundedVec::from_parts_unchecked([1, 2, 3, 2], 3); + + // both vecs have length 3 so we'd expect them to be equal, but this + // fails because elements past the length are still checked in eq + assert(vec1 != vec2); + // docs:end:from-parts-unchecked + } + } } diff --git a/noir/noir-repo/noir_stdlib/src/hash/poseidon2.nr b/noir/noir-repo/noir_stdlib/src/hash/poseidon2.nr index f2167c43c2c..419f07a2aca 100644 --- a/noir/noir-repo/noir_stdlib/src/hash/poseidon2.nr +++ b/noir/noir-repo/noir_stdlib/src/hash/poseidon2.nr @@ -13,11 +13,7 @@ pub struct Poseidon2 { impl Poseidon2 { #[no_predicates] pub fn hash(input: [Field; N], message_size: u32) -> Field { - if message_size == N { - Poseidon2::hash_internal(input, N, false) - } else { - Poseidon2::hash_internal(input, message_size, true) - } + Poseidon2::hash_internal(input, message_size, message_size != N) } pub(crate) fn new(iv: Field) -> Poseidon2 { diff --git a/noir/noir-repo/test_programs/execution_success/array_dedup_regression/Nargo.toml b/noir/noir-repo/test_programs/execution_success/array_dedup_regression/Nargo.toml new file mode 100644 index 00000000000..16a708743ed --- /dev/null +++ b/noir/noir-repo/test_programs/execution_success/array_dedup_regression/Nargo.toml @@ -0,0 +1,6 @@ +[package] +name = "array_dedup_regression" +type = "bin" +authors = [""] + +[dependencies] diff --git a/noir/noir-repo/test_programs/execution_success/array_dedup_regression/Prover.toml b/noir/noir-repo/test_programs/execution_success/array_dedup_regression/Prover.toml new file mode 100644 index 00000000000..3aea0c58ce5 --- /dev/null +++ b/noir/noir-repo/test_programs/execution_success/array_dedup_regression/Prover.toml @@ -0,0 +1 @@ +x = 0 diff --git a/noir/noir-repo/test_programs/execution_success/array_dedup_regression/src/main.nr b/noir/noir-repo/test_programs/execution_success/array_dedup_regression/src/main.nr new file mode 100644 index 00000000000..5506d55b9e7 --- /dev/null +++ b/noir/noir-repo/test_programs/execution_success/array_dedup_regression/src/main.nr @@ -0,0 +1,21 @@ +unconstrained fn main(x: u32) { + let a1 = [1, 2, 3, 4, 5]; + + for i in 0..5 { + let mut a2 = [1, 2, 3, 4, 5]; + a2[x + i] = 128; + println(a2); + + if i != 0 { + assert(a2[x + i - 1] != 128); + } + } + + // Can't use `== [1, 2, 3, 4, 5]` here, that make_array may get + // deduplicated to equal a1 in the bugged version + assert_eq(a1[0], 1); + assert_eq(a1[1], 2); + assert_eq(a1[2], 3); + assert_eq(a1[3], 4); + assert_eq(a1[4], 5); +} diff --git a/noir/noir-repo/test_programs/execution_success/inline_decompose_hint_brillig_call/Nargo.toml b/noir/noir-repo/test_programs/execution_success/inline_decompose_hint_brillig_call/Nargo.toml new file mode 100644 index 00000000000..ecac2dfb197 --- /dev/null +++ b/noir/noir-repo/test_programs/execution_success/inline_decompose_hint_brillig_call/Nargo.toml @@ -0,0 +1,7 @@ +[package] +name = "inline_decompose_hint_brillig_call" +version = "0.1.0" +type = "bin" +authors = [""] + +[dependencies] diff --git a/noir/noir-repo/test_programs/execution_success/inline_decompose_hint_brillig_call/src/main.nr b/noir/noir-repo/test_programs/execution_success/inline_decompose_hint_brillig_call/src/main.nr new file mode 100644 index 00000000000..e500f0f976d --- /dev/null +++ b/noir/noir-repo/test_programs/execution_success/inline_decompose_hint_brillig_call/src/main.nr @@ -0,0 +1,15 @@ +use std::embedded_curve_ops::{EmbeddedCurvePoint, EmbeddedCurveScalar, fixed_base_scalar_mul}; + +fn main() -> pub Field { + let pre_address = 0x23d95e303879a5d0bbef78ecbc335e559da37431f6dcd11da54ed375c2846813; + let (a, b) = std::field::bn254::decompose(pre_address); + let curve = EmbeddedCurveScalar { lo: a, hi: b }; + let key = fixed_base_scalar_mul(curve); + let point = EmbeddedCurvePoint { + x: 0x111223493147f6785514b1c195bb37a2589f22a6596d30bb2bb145fdc9ca8f1e, + y: 0x273bbffd678edce8fe30e0deafc4f66d58357c06fd4a820285294b9746c3be95, + is_infinite: false, + }; + let address_point = key.add(point); + address_point.x +} diff --git a/noir/noir-repo/test_programs/execution_success/loop_invariant_regression/src/main.nr b/noir/noir-repo/test_programs/execution_success/loop_invariant_regression/src/main.nr index 25f6e92f868..c28ce063116 100644 --- a/noir/noir-repo/test_programs/execution_success/loop_invariant_regression/src/main.nr +++ b/noir/noir-repo/test_programs/execution_success/loop_invariant_regression/src/main.nr @@ -2,6 +2,7 @@ // to be hoisted to the loop's pre-header block. fn main(x: u32, y: u32) { loop(4, x, y); + array_read_loop(4, x); } fn loop(upper_bound: u32, x: u32, y: u32) { @@ -11,3 +12,15 @@ fn loop(upper_bound: u32, x: u32, y: u32) { assert_eq(z, 12); } } + +fn array_read_loop(upper_bound: u32, x: u32) { + let arr = [2; 5]; + for i in 0..upper_bound { + for j in 0..upper_bound { + for _ in 0..upper_bound { + assert_eq(arr[i], x); + assert_eq(arr[j], x); + } + } + } +} diff --git a/noir/noir-repo/test_programs/execution_success/reference_counts/src/main.nr b/noir/noir-repo/test_programs/execution_success/reference_counts/src/main.nr index 7ab7de893fa..8de4d0f2508 100644 --- a/noir/noir-repo/test_programs/execution_success/reference_counts/src/main.nr +++ b/noir/noir-repo/test_programs/execution_success/reference_counts/src/main.nr @@ -1,10 +1,19 @@ +use std::mem::array_refcount; + fn main() { let mut array = [0, 1, 2]; assert_refcount(array, 1); - borrow(array, std::mem::array_refcount(array)); - borrow_mut(&mut array, std::mem::array_refcount(array)); - copy_mut(array, std::mem::array_refcount(array)); + borrow(array, array_refcount(array)); + borrow_mut(&mut array, array_refcount(array)); + copy_mut(array, array_refcount(array)); + + borrow_mut_two(&mut array, &mut array, array_refcount(array)); + + let mut u32_array = [0, 1, 2]; + let rc1 = array_refcount(array); + let rc2 = array_refcount(u32_array); + borrow_mut_two_separate(&mut array, &mut u32_array, rc1, rc2); } fn borrow(array: [Field; 3], rc_before_call: u32) { @@ -13,19 +22,48 @@ fn borrow(array: [Field; 3], rc_before_call: u32) { } fn borrow_mut(array: &mut [Field; 3], rc_before_call: u32) { - assert_refcount(*array, rc_before_call + 0); // Issue! This should be rc_before_call + 1 - array[0] = 5; + // Optimization: inc_rc isn't needed since there is only one array (`array`) + // of the same type that `array` can be modified through + assert_refcount(*array, rc_before_call + 0); + array[0] = 3; println(array[0]); } fn copy_mut(mut array: [Field; 3], rc_before_call: u32) { - assert_refcount(array, rc_before_call + 0); // Issue! This should be rc_before_call + 1 - array[0] = 6; + assert_refcount(array, rc_before_call + 1); + array[0] = 4; println(array[0]); } -fn assert_refcount(array: [Field; 3], expected: u32) { - let count = std::mem::array_refcount(array); +/// Borrow the same array mutably through both parameters, inc_rc is necessary here, although +/// only one is needed to bring the rc from 1 to 2. +fn borrow_mut_two(array1: &mut [Field; 3], array2: &mut [Field; 3], rc_before_call: u32) { + assert_refcount(*array1, rc_before_call + 1); + assert_refcount(*array2, rc_before_call + 1); + array1[0] = 5; + array2[0] = 6; + println(array1[0]); // array1 & 2 alias, so this should also print 6 + println(array2[0]); +} + +/// Borrow a different array: we should be able to reason that these types cannot be mutably +/// aliased since they're different types so we don't need any inc_rc instructions. +fn borrow_mut_two_separate( + array1: &mut [Field; 3], + array2: &mut [u32; 3], + rc_before_call1: u32, + rc_before_call2: u32, +) { + assert_refcount(*array1, rc_before_call1 + 0); + assert_refcount(*array2, rc_before_call2 + 0); + array1[0] = 7; + array2[0] = 8; + println(array1[0]); + println(array2[0]); +} + +fn assert_refcount(array: [T; 3], expected: u32) { + let count = array_refcount(array); // All refcounts are zero when running this as a constrained program if std::runtime::is_unconstrained() { diff --git a/noir/noir-repo/test_programs/gates_report_brillig.sh b/noir/noir-repo/test_programs/gates_report_brillig.sh old mode 100644 new mode 100755 diff --git a/noir/noir-repo/test_programs/gates_report_brillig_execution.sh b/noir/noir-repo/test_programs/gates_report_brillig_execution.sh old mode 100644 new mode 100755 diff --git a/noir/noir-repo/test_programs/memory_report.sh b/noir/noir-repo/test_programs/memory_report.sh new file mode 100755 index 00000000000..1b8274b76cc --- /dev/null +++ b/noir/noir-repo/test_programs/memory_report.sh @@ -0,0 +1,48 @@ +#!/usr/bin/env bash +set -e + +sudo apt-get install heaptrack + +NARGO="nargo" + + +# Tests to be profiled for memory report +tests_to_profile=("keccak256" "workspace" "regression_4709" "ram_blowup_regression") + +current_dir=$(pwd) +execution_success_path="$current_dir/execution_success" +test_dirs=$(ls $execution_success_path) + +FIRST="1" + +echo "{\"memory_reports\": [ " > memory_report.json + + +for test_name in ${tests_to_profile[@]}; do + full_path=$execution_success_path"/"$test_name + cd $full_path + + if [ $FIRST = "1" ] + then + FIRST="0" + else + echo " ," >> $current_dir"/memory_report.json" + fi + heaptrack --output $current_dir/$test_name"_heap" $NARGO compile --force + if test -f $current_dir/$test_name"_heap.gz"; + then + heaptrack --analyze $current_dir/$test_name"_heap.gz" > $current_dir/$test_name"_heap_analysis.txt" + rm $current_dir/$test_name"_heap.gz" + else + heaptrack --analyze $current_dir/$test_name"_heap.zst" > $current_dir/$test_name"_heap_analysis.txt" + rm $current_dir/$test_name"_heap.zst" + fi + consumption="$(grep 'peak heap memory consumption' $current_dir/$test_name'_heap_analysis.txt')" + len=${#consumption}-30 + peak=${consumption:30:len} + rm $current_dir/$test_name"_heap_analysis.txt" + echo -e " {\n \"artifact_name\":\"$test_name\",\n \"peak_memory\":\"$peak\"\n }" >> $current_dir"/memory_report.json" +done + +echo "]}" >> $current_dir"/memory_report.json" + diff --git a/noir/noir-repo/tooling/acvm_cli/src/cli/execute_cmd.rs b/noir/noir-repo/tooling/acvm_cli/src/cli/execute_cmd.rs index c453936568c..bf5969718e5 100644 --- a/noir/noir-repo/tooling/acvm_cli/src/cli/execute_cmd.rs +++ b/noir/noir-repo/tooling/acvm_cli/src/cli/execute_cmd.rs @@ -8,7 +8,7 @@ use clap::Args; use crate::cli::fs::inputs::{read_bytecode_from_file, read_inputs_from_file}; use crate::errors::CliError; -use nargo::ops::{execute_program, DefaultForeignCallExecutor}; +use nargo::{foreign_calls::DefaultForeignCallExecutor, ops::execute_program}; use super::fs::witness::{create_output_witness_string, save_witness_to_dir}; diff --git a/noir/noir-repo/tooling/debugger/src/foreign_calls.rs b/noir/noir-repo/tooling/debugger/src/foreign_calls.rs index 6a773a4b348..ecf27a22f29 100644 --- a/noir/noir-repo/tooling/debugger/src/foreign_calls.rs +++ b/noir/noir-repo/tooling/debugger/src/foreign_calls.rs @@ -3,7 +3,7 @@ use acvm::{ pwg::ForeignCallWaitInfo, AcirField, FieldElement, }; -use nargo::ops::{DefaultForeignCallExecutor, ForeignCallExecutor}; +use nargo::foreign_calls::{DefaultForeignCallExecutor, ForeignCallExecutor}; use noirc_artifacts::debug::{DebugArtifact, DebugVars, StackFrame}; use noirc_errors::debug_info::{DebugFnId, DebugVarId}; use noirc_printable_type::ForeignCallError; diff --git a/noir/noir-repo/tooling/lsp/src/requests/completion/tests.rs b/noir/noir-repo/tooling/lsp/src/requests/completion/tests.rs index 9306e38a48a..97c7ad86d5a 100644 --- a/noir/noir-repo/tooling/lsp/src/requests/completion/tests.rs +++ b/noir/noir-repo/tooling/lsp/src/requests/completion/tests.rs @@ -1586,6 +1586,54 @@ fn main() { assert_eq!(changed, expected); } + #[test] + async fn test_auto_import_inserts_after_last_use_in_nested_module() { + let src = r#"mod foo { + pub mod bar { + pub fn hello_world() {} + } +} + +mod baz { + fn qux() {} +} + +mod other { + use baz::qux; + + fn main() { + hel>|< + } +}"#; + + let expected = r#"mod foo { + pub mod bar { + pub fn hello_world() {} + } +} + +mod baz { + fn qux() {} +} + +mod other { + use baz::qux; + use super::foo::bar::hello_world; + + fn main() { + hel + } +}"#; + let mut items = get_completions(src).await; + assert_eq!(items.len(), 1); + + let item = items.remove(0); + + let changed = + apply_text_edits(&src.replace(">|<", ""), &item.additional_text_edits.unwrap()); + assert_eq!(changed, expected); + } + #[test] async fn test_does_not_auto_import_test_functions() { let src = r#" diff --git a/noir/noir-repo/tooling/lsp/src/requests/test_run.rs b/noir/noir-repo/tooling/lsp/src/requests/test_run.rs index 50c699bb6a6..937fdcc0a5e 100644 --- a/noir/noir-repo/tooling/lsp/src/requests/test_run.rs +++ b/noir/noir-repo/tooling/lsp/src/requests/test_run.rs @@ -101,6 +101,11 @@ fn on_test_run_request_inner( result: "fail".to_string(), message: Some(message), }, + TestStatus::Skipped => NargoTestRunResult { + id: params.id.clone(), + result: "skipped".to_string(), + message: None, + }, TestStatus::CompileError(diag) => NargoTestRunResult { id: params.id.clone(), result: "error".to_string(), diff --git a/noir/noir-repo/tooling/lsp/src/use_segment_positions.rs b/noir/noir-repo/tooling/lsp/src/use_segment_positions.rs index f9a3f429029..246ff653245 100644 --- a/noir/noir-repo/tooling/lsp/src/use_segment_positions.rs +++ b/noir/noir-repo/tooling/lsp/src/use_segment_positions.rs @@ -318,7 +318,7 @@ fn new_use_completion_item_additional_text_edits( request: UseCompletionItemAdditionTextEditsRequest, ) -> Vec { let line = request.auto_import_line as u32; - let character = (request.nesting * 4) as u32; + let character = 0; let indent = " ".repeat(request.nesting * 4); let mut newlines = "\n"; @@ -331,6 +331,6 @@ fn new_use_completion_item_additional_text_edits( vec![TextEdit { range: Range { start: Position { line, character }, end: Position { line, character } }, - new_text: format!("use {};{}{}", request.full_path, newlines, indent), + new_text: format!("{}use {};{}", indent, request.full_path, newlines), }] } diff --git a/noir/noir-repo/tooling/nargo/src/foreign_calls/mocker.rs b/noir/noir-repo/tooling/nargo/src/foreign_calls/mocker.rs new file mode 100644 index 00000000000..c93d16bbaf6 --- /dev/null +++ b/noir/noir-repo/tooling/nargo/src/foreign_calls/mocker.rs @@ -0,0 +1,176 @@ +use acvm::{ + acir::brillig::{ForeignCallParam, ForeignCallResult}, + pwg::ForeignCallWaitInfo, + AcirField, +}; +use noirc_printable_type::{decode_string_value, ForeignCallError}; +use serde::{Deserialize, Serialize}; + +use super::{ForeignCall, ForeignCallExecutor}; + +/// This struct represents an oracle mock. It can be used for testing programs that use oracles. +#[derive(Debug, PartialEq, Eq, Clone)] +struct MockedCall { + /// The id of the mock, used to update or remove it + id: usize, + /// The oracle it's mocking + name: String, + /// Optionally match the parameters + params: Option>>, + /// The parameters with which the mock was last called + last_called_params: Option>>, + /// The result to return when this mock is called + result: ForeignCallResult, + /// How many times should this mock be called before it is removed + times_left: Option, +} + +impl MockedCall { + fn new(id: usize, name: String) -> Self { + Self { + id, + name, + params: None, + last_called_params: None, + result: ForeignCallResult { values: vec![] }, + times_left: None, + } + } +} + +impl MockedCall { + fn matches(&self, name: &str, params: &[ForeignCallParam]) -> bool { + self.name == name && (self.params.is_none() || self.params.as_deref() == Some(params)) + } +} + +#[derive(Debug, Default)] +pub(crate) struct MockForeignCallExecutor { + /// Mocks have unique ids used to identify them in Noir, allowing to update or remove them. + last_mock_id: usize, + /// The registered mocks + mocked_responses: Vec>, +} + +impl MockForeignCallExecutor { + fn extract_mock_id( + foreign_call_inputs: &[ForeignCallParam], + ) -> Result<(usize, &[ForeignCallParam]), ForeignCallError> { + let (id, params) = + foreign_call_inputs.split_first().ok_or(ForeignCallError::MissingForeignCallInputs)?; + let id = + usize::try_from(id.unwrap_field().try_to_u64().expect("value does not fit into u64")) + .expect("value does not fit into usize"); + Ok((id, params)) + } + + fn find_mock_by_id(&self, id: usize) -> Option<&MockedCall> { + self.mocked_responses.iter().find(|response| response.id == id) + } + + fn find_mock_by_id_mut(&mut self, id: usize) -> Option<&mut MockedCall> { + self.mocked_responses.iter_mut().find(|response| response.id == id) + } + + fn parse_string(param: &ForeignCallParam) -> String { + let fields: Vec<_> = param.fields().to_vec(); + decode_string_value(&fields) + } +} + +impl Deserialize<'a>> ForeignCallExecutor + for MockForeignCallExecutor +{ + fn execute( + &mut self, + foreign_call: &ForeignCallWaitInfo, + ) -> Result, ForeignCallError> { + let foreign_call_name = foreign_call.function.as_str(); + match ForeignCall::lookup(foreign_call_name) { + Some(ForeignCall::CreateMock) => { + let mock_oracle_name = Self::parse_string(&foreign_call.inputs[0]); + assert!(ForeignCall::lookup(&mock_oracle_name).is_none()); + let id = self.last_mock_id; + self.mocked_responses.push(MockedCall::new(id, mock_oracle_name)); + self.last_mock_id += 1; + + Ok(F::from(id).into()) + } + Some(ForeignCall::SetMockParams) => { + let (id, params) = Self::extract_mock_id(&foreign_call.inputs)?; + self.find_mock_by_id_mut(id) + .unwrap_or_else(|| panic!("Unknown mock id {}", id)) + .params = Some(params.to_vec()); + + Ok(ForeignCallResult::default()) + } + Some(ForeignCall::GetMockLastParams) => { + let (id, _) = Self::extract_mock_id(&foreign_call.inputs)?; + let mock = + self.find_mock_by_id(id).unwrap_or_else(|| panic!("Unknown mock id {}", id)); + + let last_called_params = mock + .last_called_params + .clone() + .unwrap_or_else(|| panic!("Mock {} was never called", mock.name)); + + Ok(last_called_params.into()) + } + Some(ForeignCall::SetMockReturns) => { + let (id, params) = Self::extract_mock_id(&foreign_call.inputs)?; + self.find_mock_by_id_mut(id) + .unwrap_or_else(|| panic!("Unknown mock id {}", id)) + .result = ForeignCallResult { values: params.to_vec() }; + + Ok(ForeignCallResult::default()) + } + Some(ForeignCall::SetMockTimes) => { + let (id, params) = Self::extract_mock_id(&foreign_call.inputs)?; + let times = + params[0].unwrap_field().try_to_u64().expect("Invalid bit size of times"); + + self.find_mock_by_id_mut(id) + .unwrap_or_else(|| panic!("Unknown mock id {}", id)) + .times_left = Some(times); + + Ok(ForeignCallResult::default()) + } + Some(ForeignCall::ClearMock) => { + let (id, _) = Self::extract_mock_id(&foreign_call.inputs)?; + self.mocked_responses.retain(|response| response.id != id); + Ok(ForeignCallResult::default()) + } + _ => { + let mock_response_position = self + .mocked_responses + .iter() + .position(|response| response.matches(foreign_call_name, &foreign_call.inputs)); + + if let Some(response_position) = mock_response_position { + // If the program has registered a mocked response to this oracle call then we prefer responding + // with that. + + let mock = self + .mocked_responses + .get_mut(response_position) + .expect("Invalid position of mocked response"); + + mock.last_called_params = Some(foreign_call.inputs.clone()); + + let result = mock.result.values.clone(); + + if let Some(times_left) = &mut mock.times_left { + *times_left -= 1; + if *times_left == 0 { + self.mocked_responses.remove(response_position); + } + } + + Ok(result.into()) + } else { + Err(ForeignCallError::NoHandler(foreign_call_name.to_string())) + } + } + } + } +} diff --git a/noir/noir-repo/tooling/nargo/src/foreign_calls/mod.rs b/noir/noir-repo/tooling/nargo/src/foreign_calls/mod.rs new file mode 100644 index 00000000000..16ed71e11e3 --- /dev/null +++ b/noir/noir-repo/tooling/nargo/src/foreign_calls/mod.rs @@ -0,0 +1,146 @@ +use std::path::PathBuf; + +use acvm::{acir::brillig::ForeignCallResult, pwg::ForeignCallWaitInfo, AcirField}; +use mocker::MockForeignCallExecutor; +use noirc_printable_type::ForeignCallError; +use print::PrintForeignCallExecutor; +use rand::Rng; +use rpc::RPCForeignCallExecutor; +use serde::{Deserialize, Serialize}; + +pub(crate) mod mocker; +pub(crate) mod print; +pub(crate) mod rpc; + +pub trait ForeignCallExecutor { + fn execute( + &mut self, + foreign_call: &ForeignCallWaitInfo, + ) -> Result, ForeignCallError>; +} + +/// This enumeration represents the Brillig foreign calls that are natively supported by nargo. +/// After resolution of a foreign call, nargo will restart execution of the ACVM +pub enum ForeignCall { + Print, + CreateMock, + SetMockParams, + GetMockLastParams, + SetMockReturns, + SetMockTimes, + ClearMock, +} + +impl std::fmt::Display for ForeignCall { + fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result { + write!(f, "{}", self.name()) + } +} + +impl ForeignCall { + pub(crate) fn name(&self) -> &'static str { + match self { + ForeignCall::Print => "print", + ForeignCall::CreateMock => "create_mock", + ForeignCall::SetMockParams => "set_mock_params", + ForeignCall::GetMockLastParams => "get_mock_last_params", + ForeignCall::SetMockReturns => "set_mock_returns", + ForeignCall::SetMockTimes => "set_mock_times", + ForeignCall::ClearMock => "clear_mock", + } + } + + pub(crate) fn lookup(op_name: &str) -> Option { + match op_name { + "print" => Some(ForeignCall::Print), + "create_mock" => Some(ForeignCall::CreateMock), + "set_mock_params" => Some(ForeignCall::SetMockParams), + "get_mock_last_params" => Some(ForeignCall::GetMockLastParams), + "set_mock_returns" => Some(ForeignCall::SetMockReturns), + "set_mock_times" => Some(ForeignCall::SetMockTimes), + "clear_mock" => Some(ForeignCall::ClearMock), + _ => None, + } + } +} + +#[derive(Debug, Default)] +pub struct DefaultForeignCallExecutor { + /// The executor for any [`ForeignCall::Print`] calls. + printer: Option, + mocker: MockForeignCallExecutor, + external: Option, +} + +impl DefaultForeignCallExecutor { + pub fn new( + show_output: bool, + resolver_url: Option<&str>, + root_path: Option, + package_name: Option, + ) -> Self { + let id = rand::thread_rng().gen(); + let printer = if show_output { Some(PrintForeignCallExecutor) } else { None }; + let external_resolver = resolver_url.map(|resolver_url| { + RPCForeignCallExecutor::new(resolver_url, id, root_path, package_name) + }); + DefaultForeignCallExecutor { + printer, + mocker: MockForeignCallExecutor::default(), + external: external_resolver, + } + } +} + +impl Deserialize<'a>> ForeignCallExecutor + for DefaultForeignCallExecutor +{ + fn execute( + &mut self, + foreign_call: &ForeignCallWaitInfo, + ) -> Result, ForeignCallError> { + let foreign_call_name = foreign_call.function.as_str(); + match ForeignCall::lookup(foreign_call_name) { + Some(ForeignCall::Print) => { + if let Some(printer) = &mut self.printer { + printer.execute(foreign_call) + } else { + Ok(ForeignCallResult::default()) + } + } + Some( + ForeignCall::CreateMock + | ForeignCall::SetMockParams + | ForeignCall::GetMockLastParams + | ForeignCall::SetMockReturns + | ForeignCall::SetMockTimes + | ForeignCall::ClearMock, + ) => self.mocker.execute(foreign_call), + + None => { + // First check if there's any defined mock responses for this foreign call. + match self.mocker.execute(foreign_call) { + Err(ForeignCallError::NoHandler(_)) => (), + response_or_error => return response_or_error, + }; + + if let Some(external_resolver) = &mut self.external { + // If the user has registered an external resolver then we forward any remaining oracle calls there. + match external_resolver.execute(foreign_call) { + Err(ForeignCallError::NoHandler(_)) => (), + response_or_error => return response_or_error, + }; + } + + // If all executors have no handler for the given foreign call then we cannot + // return a correct response to the ACVM. The best we can do is to return an empty response, + // this allows us to ignore any foreign calls which exist solely to pass information from inside + // the circuit to the environment (e.g. custom logging) as the execution will still be able to progress. + // + // We optimistically return an empty response for all oracle calls as the ACVM will error + // should a response have been required. + Ok(ForeignCallResult::default()) + } + } + } +} diff --git a/noir/noir-repo/tooling/nargo/src/foreign_calls/print.rs b/noir/noir-repo/tooling/nargo/src/foreign_calls/print.rs new file mode 100644 index 00000000000..92fcd65ae28 --- /dev/null +++ b/noir/noir-repo/tooling/nargo/src/foreign_calls/print.rs @@ -0,0 +1,36 @@ +use acvm::{acir::brillig::ForeignCallResult, pwg::ForeignCallWaitInfo, AcirField}; +use noirc_printable_type::{ForeignCallError, PrintableValueDisplay}; + +use super::{ForeignCall, ForeignCallExecutor}; + +#[derive(Debug, Default)] +pub(crate) struct PrintForeignCallExecutor; + +impl ForeignCallExecutor for PrintForeignCallExecutor { + fn execute( + &mut self, + foreign_call: &ForeignCallWaitInfo, + ) -> Result, ForeignCallError> { + let foreign_call_name = foreign_call.function.as_str(); + match ForeignCall::lookup(foreign_call_name) { + Some(ForeignCall::Print) => { + let skip_newline = foreign_call.inputs[0].unwrap_field().is_zero(); + + let foreign_call_inputs = foreign_call + .inputs + .split_first() + .ok_or(ForeignCallError::MissingForeignCallInputs)? + .1; + + let display_values: PrintableValueDisplay = foreign_call_inputs.try_into()?; + let display_string = + format!("{display_values}{}", if skip_newline { "" } else { "\n" }); + + print!("{display_string}"); + + Ok(ForeignCallResult::default()) + } + _ => Err(ForeignCallError::NoHandler(foreign_call_name.to_string())), + } + } +} diff --git a/noir/noir-repo/tooling/nargo/src/foreign_calls/rpc.rs b/noir/noir-repo/tooling/nargo/src/foreign_calls/rpc.rs new file mode 100644 index 00000000000..0653eb1c7e3 --- /dev/null +++ b/noir/noir-repo/tooling/nargo/src/foreign_calls/rpc.rs @@ -0,0 +1,227 @@ +use std::path::PathBuf; + +use acvm::{acir::brillig::ForeignCallResult, pwg::ForeignCallWaitInfo, AcirField}; +use jsonrpc::{arg as build_json_rpc_arg, minreq_http::Builder, Client}; +use noirc_printable_type::ForeignCallError; +use serde::{Deserialize, Serialize}; + +use super::ForeignCallExecutor; + +#[derive(Debug)] +pub(crate) struct RPCForeignCallExecutor { + /// A randomly generated id for this `DefaultForeignCallExecutor`. + /// + /// This is used so that a single `external_resolver` can distinguish between requests from multiple + /// instantiations of `DefaultForeignCallExecutor`. + id: u64, + /// JSON RPC client to resolve foreign calls + external_resolver: Client, + /// Root path to the program or workspace in execution. + root_path: Option, + /// Name of the package in execution + package_name: Option, +} + +#[derive(Debug, Serialize, Deserialize)] +struct ResolveForeignCallRequest { + /// A session ID which allows the external RPC server to link this foreign call request to other foreign calls + /// for the same program execution. + /// + /// This is intended to allow a single RPC server to maintain state related to multiple program executions being + /// performed in parallel. + session_id: u64, + + #[serde(flatten)] + /// The foreign call which the external RPC server is to provide a response for. + function_call: ForeignCallWaitInfo, + + #[serde(skip_serializing_if = "Option::is_none")] + /// Root path to the program or workspace in execution. + root_path: Option, + #[serde(skip_serializing_if = "Option::is_none")] + /// Name of the package in execution + package_name: Option, +} + +impl RPCForeignCallExecutor { + pub(crate) fn new( + resolver_url: &str, + id: u64, + root_path: Option, + package_name: Option, + ) -> Self { + let mut transport_builder = + Builder::new().url(resolver_url).expect("Invalid oracle resolver URL"); + + if let Some(Ok(timeout)) = + std::env::var("NARGO_FOREIGN_CALL_TIMEOUT").ok().map(|timeout| timeout.parse()) + { + let timeout_duration = std::time::Duration::from_millis(timeout); + transport_builder = transport_builder.timeout(timeout_duration); + }; + let oracle_resolver = Client::with_transport(transport_builder.build()); + + RPCForeignCallExecutor { external_resolver: oracle_resolver, id, root_path, package_name } + } +} + +impl Deserialize<'a>> ForeignCallExecutor + for RPCForeignCallExecutor +{ + fn execute( + &mut self, + foreign_call: &ForeignCallWaitInfo, + ) -> Result, ForeignCallError> { + let encoded_params = vec![build_json_rpc_arg(ResolveForeignCallRequest { + session_id: self.id, + function_call: foreign_call.clone(), + root_path: self.root_path.clone().map(|path| path.to_str().unwrap().to_string()), + package_name: self.package_name.clone(), + })]; + + let req = self.external_resolver.build_request("resolve_foreign_call", &encoded_params); + + let response = self.external_resolver.send_request(req)?; + + let parsed_response: ForeignCallResult = response.result()?; + + Ok(parsed_response) + } +} + +#[cfg(test)] +mod tests { + use acvm::{ + acir::brillig::ForeignCallParam, brillig_vm::brillig::ForeignCallResult, + pwg::ForeignCallWaitInfo, FieldElement, + }; + use jsonrpc_core::Result as RpcResult; + use jsonrpc_derive::rpc; + use jsonrpc_http_server::{Server, ServerBuilder}; + + use super::{ForeignCallExecutor, RPCForeignCallExecutor, ResolveForeignCallRequest}; + + #[allow(unreachable_pub)] + #[rpc] + pub trait OracleResolver { + #[rpc(name = "resolve_foreign_call")] + fn resolve_foreign_call( + &self, + req: ResolveForeignCallRequest, + ) -> RpcResult>; + } + + struct OracleResolverImpl; + + impl OracleResolverImpl { + fn echo(&self, param: ForeignCallParam) -> ForeignCallResult { + vec![param].into() + } + + fn sum(&self, array: ForeignCallParam) -> ForeignCallResult { + let mut res: FieldElement = 0_usize.into(); + + for value in array.fields() { + res += value; + } + + res.into() + } + } + + impl OracleResolver for OracleResolverImpl { + fn resolve_foreign_call( + &self, + req: ResolveForeignCallRequest, + ) -> RpcResult> { + let response = match req.function_call.function.as_str() { + "sum" => self.sum(req.function_call.inputs[0].clone()), + "echo" => self.echo(req.function_call.inputs[0].clone()), + "id" => FieldElement::from(req.session_id as u128).into(), + + _ => panic!("unexpected foreign call"), + }; + Ok(response) + } + } + + fn build_oracle_server() -> (Server, String) { + let mut io = jsonrpc_core::IoHandler::new(); + io.extend_with(OracleResolverImpl.to_delegate()); + + // Choosing port 0 results in a random port being assigned. + let server = ServerBuilder::new(io) + .start_http(&"127.0.0.1:0".parse().expect("Invalid address")) + .expect("Could not start server"); + + let url = format!("http://{}", server.address()); + (server, url) + } + + #[test] + fn test_oracle_resolver_echo() { + let (server, url) = build_oracle_server(); + + let mut executor = RPCForeignCallExecutor::new(&url, 1, None, None); + + let foreign_call: ForeignCallWaitInfo = ForeignCallWaitInfo { + function: "echo".to_string(), + inputs: vec![ForeignCallParam::Single(1_u128.into())], + }; + + let result = executor.execute(&foreign_call); + assert_eq!(result.unwrap(), ForeignCallResult { values: foreign_call.inputs }); + + server.close(); + } + + #[test] + fn test_oracle_resolver_sum() { + let (server, url) = build_oracle_server(); + + let mut executor = RPCForeignCallExecutor::new(&url, 2, None, None); + + let foreign_call: ForeignCallWaitInfo = ForeignCallWaitInfo { + function: "sum".to_string(), + inputs: vec![ForeignCallParam::Array(vec![1_usize.into(), 2_usize.into()])], + }; + + let result = executor.execute(&foreign_call); + assert_eq!(result.unwrap(), FieldElement::from(3_usize).into()); + + server.close(); + } + + #[test] + fn foreign_call_executor_id_is_persistent() { + let (server, url) = build_oracle_server(); + + let mut executor = RPCForeignCallExecutor::new(&url, 3, None, None); + + let foreign_call: ForeignCallWaitInfo = + ForeignCallWaitInfo { function: "id".to_string(), inputs: Vec::new() }; + + let result_1 = executor.execute(&foreign_call).unwrap(); + let result_2 = executor.execute(&foreign_call).unwrap(); + assert_eq!(result_1, result_2); + + server.close(); + } + + #[test] + fn oracle_resolver_rpc_can_distinguish_executors() { + let (server, url) = build_oracle_server(); + + let mut executor_1 = RPCForeignCallExecutor::new(&url, 4, None, None); + let mut executor_2 = RPCForeignCallExecutor::new(&url, 5, None, None); + + let foreign_call: ForeignCallWaitInfo = + ForeignCallWaitInfo { function: "id".to_string(), inputs: Vec::new() }; + + let result_1 = executor_1.execute(&foreign_call).unwrap(); + let result_2 = executor_2.execute(&foreign_call).unwrap(); + assert_ne!(result_1, result_2); + + server.close(); + } +} diff --git a/noir/noir-repo/tooling/nargo/src/lib.rs b/noir/noir-repo/tooling/nargo/src/lib.rs index 88f07e0c292..74b7f54d860 100644 --- a/noir/noir-repo/tooling/nargo/src/lib.rs +++ b/noir/noir-repo/tooling/nargo/src/lib.rs @@ -9,6 +9,7 @@ pub mod constants; pub mod errors; +pub mod foreign_calls; pub mod ops; pub mod package; pub mod workspace; diff --git a/noir/noir-repo/tooling/nargo/src/ops/check.rs b/noir/noir-repo/tooling/nargo/src/ops/check.rs index 14d629ab0f6..707353ccdad 100644 --- a/noir/noir-repo/tooling/nargo/src/ops/check.rs +++ b/noir/noir-repo/tooling/nargo/src/ops/check.rs @@ -2,8 +2,8 @@ use acvm::compiler::CircuitSimulator; use noirc_driver::{CompiledProgram, ErrorsAndWarnings}; use noirc_errors::{CustomDiagnostic, FileDiagnostic}; +/// Run each function through a circuit simulator to check that they are solvable. pub fn check_program(compiled_program: &CompiledProgram) -> Result<(), ErrorsAndWarnings> { - // Check if the program is solvable for (i, circuit) in compiled_program.program.functions.iter().enumerate() { let mut simulator = CircuitSimulator::default(); if !simulator.check_circuit(circuit) { diff --git a/noir/noir-repo/tooling/nargo/src/ops/execute.rs b/noir/noir-repo/tooling/nargo/src/ops/execute.rs index 09ef554d2aa..57116ec2efd 100644 --- a/noir/noir-repo/tooling/nargo/src/ops/execute.rs +++ b/noir/noir-repo/tooling/nargo/src/ops/execute.rs @@ -10,10 +10,9 @@ use acvm::{acir::circuit::Circuit, acir::native_types::WitnessMap}; use acvm::{AcirField, BlackBoxFunctionSolver}; use crate::errors::ExecutionError; +use crate::foreign_calls::ForeignCallExecutor; use crate::NargoError; -use super::foreign_calls::ForeignCallExecutor; - struct ProgramExecutor<'a, F, B: BlackBoxFunctionSolver, E: ForeignCallExecutor> { functions: &'a [Circuit], diff --git a/noir/noir-repo/tooling/nargo/src/ops/foreign_calls.rs b/noir/noir-repo/tooling/nargo/src/ops/foreign_calls.rs deleted file mode 100644 index 30785949a46..00000000000 --- a/noir/noir-repo/tooling/nargo/src/ops/foreign_calls.rs +++ /dev/null @@ -1,494 +0,0 @@ -use std::path::PathBuf; - -use acvm::{ - acir::brillig::{ForeignCallParam, ForeignCallResult}, - pwg::ForeignCallWaitInfo, - AcirField, -}; -use jsonrpc::{arg as build_json_rpc_arg, minreq_http::Builder, Client}; -use noirc_printable_type::{decode_string_value, ForeignCallError, PrintableValueDisplay}; -use rand::Rng; -use serde::{Deserialize, Serialize}; - -pub trait ForeignCallExecutor { - fn execute( - &mut self, - foreign_call: &ForeignCallWaitInfo, - ) -> Result, ForeignCallError>; -} - -/// This enumeration represents the Brillig foreign calls that are natively supported by nargo. -/// After resolution of a foreign call, nargo will restart execution of the ACVM -pub enum ForeignCall { - Print, - CreateMock, - SetMockParams, - GetMockLastParams, - SetMockReturns, - SetMockTimes, - ClearMock, -} - -impl std::fmt::Display for ForeignCall { - fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result { - write!(f, "{}", self.name()) - } -} - -impl ForeignCall { - pub(crate) fn name(&self) -> &'static str { - match self { - ForeignCall::Print => "print", - ForeignCall::CreateMock => "create_mock", - ForeignCall::SetMockParams => "set_mock_params", - ForeignCall::GetMockLastParams => "get_mock_last_params", - ForeignCall::SetMockReturns => "set_mock_returns", - ForeignCall::SetMockTimes => "set_mock_times", - ForeignCall::ClearMock => "clear_mock", - } - } - - pub(crate) fn lookup(op_name: &str) -> Option { - match op_name { - "print" => Some(ForeignCall::Print), - "create_mock" => Some(ForeignCall::CreateMock), - "set_mock_params" => Some(ForeignCall::SetMockParams), - "get_mock_last_params" => Some(ForeignCall::GetMockLastParams), - "set_mock_returns" => Some(ForeignCall::SetMockReturns), - "set_mock_times" => Some(ForeignCall::SetMockTimes), - "clear_mock" => Some(ForeignCall::ClearMock), - _ => None, - } - } -} - -/// This struct represents an oracle mock. It can be used for testing programs that use oracles. -#[derive(Debug, PartialEq, Eq, Clone)] -struct MockedCall { - /// The id of the mock, used to update or remove it - id: usize, - /// The oracle it's mocking - name: String, - /// Optionally match the parameters - params: Option>>, - /// The parameters with which the mock was last called - last_called_params: Option>>, - /// The result to return when this mock is called - result: ForeignCallResult, - /// How many times should this mock be called before it is removed - times_left: Option, -} - -impl MockedCall { - fn new(id: usize, name: String) -> Self { - Self { - id, - name, - params: None, - last_called_params: None, - result: ForeignCallResult { values: vec![] }, - times_left: None, - } - } -} - -impl MockedCall { - fn matches(&self, name: &str, params: &[ForeignCallParam]) -> bool { - self.name == name && (self.params.is_none() || self.params.as_deref() == Some(params)) - } -} - -#[derive(Debug, Default)] -pub struct DefaultForeignCallExecutor { - /// A randomly generated id for this `DefaultForeignCallExecutor`. - /// - /// This is used so that a single `external_resolver` can distinguish between requests from multiple - /// instantiations of `DefaultForeignCallExecutor`. - id: u64, - - /// Mocks have unique ids used to identify them in Noir, allowing to update or remove them. - last_mock_id: usize, - /// The registered mocks - mocked_responses: Vec>, - /// Whether to print [`ForeignCall::Print`] output. - show_output: bool, - /// JSON RPC client to resolve foreign calls - external_resolver: Option, - /// Root path to the program or workspace in execution. - root_path: Option, - /// Name of the package in execution - package_name: Option, -} - -#[derive(Debug, Serialize, Deserialize)] -struct ResolveForeignCallRequest { - /// A session ID which allows the external RPC server to link this foreign call request to other foreign calls - /// for the same program execution. - /// - /// This is intended to allow a single RPC server to maintain state related to multiple program executions being - /// performed in parallel. - session_id: u64, - - #[serde(flatten)] - /// The foreign call which the external RPC server is to provide a response for. - function_call: ForeignCallWaitInfo, - - #[serde(skip_serializing_if = "Option::is_none")] - /// Root path to the program or workspace in execution. - root_path: Option, - #[serde(skip_serializing_if = "Option::is_none")] - /// Name of the package in execution - package_name: Option, -} - -impl DefaultForeignCallExecutor { - pub fn new( - show_output: bool, - resolver_url: Option<&str>, - root_path: Option, - package_name: Option, - ) -> Self { - let oracle_resolver = resolver_url.map(|resolver_url| { - let mut transport_builder = - Builder::new().url(resolver_url).expect("Invalid oracle resolver URL"); - - if let Some(Ok(timeout)) = - std::env::var("NARGO_FOREIGN_CALL_TIMEOUT").ok().map(|timeout| timeout.parse()) - { - let timeout_duration = std::time::Duration::from_millis(timeout); - transport_builder = transport_builder.timeout(timeout_duration); - }; - Client::with_transport(transport_builder.build()) - }); - DefaultForeignCallExecutor { - show_output, - external_resolver: oracle_resolver, - id: rand::thread_rng().gen(), - mocked_responses: Vec::new(), - last_mock_id: 0, - root_path, - package_name, - } - } -} - -impl DefaultForeignCallExecutor { - fn extract_mock_id( - foreign_call_inputs: &[ForeignCallParam], - ) -> Result<(usize, &[ForeignCallParam]), ForeignCallError> { - let (id, params) = - foreign_call_inputs.split_first().ok_or(ForeignCallError::MissingForeignCallInputs)?; - let id = - usize::try_from(id.unwrap_field().try_to_u64().expect("value does not fit into u64")) - .expect("value does not fit into usize"); - Ok((id, params)) - } - - fn find_mock_by_id(&self, id: usize) -> Option<&MockedCall> { - self.mocked_responses.iter().find(|response| response.id == id) - } - - fn find_mock_by_id_mut(&mut self, id: usize) -> Option<&mut MockedCall> { - self.mocked_responses.iter_mut().find(|response| response.id == id) - } - - fn parse_string(param: &ForeignCallParam) -> String { - let fields: Vec<_> = param.fields().to_vec(); - decode_string_value(&fields) - } - - fn execute_print(foreign_call_inputs: &[ForeignCallParam]) -> Result<(), ForeignCallError> { - let skip_newline = foreign_call_inputs[0].unwrap_field().is_zero(); - - let foreign_call_inputs = - foreign_call_inputs.split_first().ok_or(ForeignCallError::MissingForeignCallInputs)?.1; - let display_string = Self::format_printable_value(foreign_call_inputs, skip_newline)?; - - print!("{display_string}"); - - Ok(()) - } - - fn format_printable_value( - foreign_call_inputs: &[ForeignCallParam], - skip_newline: bool, - ) -> Result { - let display_values: PrintableValueDisplay = foreign_call_inputs.try_into()?; - - let result = format!("{display_values}{}", if skip_newline { "" } else { "\n" }); - - Ok(result) - } -} - -impl Deserialize<'a>> ForeignCallExecutor - for DefaultForeignCallExecutor -{ - fn execute( - &mut self, - foreign_call: &ForeignCallWaitInfo, - ) -> Result, ForeignCallError> { - let foreign_call_name = foreign_call.function.as_str(); - match ForeignCall::lookup(foreign_call_name) { - Some(ForeignCall::Print) => { - if self.show_output { - Self::execute_print(&foreign_call.inputs)?; - } - Ok(ForeignCallResult::default()) - } - Some(ForeignCall::CreateMock) => { - let mock_oracle_name = Self::parse_string(&foreign_call.inputs[0]); - assert!(ForeignCall::lookup(&mock_oracle_name).is_none()); - let id = self.last_mock_id; - self.mocked_responses.push(MockedCall::new(id, mock_oracle_name)); - self.last_mock_id += 1; - - Ok(F::from(id).into()) - } - Some(ForeignCall::SetMockParams) => { - let (id, params) = Self::extract_mock_id(&foreign_call.inputs)?; - self.find_mock_by_id_mut(id) - .unwrap_or_else(|| panic!("Unknown mock id {}", id)) - .params = Some(params.to_vec()); - - Ok(ForeignCallResult::default()) - } - Some(ForeignCall::GetMockLastParams) => { - let (id, _) = Self::extract_mock_id(&foreign_call.inputs)?; - let mock = - self.find_mock_by_id(id).unwrap_or_else(|| panic!("Unknown mock id {}", id)); - - let last_called_params = mock - .last_called_params - .clone() - .unwrap_or_else(|| panic!("Mock {} was never called", mock.name)); - - Ok(last_called_params.into()) - } - Some(ForeignCall::SetMockReturns) => { - let (id, params) = Self::extract_mock_id(&foreign_call.inputs)?; - self.find_mock_by_id_mut(id) - .unwrap_or_else(|| panic!("Unknown mock id {}", id)) - .result = ForeignCallResult { values: params.to_vec() }; - - Ok(ForeignCallResult::default()) - } - Some(ForeignCall::SetMockTimes) => { - let (id, params) = Self::extract_mock_id(&foreign_call.inputs)?; - let times = - params[0].unwrap_field().try_to_u64().expect("Invalid bit size of times"); - - self.find_mock_by_id_mut(id) - .unwrap_or_else(|| panic!("Unknown mock id {}", id)) - .times_left = Some(times); - - Ok(ForeignCallResult::default()) - } - Some(ForeignCall::ClearMock) => { - let (id, _) = Self::extract_mock_id(&foreign_call.inputs)?; - self.mocked_responses.retain(|response| response.id != id); - Ok(ForeignCallResult::default()) - } - None => { - let mock_response_position = self - .mocked_responses - .iter() - .position(|response| response.matches(foreign_call_name, &foreign_call.inputs)); - - if let Some(response_position) = mock_response_position { - // If the program has registered a mocked response to this oracle call then we prefer responding - // with that. - - let mock = self - .mocked_responses - .get_mut(response_position) - .expect("Invalid position of mocked response"); - - mock.last_called_params = Some(foreign_call.inputs.clone()); - - let result = mock.result.values.clone(); - - if let Some(times_left) = &mut mock.times_left { - *times_left -= 1; - if *times_left == 0 { - self.mocked_responses.remove(response_position); - } - } - - Ok(result.into()) - } else if let Some(external_resolver) = &self.external_resolver { - // If the user has registered an external resolver then we forward any remaining oracle calls there. - - let encoded_params = vec![build_json_rpc_arg(ResolveForeignCallRequest { - session_id: self.id, - function_call: foreign_call.clone(), - root_path: self - .root_path - .clone() - .map(|path| path.to_str().unwrap().to_string()), - package_name: self.package_name.clone(), - })]; - - let req = - external_resolver.build_request("resolve_foreign_call", &encoded_params); - - let response = external_resolver.send_request(req)?; - - let parsed_response: ForeignCallResult = response.result()?; - - Ok(parsed_response) - } else { - // If there's no registered mock oracle response and no registered resolver then we cannot - // return a correct response to the ACVM. The best we can do is to return an empty response, - // this allows us to ignore any foreign calls which exist solely to pass information from inside - // the circuit to the environment (e.g. custom logging) as the execution will still be able to progress. - // - // We optimistically return an empty response for all oracle calls as the ACVM will error - // should a response have been required. - Ok(ForeignCallResult::default()) - } - } - } - } -} - -#[cfg(test)] -mod tests { - use acvm::{ - acir::brillig::ForeignCallParam, brillig_vm::brillig::ForeignCallResult, - pwg::ForeignCallWaitInfo, FieldElement, - }; - use jsonrpc_core::Result as RpcResult; - use jsonrpc_derive::rpc; - use jsonrpc_http_server::{Server, ServerBuilder}; - - use crate::ops::{DefaultForeignCallExecutor, ForeignCallExecutor}; - - use super::ResolveForeignCallRequest; - - #[allow(unreachable_pub)] - #[rpc] - pub trait OracleResolver { - #[rpc(name = "resolve_foreign_call")] - fn resolve_foreign_call( - &self, - req: ResolveForeignCallRequest, - ) -> RpcResult>; - } - - struct OracleResolverImpl; - - impl OracleResolverImpl { - fn echo(&self, param: ForeignCallParam) -> ForeignCallResult { - vec![param].into() - } - - fn sum(&self, array: ForeignCallParam) -> ForeignCallResult { - let mut res: FieldElement = 0_usize.into(); - - for value in array.fields() { - res += value; - } - - res.into() - } - } - - impl OracleResolver for OracleResolverImpl { - fn resolve_foreign_call( - &self, - req: ResolveForeignCallRequest, - ) -> RpcResult> { - let response = match req.function_call.function.as_str() { - "sum" => self.sum(req.function_call.inputs[0].clone()), - "echo" => self.echo(req.function_call.inputs[0].clone()), - "id" => FieldElement::from(req.session_id as u128).into(), - - _ => panic!("unexpected foreign call"), - }; - Ok(response) - } - } - - fn build_oracle_server() -> (Server, String) { - let mut io = jsonrpc_core::IoHandler::new(); - io.extend_with(OracleResolverImpl.to_delegate()); - - // Choosing port 0 results in a random port being assigned. - let server = ServerBuilder::new(io) - .start_http(&"127.0.0.1:0".parse().expect("Invalid address")) - .expect("Could not start server"); - - let url = format!("http://{}", server.address()); - (server, url) - } - - #[test] - fn test_oracle_resolver_echo() { - let (server, url) = build_oracle_server(); - - let mut executor = - DefaultForeignCallExecutor::::new(false, Some(&url), None, None); - - let foreign_call = ForeignCallWaitInfo { - function: "echo".to_string(), - inputs: vec![ForeignCallParam::Single(1_u128.into())], - }; - - let result = executor.execute(&foreign_call); - assert_eq!(result.unwrap(), ForeignCallResult { values: foreign_call.inputs }); - - server.close(); - } - - #[test] - fn test_oracle_resolver_sum() { - let (server, url) = build_oracle_server(); - - let mut executor = DefaultForeignCallExecutor::new(false, Some(&url), None, None); - - let foreign_call = ForeignCallWaitInfo { - function: "sum".to_string(), - inputs: vec![ForeignCallParam::Array(vec![1_usize.into(), 2_usize.into()])], - }; - - let result = executor.execute(&foreign_call); - assert_eq!(result.unwrap(), FieldElement::from(3_usize).into()); - - server.close(); - } - - #[test] - fn foreign_call_executor_id_is_persistent() { - let (server, url) = build_oracle_server(); - - let mut executor = - DefaultForeignCallExecutor::::new(false, Some(&url), None, None); - - let foreign_call = ForeignCallWaitInfo { function: "id".to_string(), inputs: Vec::new() }; - - let result_1 = executor.execute(&foreign_call).unwrap(); - let result_2 = executor.execute(&foreign_call).unwrap(); - assert_eq!(result_1, result_2); - - server.close(); - } - - #[test] - fn oracle_resolver_rpc_can_distinguish_executors() { - let (server, url) = build_oracle_server(); - - let mut executor_1 = - DefaultForeignCallExecutor::::new(false, Some(&url), None, None); - let mut executor_2 = - DefaultForeignCallExecutor::::new(false, Some(&url), None, None); - - let foreign_call = ForeignCallWaitInfo { function: "id".to_string(), inputs: Vec::new() }; - - let result_1 = executor_1.execute(&foreign_call).unwrap(); - let result_2 = executor_2.execute(&foreign_call).unwrap(); - assert_ne!(result_1, result_2); - - server.close(); - } -} diff --git a/noir/noir-repo/tooling/nargo/src/ops/mod.rs b/noir/noir-repo/tooling/nargo/src/ops/mod.rs index f70577a14f1..04efeb5a9ec 100644 --- a/noir/noir-repo/tooling/nargo/src/ops/mod.rs +++ b/noir/noir-repo/tooling/nargo/src/ops/mod.rs @@ -4,7 +4,6 @@ pub use self::compile::{ compile_workspace, report_errors, }; pub use self::execute::{execute_program, execute_program_with_profiling}; -pub use self::foreign_calls::{DefaultForeignCallExecutor, ForeignCall, ForeignCallExecutor}; pub use self::optimize::{optimize_contract, optimize_program}; pub use self::transform::{transform_contract, transform_program}; @@ -13,7 +12,6 @@ pub use self::test::{run_test, TestStatus}; mod check; mod compile; mod execute; -mod foreign_calls; mod optimize; mod test; mod transform; diff --git a/noir/noir-repo/tooling/nargo/src/ops/test.rs b/noir/noir-repo/tooling/nargo/src/ops/test.rs index 370a4235f61..e258627b522 100644 --- a/noir/noir-repo/tooling/nargo/src/ops/test.rs +++ b/noir/noir-repo/tooling/nargo/src/ops/test.rs @@ -1,27 +1,42 @@ use std::path::PathBuf; use acvm::{ - acir::native_types::{WitnessMap, WitnessStack}, - BlackBoxFunctionSolver, FieldElement, + acir::{ + brillig::ForeignCallResult, + native_types::{WitnessMap, WitnessStack}, + }, + pwg::ForeignCallWaitInfo, + AcirField, BlackBoxFunctionSolver, FieldElement, }; use noirc_abi::Abi; use noirc_driver::{compile_no_check, CompileError, CompileOptions}; use noirc_errors::{debug_info::DebugInfo, FileDiagnostic}; use noirc_frontend::hir::{def_map::TestFunction, Context}; +use noirc_printable_type::ForeignCallError; +use rand::Rng; +use serde::{Deserialize, Serialize}; -use crate::{errors::try_to_diagnose_runtime_error, NargoError}; +use crate::{ + errors::try_to_diagnose_runtime_error, + foreign_calls::{ + mocker::MockForeignCallExecutor, print::PrintForeignCallExecutor, + rpc::RPCForeignCallExecutor, ForeignCall, ForeignCallExecutor, + }, + NargoError, +}; -use super::{execute_program, DefaultForeignCallExecutor}; +use super::execute_program; pub enum TestStatus { Pass, Fail { message: String, error_diagnostic: Option }, + Skipped, CompileError(FileDiagnostic), } impl TestStatus { pub fn failed(&self) -> bool { - !matches!(self, TestStatus::Pass) + !matches!(self, TestStatus::Pass | TestStatus::Skipped) } } @@ -48,23 +63,42 @@ pub fn run_test>( if test_function_has_no_arguments { // Run the backend to ensure the PWG evaluates functions like std::hash::pedersen, // otherwise constraints involving these expressions will not error. + let mut foreign_call_executor = TestForeignCallExecutor::new( + show_output, + foreign_call_resolver_url, + root_path, + package_name, + ); + let circuit_execution = execute_program( &compiled_program.program, WitnessMap::new(), blackbox_solver, - &mut DefaultForeignCallExecutor::new( - show_output, - foreign_call_resolver_url, - root_path, - package_name, - ), + &mut foreign_call_executor, ); - test_status_program_compile_pass( + + let status = test_status_program_compile_pass( test_function, compiled_program.abi, compiled_program.debug, circuit_execution, - ) + ); + + let ignore_foreign_call_failures = + std::env::var("NARGO_IGNORE_TEST_FAILURES_FROM_FOREIGN_CALLS") + .is_ok_and(|var| &var == "true"); + + if let TestStatus::Fail { .. } = status { + if ignore_foreign_call_failures + && foreign_call_executor.encountered_unknown_foreign_call + { + TestStatus::Skipped + } else { + status + } + } else { + status + } } else { #[cfg(target_arch = "wasm32")] { @@ -90,7 +124,7 @@ pub fn run_test>( program, initial_witness, blackbox_solver, - &mut DefaultForeignCallExecutor::::new( + &mut TestForeignCallExecutor::::new( false, foreign_call_resolver_url, root_path.clone(), @@ -215,3 +249,93 @@ fn check_expected_failure_message( error_diagnostic, } } + +/// A specialized foreign call executor which tracks whether it has encountered any unknown foreign calls +struct TestForeignCallExecutor { + /// The executor for any [`ForeignCall::Print`] calls. + printer: Option, + mocker: MockForeignCallExecutor, + external: Option, + + encountered_unknown_foreign_call: bool, +} + +impl TestForeignCallExecutor { + fn new( + show_output: bool, + resolver_url: Option<&str>, + root_path: Option, + package_name: Option, + ) -> Self { + let id = rand::thread_rng().gen(); + let printer = if show_output { Some(PrintForeignCallExecutor) } else { None }; + let external_resolver = resolver_url.map(|resolver_url| { + RPCForeignCallExecutor::new(resolver_url, id, root_path, package_name) + }); + TestForeignCallExecutor { + printer, + mocker: MockForeignCallExecutor::default(), + external: external_resolver, + encountered_unknown_foreign_call: false, + } + } +} + +impl Deserialize<'a>> ForeignCallExecutor + for TestForeignCallExecutor +{ + fn execute( + &mut self, + foreign_call: &ForeignCallWaitInfo, + ) -> Result, ForeignCallError> { + // If the circuit has reached a new foreign call opcode then it can't have failed from any previous unknown foreign calls. + self.encountered_unknown_foreign_call = false; + + let foreign_call_name = foreign_call.function.as_str(); + match ForeignCall::lookup(foreign_call_name) { + Some(ForeignCall::Print) => { + if let Some(printer) = &mut self.printer { + printer.execute(foreign_call) + } else { + Ok(ForeignCallResult::default()) + } + } + + Some( + ForeignCall::CreateMock + | ForeignCall::SetMockParams + | ForeignCall::GetMockLastParams + | ForeignCall::SetMockReturns + | ForeignCall::SetMockTimes + | ForeignCall::ClearMock, + ) => self.mocker.execute(foreign_call), + + None => { + // First check if there's any defined mock responses for this foreign call. + match self.mocker.execute(foreign_call) { + Err(ForeignCallError::NoHandler(_)) => (), + response_or_error => return response_or_error, + }; + + if let Some(external_resolver) = &mut self.external { + // If the user has registered an external resolver then we forward any remaining oracle calls there. + match external_resolver.execute(foreign_call) { + Err(ForeignCallError::NoHandler(_)) => (), + response_or_error => return response_or_error, + }; + } + + self.encountered_unknown_foreign_call = true; + + // If all executors have no handler for the given foreign call then we cannot + // return a correct response to the ACVM. The best we can do is to return an empty response, + // this allows us to ignore any foreign calls which exist solely to pass information from inside + // the circuit to the environment (e.g. custom logging) as the execution will still be able to progress. + // + // We optimistically return an empty response for all oracle calls as the ACVM will error + // should a response have been required. + Ok(ForeignCallResult::default()) + } + } + } +} diff --git a/noir/noir-repo/tooling/nargo/src/ops/transform.rs b/noir/noir-repo/tooling/nargo/src/ops/transform.rs index 9255ac3e0ec..fdda368d150 100644 --- a/noir/noir-repo/tooling/nargo/src/ops/transform.rs +++ b/noir/noir-repo/tooling/nargo/src/ops/transform.rs @@ -6,6 +6,7 @@ use iter_extended::vecmap; use noirc_driver::{CompiledContract, CompiledProgram}; use noirc_errors::debug_info::DebugInfo; +/// Apply ACVM optimizations on the circuit. pub fn transform_program( mut compiled_program: CompiledProgram, expression_width: ExpressionWidth, @@ -18,6 +19,7 @@ pub fn transform_program( compiled_program } +/// Apply the optimizing transformation on each function in the contract. pub fn transform_contract( contract: CompiledContract, expression_width: ExpressionWidth, @@ -25,7 +27,6 @@ pub fn transform_contract( let functions = vecmap(contract.functions, |mut func| { func.bytecode = transform_program_internal(func.bytecode, &mut func.debug, expression_width); - func }); diff --git a/noir/noir-repo/tooling/nargo_cli/Cargo.toml b/noir/noir-repo/tooling/nargo_cli/Cargo.toml index 02e669f5c68..5603b7f4fca 100644 --- a/noir/noir-repo/tooling/nargo_cli/Cargo.toml +++ b/noir/noir-repo/tooling/nargo_cli/Cargo.toml @@ -25,6 +25,7 @@ toml.workspace = true [dependencies] clap.workspace = true fm.workspace = true +fxhash.workspace = true iter-extended.workspace = true nargo.workspace = true nargo_fmt.workspace = true diff --git a/noir/noir-repo/tooling/nargo_cli/benches/criterion.rs b/noir/noir-repo/tooling/nargo_cli/benches/criterion.rs index 488cbfcd243..51de97df139 100644 --- a/noir/noir-repo/tooling/nargo_cli/benches/criterion.rs +++ b/noir/noir-repo/tooling/nargo_cli/benches/criterion.rs @@ -115,7 +115,7 @@ fn criterion_test_execution(c: &mut Criterion, test_program_dir: &Path, force_br let artifacts = RefCell::new(None); let mut foreign_call_executor = - nargo::ops::DefaultForeignCallExecutor::new(false, None, None, None); + nargo::foreign_calls::DefaultForeignCallExecutor::new(false, None, None, None); c.bench_function(&benchmark_name, |b| { b.iter_batched( diff --git a/noir/noir-repo/tooling/nargo_cli/build.rs b/noir/noir-repo/tooling/nargo_cli/build.rs index 740e5ed2052..41b3c0c9cf7 100644 --- a/noir/noir-repo/tooling/nargo_cli/build.rs +++ b/noir/noir-repo/tooling/nargo_cli/build.rs @@ -60,13 +60,9 @@ const IGNORED_BRILLIG_TESTS: [&str; 11] = [ ]; /// Tests which aren't expected to work with the default inliner cases. -const INLINER_MIN_OVERRIDES: [(&str, i64); 2] = [ +const INLINER_MIN_OVERRIDES: [(&str, i64); 1] = [ // 0 works if PoseidonHasher::write is tagged as `inline_always`, otherwise 22. ("eddsa", 0), - // (#6583): The RcTracker in the DIE SSA pass is removing inc_rcs that are still needed. - // This triggers differently depending on the optimization level (although all are wrong), - // so we arbitrarily only run with the inlined versions. - ("reference_counts", 0), ]; /// Some tests are expected to have warnings @@ -213,8 +209,13 @@ fn test_{test_name}(force_brillig: ForceBrillig, inliner_aggressiveness: Inliner nargo.arg("--program-dir").arg(test_program_dir); nargo.arg("{test_command}").arg("--force"); nargo.arg("--inliner-aggressiveness").arg(inliner_aggressiveness.0.to_string()); + if force_brillig.0 {{ nargo.arg("--force-brillig"); + + // Set the maximum increase so that part of the optimization is exercised (it might fail). + nargo.arg("--max-bytecode-increase-percent"); + nargo.arg("50"); }} {test_content} diff --git a/noir/noir-repo/tooling/nargo_cli/src/cli/compile_cmd.rs b/noir/noir-repo/tooling/nargo_cli/src/cli/compile_cmd.rs index 304988ed516..ff6009981c7 100644 --- a/noir/noir-repo/tooling/nargo_cli/src/cli/compile_cmd.rs +++ b/noir/noir-repo/tooling/nargo_cli/src/cli/compile_cmd.rs @@ -65,6 +65,7 @@ pub(crate) fn run(args: CompileCommand, config: NargoConfig) -> Result<(), CliEr Ok(()) } +/// Continuously recompile the workspace on any Noir file change event. fn watch_workspace(workspace: &Workspace, compile_options: &CompileOptions) -> notify::Result<()> { let (tx, rx) = std::sync::mpsc::channel(); @@ -108,6 +109,8 @@ fn watch_workspace(workspace: &Workspace, compile_options: &CompileOptions) -> n Ok(()) } +/// Parse and compile the entire workspace, then report errors. +/// This is the main entry point used by all other commands that need compilation. pub(super) fn compile_workspace_full( workspace: &Workspace, compile_options: &CompileOptions, @@ -129,6 +132,8 @@ pub(super) fn compile_workspace_full( Ok(()) } +/// Compile binary and contract packages. +/// Returns the merged warnings or errors. fn compile_workspace( file_manager: &FileManager, parsed_files: &ParsedFiles, @@ -144,6 +149,7 @@ fn compile_workspace( // Compile all of the packages in parallel. let program_warnings_or_errors: CompilationResult<()> = compile_programs(file_manager, parsed_files, workspace, &binary_packages, compile_options); + let contract_warnings_or_errors: CompilationResult<()> = compiled_contracts( file_manager, parsed_files, @@ -164,6 +170,7 @@ fn compile_workspace( } } +/// Compile the given binary packages in the workspace. fn compile_programs( file_manager: &FileManager, parsed_files: &ParsedFiles, @@ -171,6 +178,8 @@ fn compile_programs( binary_packages: &[Package], compile_options: &CompileOptions, ) -> CompilationResult<()> { + // Load any existing artifact for a given package, _iff_ it was compiled with the same nargo version. + // The loaded circuit includes backend specific transformations, which might be different from the current target. let load_cached_program = |package| { let program_artifact_path = workspace.package_build_path(package); read_program_from_file(program_artifact_path) @@ -180,19 +189,45 @@ fn compile_programs( }; let compile_package = |package| { + let cached_program = load_cached_program(package); + + // Hash over the entire compiled program, including any post-compile transformations. + // This is used to detect whether `cached_program` is returned by `compile_program`. + let cached_hash = cached_program.as_ref().map(fxhash::hash64); + + // Compile the program, or use the cached artifacts if it matches. let (program, warnings) = compile_program( file_manager, parsed_files, workspace, package, compile_options, - load_cached_program(package), + cached_program, )?; + // Choose the target width for the final, backend specific transformation. let target_width = get_target_width(package.expression_width, compile_options.expression_width); + + // If the compiled program is the same as the cached one, we don't apply transformations again, unless the target width has changed. + // The transformations might not be idempotent, which would risk creating witnesses that don't work with earlier versions, + // based on which we might have generated a verifier already. + if cached_hash == Some(fxhash::hash64(&program)) { + let width_matches = program + .program + .functions + .iter() + .all(|circuit| circuit.expression_width == target_width); + + if width_matches { + return Ok(((), warnings)); + } + } + // Run ACVM optimizations and set the target width. let program = nargo::ops::transform_program(program, target_width); + // Check solvability. nargo::ops::check_program(&program)?; + // Overwrite the build artifacts with the final circuit, which includes the backend specific transformations. save_program_to_file(&program.into(), &package.name, workspace.target_directory_path()); Ok(((), warnings)) @@ -208,6 +243,7 @@ fn compile_programs( collect_errors(program_results).map(|(_, warnings)| ((), warnings)) } +/// Compile the given contracts in the workspace. fn compiled_contracts( file_manager: &FileManager, parsed_files: &ParsedFiles, diff --git a/noir/noir-repo/tooling/nargo_cli/src/cli/execute_cmd.rs b/noir/noir-repo/tooling/nargo_cli/src/cli/execute_cmd.rs index 8dc71b1c7e5..fa95d3123c6 100644 --- a/noir/noir-repo/tooling/nargo_cli/src/cli/execute_cmd.rs +++ b/noir/noir-repo/tooling/nargo_cli/src/cli/execute_cmd.rs @@ -7,7 +7,7 @@ use clap::Args; use nargo::constants::PROVER_INPUT_FILE; use nargo::errors::try_to_diagnose_runtime_error; -use nargo::ops::DefaultForeignCallExecutor; +use nargo::foreign_calls::DefaultForeignCallExecutor; use nargo::package::{CrateName, Package}; use nargo_toml::{get_package_manifest, resolve_workspace_from_toml, PackageSelection}; use noirc_abi::input_parser::{Format, InputValue}; diff --git a/noir/noir-repo/tooling/nargo_cli/src/cli/info_cmd.rs b/noir/noir-repo/tooling/nargo_cli/src/cli/info_cmd.rs index cf416b1fa5f..769a1f79d81 100644 --- a/noir/noir-repo/tooling/nargo_cli/src/cli/info_cmd.rs +++ b/noir/noir-repo/tooling/nargo_cli/src/cli/info_cmd.rs @@ -4,7 +4,7 @@ use clap::Args; use iter_extended::vecmap; use nargo::{ constants::PROVER_INPUT_FILE, - ops::DefaultForeignCallExecutor, + foreign_calls::DefaultForeignCallExecutor, package::{CrateName, Package}, }; use nargo_toml::{get_package_manifest, resolve_workspace_from_toml, PackageSelection}; diff --git a/noir/noir-repo/tooling/nargo_cli/src/cli/test_cmd.rs b/noir/noir-repo/tooling/nargo_cli/src/cli/test_cmd.rs index 7b0201226ef..aa0ee1bb94b 100644 --- a/noir/noir-repo/tooling/nargo_cli/src/cli/test_cmd.rs +++ b/noir/noir-repo/tooling/nargo_cli/src/cli/test_cmd.rs @@ -255,6 +255,12 @@ fn display_test_report( ); } } + TestStatus::Skipped { .. } => { + writer + .set_color(ColorSpec::new().set_fg(Some(Color::Yellow))) + .expect("Failed to set color"); + writeln!(writer, "skipped").expect("Failed to write to stderr"); + } TestStatus::CompileError(err) => { noirc_errors::reporter::report_all( file_manager.as_file_map(), diff --git a/noir/noir-repo/tooling/nargo_cli/tests/stdlib-props.rs b/noir/noir-repo/tooling/nargo_cli/tests/stdlib-props.rs index 0013a90b4ff..86c225831b9 100644 --- a/noir/noir-repo/tooling/nargo_cli/tests/stdlib-props.rs +++ b/noir/noir-repo/tooling/nargo_cli/tests/stdlib-props.rs @@ -2,10 +2,7 @@ use std::{cell::RefCell, collections::BTreeMap, path::Path}; use acvm::{acir::native_types::WitnessStack, AcirField, FieldElement}; use iter_extended::vecmap; -use nargo::{ - ops::{execute_program, DefaultForeignCallExecutor}, - parse_all, -}; +use nargo::{foreign_calls::DefaultForeignCallExecutor, ops::execute_program, parse_all}; use noirc_abi::input_parser::InputValue; use noirc_driver::{ compile_main, file_manager_with_stdlib, prepare_crate, CompilationResult, CompileOptions, @@ -64,6 +61,7 @@ fn prepare_and_compile_snippet( ) -> CompilationResult { let (mut context, root_crate_id) = prepare_snippet(source); let options = CompileOptions { force_brillig, ..Default::default() }; + // TODO: Run nargo::ops::transform_program? compile_main(&mut context, root_crate_id, &options, None) } diff --git a/noir/noir-repo/tooling/nargo_cli/tests/stdlib-tests.rs b/noir/noir-repo/tooling/nargo_cli/tests/stdlib-tests.rs index bdc92e625ab..99f0c9a2e7f 100644 --- a/noir/noir-repo/tooling/nargo_cli/tests/stdlib-tests.rs +++ b/noir/noir-repo/tooling/nargo_cli/tests/stdlib-tests.rs @@ -138,6 +138,12 @@ fn display_test_report( ); } } + TestStatus::Skipped { .. } => { + writer + .set_color(ColorSpec::new().set_fg(Some(Color::Yellow))) + .expect("Failed to set color"); + writeln!(writer, "skipped").expect("Failed to write to stderr"); + } TestStatus::CompileError(err) => { noirc_errors::reporter::report_all( file_manager.as_file_map(), diff --git a/noir/noir-repo/tooling/noirc_abi/proptest-regressions/input_parser/json.txt b/noir/noir-repo/tooling/noirc_abi/proptest-regressions/input_parser/json.txt new file mode 100644 index 00000000000..19de8eeaf48 --- /dev/null +++ b/noir/noir-repo/tooling/noirc_abi/proptest-regressions/input_parser/json.txt @@ -0,0 +1,7 @@ +# Seeds for failure cases proptest has generated in the past. It is +# automatically read and these particular cases re-run before any +# novel cases are generated. +# +# It is recommended to check this file in to source control so that +# everyone who runs the test benefits from these saved cases. +cc b3f9ae88d54944ca274764f4d99a2023d4b0ac09beb89bc599cbba1e45dd3620 # shrinks to (typ, value) = (Integer { sign: Signed, width: 1 }, -1) diff --git a/noir/noir-repo/tooling/noirc_abi/proptest-regressions/input_parser/toml.txt b/noir/noir-repo/tooling/noirc_abi/proptest-regressions/input_parser/toml.txt new file mode 100644 index 00000000000..1448cb67ef1 --- /dev/null +++ b/noir/noir-repo/tooling/noirc_abi/proptest-regressions/input_parser/toml.txt @@ -0,0 +1,9 @@ +# Seeds for failure cases proptest has generated in the past. It is +# automatically read and these particular cases re-run before any +# novel cases are generated. +# +# It is recommended to check this file in to source control so that +# everyone who runs the test benefits from these saved cases. +cc 9d200afb8f5c01e3414d24eebe1436a7eef5377a46a9a9235aaa7f81e0b33656 # shrinks to (typ, value) = (Integer { sign: Signed, width: 8 }, -1) +cc 7fd29637e5566d819992185c1a95438e9949a555928a911b3918eed2e3f7a1fd # shrinks to (typ, value) = (Integer { sign: Signed, width: 64 }, -1) +cc 8ecbda39d887674b53ca23a861ac30fbb10c123bb70c57e69b336c86a3d9dea8 # shrinks to (abi, input_map) = (Abi { parameters: [AbiParameter { name: "¡", typ: Struct { path: "�)\u{1b}=�?Ⱥ\u{59424}?{\u{e4d5e}%Ѩ/Q\u{36a17}/*\";\u{b}&iC_\u{d313f}S\u{1b}\u{9dfec}\r/\u{10530d}", fields: [("?p*\"/\u{202e}\u{6f038}\u{537ca}.y@~𘛶?4\u{1b}*", Field), (".Ⱥ/$\u{7f}\u{103c06}%\\\u{202e}][0\u{88479}]\"*~\u{36fd5}\u{5}\u{feff}]{/", Tuple { fields: [String { length: 937 }] }), ("r\u{ac3a5}&:", Boolean), ("$d6🕴/:|�\u{37f8b}\r\u{a13b7}C$𲁹\\&\u{f8712}?\u{db61c}t%\u{57be1}\0", Field), ("/\u{6378b}\u{a426c}¥\u{7}/\u{fcb29}$\u{53c6b}\u{12d6f}\u{12bd3}.\u{f2f82}\u{8613e}*$\u{fd32f}\u{e29f7}\0𨺉'¬\"1", Struct { path: "\\\u{4a5ac}<\u{9e505}\u{4f3af}🕴&?<:^\u{7}\u{88}\u{3e1ff}(¥\u{531f3}K{:¥𦺀", fields: [("n\0Ѩ/\u{1b}𥐰\u{a4906}�¥`{\u{389d4}`1\u{7708a})\u{3dac4}8\u{93e5f}㒭\\\"\u{e6824}\u{b}Ѩ\u{88946}Ⱥ{", Integer { sign: Signed, width: 127 })] }), ("¥🕴\u{1b}¥🕴=sR\0\u{35f36}\u{867dc}>ä\u{202e}f:BȺ?:``*·¥\u{74ca5}\"", Tuple { fields: [Boolean, Field, String { length: 205 }, String { length: 575 }, Integer { sign: Signed, width: 124 }, String { length: 923 }, String { length: 294 }] })] }, visibility: Public }], return_type: None, error_types: {} }, {"¡": Struct({"$d6🕴/:|�\u{37f8b}\r\u{a13b7}C$𲁹\\&\u{f8712}?\u{db61c}t%\u{57be1}\0": Field(-8275115097504119425402713293372777967031130481426075481525511323101167533940), ".Ⱥ/$\u{7f}\u{103c06}%\\\u{202e}][0\u{88479}]\"*~\u{36fd5}\u{5}\u{feff}]{/": Vec([String("A \0A 0 aA0 a0aa00 A\000 0 \0\0aA\0\0a \0 \0a 0A\0A\0 Aa0aAA0A\0aa\00 0\0\0\0\0\00a Aa0 \0 a A0 \0AA0A Aa Aa\00aAaAaaA0A0 aA0 \0 Aa\00 \0000AAA a \0AAaaA\0\0a A0a0AA\0aA00 aA a0A\0AAa0a\0A0a\0\0A0A \00Aaaaa a A AO.*D\r.`bD4a\n*\u{15}\\B\"ace.8&A\t[AV8w<\u{18}\"\u{f}4`^Q\u{1b}U*$Z/\0\u{b}]qw${`\"=X&A\\\u{e}%`\\:\"$\u{1}.(6_C:\u{7}a`V=N**\u{1b})#Y\u{7f}#\u{b}$l\t}.Mns5!\t*$g\u{18}\rC\u{11}\"$=\u{7}.?&\u{1}yW\t.Y|<6\u{12}\u{e}/4JJ*&/V$`\"&`x#R\np\\%'*\n:P\0K\u{b}*`\r7Ym\t_\u{b}=$\u{16}`0v\u{7f}'NV^N4J<9=G*A:!b\u{1c}:'c{ST&z![\u{7f}/.={E*pmaWC\u{7f}7p{<\"']\u{8}?`\u{1b}\"\\\u{1}$\u{18}/!\u{16}-\t:E7CUs%_qw*xf.S\t\u{4}'=\"&%t'\u{1f}\u{7f}\u{b}$.=f\u{6}\"$A}xV_$\u{1a}nH\n\u{1b}?<&\n\u{15}U\\-b\u{1d}|\u{b}\u{2}t \rwA{L\u{11}\u{6}\u{10}\0\u{1b}G[x?&Yi?&7\u{b}?\r\u{1f}b\\$=\u{b}x& Q/\t\u{4}|X\"7\"{\0\0j'.\0\\e1zR.\u{c}\n<\u{b}Q*R+y8\u{19}(o\u{1f}@m\nt+\u{7f}Q\\+.Rn?\u{17}UZ\"$\u{b}/\0B=9=\t{\u{8}qZ&`!:D{\u{6}IO.H\u{7f}:?/3@\r\u{1b}oä\u{202e}f:BȺ?:``*·¥\u{74ca5}\"": Vec([Field(1), Field(8822392870083219098626030699076694602179106416928939583840848325203494062169), String("*TXn;{}\"_)_9\nk\\#ts\u{10}%\\c\n/2._::Oj*\u{7f}\0\r&PUMl\u{10}$/u?L}\u{7f}*P&<%=\u{7}S#%A\n \u{e}\\#v!\"\nepRp.{vH{&@\t\u{1f}\u{b}?=T\u{f}\"B\u{11}\n/{HY.\u{16}\n\nj<&\u{3}{f\n/9J*&x.$/,\r\0\u{1c}'\u{5}\u{13}\u{1b}`T\0`\n&/&\u{15}\u{b}w:{SK\u{7f}\\apR%/'0`0\n'd$$\u{7f}Vs\t<{\nDTT\\F\n\u{15}y.\\\t*-)&D$*u\u{b}\u{1b}?{\u{b}/\n\u{7f}0*.7\0\n:\u{b}.rSk<6~>{#"), String(".\"JA%q6i\ra/:F\u{16}?q<\t\rN\\13?H<;?{`\u{1d}p{.\"5?*@'N\"\u{1a}P,\u{1b}\u{7f}c+dt5':Y\u{1b}k/G>k/eM$XIX')\u{1b}'&\u{7f}\\\r\u{1b}`'P_.\n.?\0p`Y\u{c}`._\u{b}B\0\ng/*v$jfJ:\u{c}\u{1b}Pv}xn7ph@#{_<{.JD?r%'E\n7s9n/],u![;%*\u{2}{y`MgRdok8\"%<*>*{GyFJ}?\0W%#\0\u{1b}\u{7f}\u{16}G:\t=w\u{7f}:q\u{7f}:{k?\u{b}(:ca{$*1X/cw\u{1b}Z6I\rX\0\u{1b}(.^14\r\\=s\u{1b}w\u{3}F~\n\u{1e})/$0:=[\u{1},\\\\\tg\u{16}:],J`\0N\n\u{1b}\u{1b}\u{1b}{.xb\u{1a}\r'12#?e\\#/\tA\u{7f}\".\\Ke=\\?!v+P\u{17}\r\u{12}x.=A.`0<&?\niR/*WW\rnV)5vY.~\n _h\0&5f#\r\u{2}-S%\t s..\u{7f}!X}\"=\"?\u{5}y\u{4}`fr&R&d: 1Ht\"4`y_/S.71#{|%$%&ehy\u{16}J_\u{e}=:.%'\"N=J:\r:{&.\u{12}\u{b})&N\u{10}R_3;11\u{b}Qd<`<{?xF:~\"%<=<<\03:t??&\r;{\u{13}?__Y\u{6})\\k,vs?\n`G(*\n!\u{1b}[@z\0$?*yKLJh_\u{13}FkY'\\?T^\u{1f}$1n`'[\n\u{7f}\0+l\u{b}\u{1a}E\u{b}&(/\u{b}\rr\t:&\0+N'N:oC:*``IN\u{b}*.:\t$7+'*U:\t Result { let json_value = match (value, abi_type) { + (InputValue::Field(f), AbiType::Integer { sign: crate::Sign::Signed, width }) => { + JsonTypes::String(field_to_signed_hex(*f, *width)) + } (InputValue::Field(f), AbiType::Field | AbiType::Integer { .. }) => { JsonTypes::String(Self::format_field_string(*f)) } @@ -143,6 +146,9 @@ impl InputValue { ) -> Result { let input_value = match (value, param_type) { (JsonTypes::String(string), AbiType::String { .. }) => InputValue::String(string), + (JsonTypes::String(string), AbiType::Integer { sign: crate::Sign::Signed, width }) => { + InputValue::Field(parse_str_to_signed(&string, *width)?) + } ( JsonTypes::String(string), AbiType::Field | AbiType::Integer { .. } | AbiType::Boolean, @@ -192,3 +198,40 @@ impl InputValue { Ok(input_value) } } + +#[cfg(test)] +mod test { + use proptest::prelude::*; + + use crate::{ + arbitrary::arb_abi_and_input_map, + input_parser::{arbitrary::arb_signed_integer_type_and_value, json::JsonTypes, InputValue}, + }; + + use super::{parse_json, serialize_to_json}; + + proptest! { + #[test] + fn serializing_and_parsing_returns_original_input((abi, input_map) in arb_abi_and_input_map()) { + let json = serialize_to_json(&input_map, &abi).expect("should be serializable"); + let parsed_input_map = parse_json(&json, &abi).expect("should be parsable"); + + prop_assert_eq!(parsed_input_map, input_map); + } + + #[test] + fn signed_integer_serialization_roundtrip((typ, value) in arb_signed_integer_type_and_value()) { + let string_input = JsonTypes::String(value.to_string()); + let input_value = InputValue::try_from_json(string_input, &typ, "foo").expect("should be parsable"); + let JsonTypes::String(output_string) = JsonTypes::try_from_input_value(&input_value, &typ).expect("should be serializable") else { + panic!("wrong type output"); + }; + let output_number = if let Some(output_string) = output_string.strip_prefix("-0x") { + -i64::from_str_radix(output_string, 16).unwrap() + } else { + i64::from_str_radix(output_string.strip_prefix("0x").unwrap(), 16).unwrap() + }; + prop_assert_eq!(output_number, value); + } + } +} diff --git a/noir/noir-repo/tooling/noirc_abi/src/input_parser/mod.rs b/noir/noir-repo/tooling/noirc_abi/src/input_parser/mod.rs index d7bbb0adfe3..b7732235eb2 100644 --- a/noir/noir-repo/tooling/noirc_abi/src/input_parser/mod.rs +++ b/noir/noir-repo/tooling/noirc_abi/src/input_parser/mod.rs @@ -248,6 +248,11 @@ mod serialization_tests { typ: AbiType::Field, visibility: AbiVisibility::Private, }, + AbiParameter { + name: "signed_example".into(), + typ: AbiType::Integer { sign: Sign::Signed, width: 8 }, + visibility: AbiVisibility::Private, + }, AbiParameter { name: "bar".into(), typ: AbiType::Struct { @@ -272,6 +277,7 @@ mod serialization_tests { let input_map: BTreeMap = BTreeMap::from([ ("foo".into(), InputValue::Field(FieldElement::one())), + ("signed_example".into(), InputValue::Field(FieldElement::from(240u128))), ( "bar".into(), InputValue::Struct(BTreeMap::from([ @@ -317,7 +323,9 @@ fn parse_str_to_field(value: &str) -> Result { } fn parse_str_to_signed(value: &str, width: u32) -> Result { - let big_num = if let Some(hex) = value.strip_prefix("0x") { + let big_num = if let Some(hex) = value.strip_prefix("-0x") { + BigInt::from_str_radix(hex, 16).map(|value| -value) + } else if let Some(hex) = value.strip_prefix("0x") { BigInt::from_str_radix(hex, 16) } else { BigInt::from_str_radix(value, 10) @@ -357,12 +365,23 @@ fn field_from_big_int(bigint: BigInt) -> FieldElement { } } +fn field_to_signed_hex(f: FieldElement, bit_size: u32) -> String { + let f_u128 = f.to_u128(); + let max = 2_u128.pow(bit_size - 1) - 1; + if f_u128 > max { + let f = FieldElement::from(2_u128.pow(bit_size) - f_u128); + format!("-0x{}", f.to_hex()) + } else { + format!("0x{}", f.to_hex()) + } +} + #[cfg(test)] mod test { use acvm::{AcirField, FieldElement}; use num_bigint::BigUint; - use super::parse_str_to_field; + use super::{parse_str_to_field, parse_str_to_signed}; fn big_uint_from_field(field: FieldElement) -> BigUint { BigUint::from_bytes_be(&field.to_be_bytes()) @@ -400,4 +419,38 @@ mod test { let noncanonical_field = FieldElement::modulus().to_string(); assert!(parse_str_to_field(&noncanonical_field).is_err()); } + + #[test] + fn test_parse_str_to_signed() { + let value = parse_str_to_signed("1", 8).unwrap(); + assert_eq!(value, FieldElement::from(1_u128)); + + let value = parse_str_to_signed("-1", 8).unwrap(); + assert_eq!(value, FieldElement::from(255_u128)); + + let value = parse_str_to_signed("-1", 16).unwrap(); + assert_eq!(value, FieldElement::from(65535_u128)); + } +} + +#[cfg(test)] +mod arbitrary { + use proptest::prelude::*; + + use crate::{AbiType, Sign}; + + pub(super) fn arb_signed_integer_type_and_value() -> BoxedStrategy<(AbiType, i64)> { + (2u32..=64) + .prop_flat_map(|width| { + let typ = Just(AbiType::Integer { width, sign: Sign::Signed }); + let value = if width == 64 { + // Avoid overflow + i64::MIN..i64::MAX + } else { + -(2i64.pow(width - 1))..(2i64.pow(width - 1) - 1) + }; + (typ, value) + }) + .boxed() + } } diff --git a/noir/noir-repo/tooling/noirc_abi/src/input_parser/toml.rs b/noir/noir-repo/tooling/noirc_abi/src/input_parser/toml.rs index 321d3511b5d..6f2be68a0c4 100644 --- a/noir/noir-repo/tooling/noirc_abi/src/input_parser/toml.rs +++ b/noir/noir-repo/tooling/noirc_abi/src/input_parser/toml.rs @@ -1,4 +1,4 @@ -use super::{parse_str_to_field, parse_str_to_signed, InputValue}; +use super::{field_to_signed_hex, parse_str_to_field, parse_str_to_signed, InputValue}; use crate::{errors::InputParserError, Abi, AbiType, MAIN_RETURN_NAME}; use acvm::{AcirField, FieldElement}; use iter_extended::{try_btree_map, try_vecmap}; @@ -60,7 +60,7 @@ pub(crate) fn serialize_to_toml( Ok(toml_string) } -#[derive(Debug, Deserialize, Serialize, Clone)] +#[derive(Debug, Deserialize, Serialize, Clone, PartialEq)] #[serde(untagged)] enum TomlTypes { // This is most likely going to be a hex string @@ -83,6 +83,9 @@ impl TomlTypes { abi_type: &AbiType, ) -> Result { let toml_value = match (value, abi_type) { + (InputValue::Field(f), AbiType::Integer { sign: crate::Sign::Signed, width }) => { + TomlTypes::String(field_to_signed_hex(*f, *width)) + } (InputValue::Field(f), AbiType::Field | AbiType::Integer { .. }) => { let f_str = format!("0x{}", f.to_hex()); TomlTypes::String(f_str) @@ -126,6 +129,7 @@ impl InputValue { ) -> Result { let input_value = match (value, param_type) { (TomlTypes::String(string), AbiType::String { .. }) => InputValue::String(string), + ( TomlTypes::String(string), AbiType::Field @@ -139,7 +143,7 @@ impl InputValue { TomlTypes::Integer(integer), AbiType::Field | AbiType::Integer { .. } | AbiType::Boolean, ) => { - let new_value = FieldElement::from(i128::from(integer)); + let new_value = FieldElement::from(u128::from(integer)); InputValue::Field(new_value) } @@ -179,3 +183,40 @@ impl InputValue { Ok(input_value) } } + +#[cfg(test)] +mod test { + use proptest::prelude::*; + + use crate::{ + arbitrary::arb_abi_and_input_map, + input_parser::{arbitrary::arb_signed_integer_type_and_value, toml::TomlTypes, InputValue}, + }; + + use super::{parse_toml, serialize_to_toml}; + + proptest! { + #[test] + fn serializing_and_parsing_returns_original_input((abi, input_map) in arb_abi_and_input_map()) { + let toml = serialize_to_toml(&input_map, &abi).expect("should be serializable"); + let parsed_input_map = parse_toml(&toml, &abi).expect("should be parsable"); + + prop_assert_eq!(parsed_input_map, input_map); + } + + #[test] + fn signed_integer_serialization_roundtrip((typ, value) in arb_signed_integer_type_and_value()) { + let string_input = TomlTypes::String(value.to_string()); + let input_value = InputValue::try_from_toml(string_input.clone(), &typ, "foo").expect("should be parsable"); + let TomlTypes::String(output_string) = TomlTypes::try_from_input_value(&input_value, &typ).expect("should be serializable") else { + panic!("wrong type output"); + }; + let output_number = if let Some(output_string) = output_string.strip_prefix("-0x") { + -i64::from_str_radix(output_string, 16).unwrap() + } else { + i64::from_str_radix(output_string.strip_prefix("0x").unwrap(), 16).unwrap() + }; + prop_assert_eq!(output_number, value); + } + } +} diff --git a/noir/noir-repo/tooling/noirc_abi/src/lib.rs b/noir/noir-repo/tooling/noirc_abi/src/lib.rs index b1b199727c2..bd5674d64f1 100644 --- a/noir/noir-repo/tooling/noirc_abi/src/lib.rs +++ b/noir/noir-repo/tooling/noirc_abi/src/lib.rs @@ -49,6 +49,7 @@ pub const MAIN_RETURN_NAME: &str = "return"; /// depends on the types of programs that users want to do. I don't envision string manipulation /// in programs, however it is possible to support, with many complications like encoding character set /// support. +#[derive(Hash)] pub enum AbiType { Field, Array { @@ -77,7 +78,7 @@ pub enum AbiType { }, } -#[derive(Clone, Copy, Debug, PartialEq, Eq, Serialize, Deserialize)] +#[derive(Clone, Copy, Debug, PartialEq, Eq, Serialize, Deserialize, Hash)] #[cfg_attr(test, derive(arbitrary::Arbitrary))] #[serde(rename_all = "lowercase")] /// Represents whether the parameter is public or known only to the prover. @@ -89,7 +90,7 @@ pub enum AbiVisibility { DataBus, } -#[derive(Clone, Copy, Debug, PartialEq, Eq, Serialize, Deserialize)] +#[derive(Clone, Copy, Debug, PartialEq, Eq, Serialize, Deserialize, Hash)] #[cfg_attr(test, derive(arbitrary::Arbitrary))] #[serde(rename_all = "lowercase")] pub enum Sign { @@ -146,7 +147,7 @@ impl From<&AbiType> for PrintableType { } } -#[derive(Clone, Debug, PartialEq, Serialize, Deserialize)] +#[derive(Clone, Debug, PartialEq, Serialize, Deserialize, Hash)] #[cfg_attr(test, derive(arbitrary::Arbitrary))] /// An argument or return value of the circuit's `main` function. pub struct AbiParameter { @@ -163,7 +164,7 @@ impl AbiParameter { } } -#[derive(Clone, Debug, Serialize, Deserialize)] +#[derive(Clone, Debug, Serialize, Deserialize, Hash)] #[cfg_attr(test, derive(arbitrary::Arbitrary))] pub struct AbiReturnType { #[cfg_attr(test, proptest(strategy = "arbitrary::arb_abi_type()"))] @@ -171,7 +172,7 @@ pub struct AbiReturnType { pub visibility: AbiVisibility, } -#[derive(Clone, Debug, Default, Serialize, Deserialize)] +#[derive(Clone, Debug, Default, Serialize, Deserialize, Hash)] #[cfg_attr(test, derive(arbitrary::Arbitrary))] pub struct Abi { /// An ordered list of the arguments to the program's `main` function, specifying their types and visibility. @@ -459,7 +460,7 @@ pub enum AbiValue { }, } -#[derive(Clone, Debug, PartialEq, Eq, Serialize, Deserialize)] +#[derive(Clone, Debug, PartialEq, Eq, Serialize, Deserialize, Hash)] #[serde(tag = "error_kind", rename_all = "lowercase")] pub enum AbiErrorType { FmtString { length: u32, item_types: Vec }, diff --git a/noir/noir-repo/tooling/noirc_abi_wasm/Cargo.toml b/noir/noir-repo/tooling/noirc_abi_wasm/Cargo.toml index daa619ca01d..b00d580515e 100644 --- a/noir/noir-repo/tooling/noirc_abi_wasm/Cargo.toml +++ b/noir/noir-repo/tooling/noirc_abi_wasm/Cargo.toml @@ -1,9 +1,11 @@ [package] name = "noirc_abi_wasm" +description = "An ABI encoder for the Noir language" version.workspace = true authors.workspace = true edition.workspace = true license.workspace = true +repository.workspace = true # See more keys and their definitions at https://doc.rust-lang.org/cargo/reference/manifest.html diff --git a/noir/noir-repo/tooling/noirc_abi_wasm/test/browser/abi_encode.test.ts b/noir/noir-repo/tooling/noirc_abi_wasm/test/browser/abi_encode.test.ts index e1aaf0dc2c0..ac18495919c 100644 --- a/noir/noir-repo/tooling/noirc_abi_wasm/test/browser/abi_encode.test.ts +++ b/noir/noir-repo/tooling/noirc_abi_wasm/test/browser/abi_encode.test.ts @@ -15,7 +15,8 @@ it('recovers original inputs when abi encoding and decoding', async () => { const foo: Field = inputs.foo as Field; const bar: Field[] = inputs.bar as Field[]; expect(BigInt(decoded_inputs.inputs.foo)).to.be.equal(BigInt(foo)); - expect(BigInt(decoded_inputs.inputs.bar[0])).to.be.equal(BigInt(bar[0])); - expect(BigInt(decoded_inputs.inputs.bar[1])).to.be.equal(BigInt(bar[1])); + expect(parseInt(decoded_inputs.inputs.bar[0])).to.be.equal(parseInt(bar[0].toString())); + expect(parseInt(decoded_inputs.inputs.bar[1])).to.be.equal(parseInt(bar[1].toString())); + expect(parseInt(decoded_inputs.inputs.bar[2])).to.be.equal(parseInt(bar[2].toString())); expect(decoded_inputs.return_value).to.be.null; }); diff --git a/noir/noir-repo/tooling/noirc_abi_wasm/test/node/abi_encode.test.ts b/noir/noir-repo/tooling/noirc_abi_wasm/test/node/abi_encode.test.ts index a49c10b6ea6..e87618d84da 100644 --- a/noir/noir-repo/tooling/noirc_abi_wasm/test/node/abi_encode.test.ts +++ b/noir/noir-repo/tooling/noirc_abi_wasm/test/node/abi_encode.test.ts @@ -11,7 +11,8 @@ it('recovers original inputs when abi encoding and decoding', async () => { const foo: Field = inputs.foo as Field; const bar: Field[] = inputs.bar as Field[]; expect(BigInt(decoded_inputs.inputs.foo)).to.be.equal(BigInt(foo)); - expect(BigInt(decoded_inputs.inputs.bar[0])).to.be.equal(BigInt(bar[0])); - expect(BigInt(decoded_inputs.inputs.bar[1])).to.be.equal(BigInt(bar[1])); + expect(parseInt(decoded_inputs.inputs.bar[0])).to.be.equal(parseInt(bar[0].toString())); + expect(parseInt(decoded_inputs.inputs.bar[1])).to.be.equal(parseInt(bar[1].toString())); + expect(parseInt(decoded_inputs.inputs.bar[2])).to.be.equal(parseInt(bar[2].toString())); expect(decoded_inputs.return_value).to.be.null; }); diff --git a/noir/noir-repo/tooling/noirc_abi_wasm/test/shared/abi_encode.ts b/noir/noir-repo/tooling/noirc_abi_wasm/test/shared/abi_encode.ts index 62eb7658f43..b789bb05371 100644 --- a/noir/noir-repo/tooling/noirc_abi_wasm/test/shared/abi_encode.ts +++ b/noir/noir-repo/tooling/noirc_abi_wasm/test/shared/abi_encode.ts @@ -5,7 +5,7 @@ export const abi: Abi = { { name: 'foo', type: { kind: 'field' }, visibility: 'private' }, { name: 'bar', - type: { kind: 'array', length: 2, type: { kind: 'field' } }, + type: { kind: 'array', length: 3, type: { kind: 'integer', sign: 'signed', width: 32 } }, visibility: 'private', }, ], @@ -15,5 +15,5 @@ export const abi: Abi = { export const inputs: InputMap = { foo: '1', - bar: ['1', '2'], + bar: ['1', '2', '-1'], }; diff --git a/noir/noir-repo/tooling/profiler/src/cli/execution_flamegraph_cmd.rs b/noir/noir-repo/tooling/profiler/src/cli/execution_flamegraph_cmd.rs index 981d08a3eb1..6d6da89f660 100644 --- a/noir/noir-repo/tooling/profiler/src/cli/execution_flamegraph_cmd.rs +++ b/noir/noir-repo/tooling/profiler/src/cli/execution_flamegraph_cmd.rs @@ -8,7 +8,7 @@ use crate::flamegraph::{BrilligExecutionSample, FlamegraphGenerator, InfernoFlam use crate::fs::{read_inputs_from_file, read_program_from_file}; use crate::opcode_formatter::format_brillig_opcode; use bn254_blackbox_solver::Bn254BlackBoxSolver; -use nargo::ops::DefaultForeignCallExecutor; +use nargo::foreign_calls::DefaultForeignCallExecutor; use noirc_abi::input_parser::Format; use noirc_artifacts::debug::DebugArtifact;