fix: compression benchmark (#3339)

Build the duckdb extension explicitly

Signed-off-by: Joe Isaacs <[email protected]>

Author: joseph-isaacs

.github/workflows/bench-pr.yml

@@ -98,6 +98,18 @@ jobs:
         run: |
           echo "${{ github.workspace }}/duckdb/build/release/" >> $GITHUB_PATH
 
+      - name: Install ninja and gzip
+        run: sudo apt-get update && sudo apt-get install -y gzip ninja-build
+
+      - name: DuckDB extension build
+        env:
+          # Build DuckDB and the Vortex extension with `-march=native`.
+          # The `NATIVE_ARCH` environment variable is picked up by `duckdb/Makefile`.
+          NATIVE_ARCH: 1
+          DUCKDB_PLATFORM: linux_amd64
+        run: GEN=ninja make release
+        working-directory: ${{ github.workspace }}/duckdb-vortex
+
       - name: Build binary
         shell: bash
         env:

.github/workflows/bench.yml

@@ -107,6 +107,15 @@ jobs:
       - name: Install gzip
         run: sudo apt-get update && sudo apt-get install -y gzip ninja-build
 
+      - name: DuckDB extension build
+        env:
+          # Build DuckDB and the Vortex extension with `-march=native`.
+          # The `NATIVE_ARCH` environment variable is picked up by `duckdb/Makefile`.
+          NATIVE_ARCH: 1
+          DUCKDB_PLATFORM: linux_amd64
+        run: GEN=ninja make release
+        working-directory: ${{ github.workspace }}/duckdb-vortex
+
       # rustup is pre-installed on the ubuntu24-full-x64 image.
       - name: Run ${{ matrix.benchmark.name }} benchmark
         shell: bash

bench-vortex/src/datasets/tpch_l_comment.rs

@@ -4,7 +4,7 @@ use vortex::dtype::FieldName;
 use vortex::{ArrayRef, IntoArray, ToCanonical};
 
 use crate::datasets::Dataset;
-use crate::ddb::{build_vortex_duckdb, duckdb_executable_path};
+use crate::ddb::duckdb_executable_path;
 use crate::tpch::duckdb::{DuckdbTpcOptions, TpcDataset, generate_tpc};
 use crate::{Format, IdempotentPath, tpch};
 
@@ -17,8 +17,6 @@ impl Dataset for TPCHLCommentChunked {
     }
 
     async fn to_vortex_array(&self) -> ArrayRef {
-        // TODO(joe): replace with a duckdb binary
-        build_vortex_duckdb();
         let duckdb_resolved_path = duckdb_executable_path(&None);
         let opts = DuckdbTpcOptions::new("tpch".to_data_path(), TpcDataset::TpcH, Format::Csv)
             .with_duckdb_path(duckdb_resolved_path.clone());