Remove heap allocations overheads in Chebyshev interpolations

.github/workflows/benchmarks.yml

@@ -1,15 +1,13 @@
 name: Run benchmarks
 
 on:
-  pull_request:
-    types: [labeled]
+  workflow_dispatch:
 
-permissions:
-  pull-requests: write
+env:
+  CARGO_TERM_COLOR: always
 
 jobs:
   benchmark:
-    if: github.event.label.name == 'benchmark'
     name: Run criterion benchmarks
     runs-on: ubuntu-latest
     steps:
@@ -22,41 +20,23 @@ jobs:
       - name: Get data
         uses: ./.github/actions/cache-data
 
-      - name: Cache Criterion baseline
+      - name: Cache Criterion results
         uses: actions/cache@v4
         with:
           path: target/criterion
-          key: criterion-benchmark-${{ runner.os }}-${{ github.ref }}
+          key: ${{ runner.os }}-criterion-${{ github.ref_name }}
           restore-keys: |
-            criterion-benchmark-${{ runner.os }}-refs/heads/main
+            ${{ runner.os }}-criterion-main
+            ${{ runner.os }}-criterion-
 
-      - name: Run benchmarks and save clean output
-        id: bench
+      - name: Run benchmarks
         run: |
           export NEOPDF_DATA_PATH=${PWD}/neopdf-data
-          # TODO: remove the version specification
-          cargo bench -p neopdf -- --nocapture | awk '/^test result: ok\./ {p=1; next} p' | tee benchmark_results.txt
+          cargo bench -p  neopdf@0.3.0-alpha4
 
-      - name: Post benchmark results as a PR comment
-        uses: actions/github-script@v7
+      - name: Upload benchmark reports
+        if: always()
+        uses: actions/upload-artifact@v4
         with:
-          script: |
-            const fs = require('fs');
-            const output = fs.readFileSync('benchmark_results.txt', 'utf8').trim();
-
-            if (output) {
-              const body = '## 🚀 Benchmark Results 🚀\n\n' +
-                           'Here are the results from the benchmark run.\n\n' +
-                           '```\n' +
-                           output + '\n' +
-                           '```';
-
-              await github.rest.issues.createComment({
-                owner: context.repo.owner,
-                repo: context.repo.repo,
-                issue_number: context.issue.number,
-                body: body
-              });
-            } else {
-              console.log("Benchmark output was empty.");
-            }
+          name: criterion-reports-${{ github.sha }}
+          path: target/criterion/

.github/workflows/run-pyapi.yml

@@ -23,7 +23,8 @@ jobs:
           cd neopdf_pyapi
           python3 -m venv --system-site-packages myenv
           . myenv/bin/activate
-          pip install maturin
+           # maturin-1.12.5 fails with: `--include-debuginfo cannot be used with --strip`
+          pip install maturin==1.12.4
           maturin develop --extras test
           export NEOPDF_DATA_PATH=/usr/share/LHAPDF
           pytest --verbose

.github/workflows/run-python.yml

@@ -34,6 +34,7 @@ jobs:
           cd neopdf_pyapi
           python -m venv env
           . env/bin/activate
-          pip install maturin
+          # maturin-1.12.5 fails with: `--include-debuginfo cannot be used with --strip`
+          pip install maturin==1.12.4
           maturin develop --extras test
           pytest -m multipleversions benches/test_versions.py

CHANGELOG.md

@@ -24,6 +24,7 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 - Added `LogFourCubic` and `LogFiveCubic` interpolation strategies for 4D and 5D data.
 - Added new methods to the Fortran and C/C++ APIs to write and compress grids
   with `xi` and `delta` dependence.
+- Added `load_by_lhaid` method to load PDF set using LHAPDF IDs.
 
 ### Changed
 

neopdf/benches/bench_pdf.rs

@@ -71,12 +71,91 @@ fn xfxq2_members(c: &mut Criterion) {
     });
 }
 
+fn xfxq2_allpids_cheby(c: &mut Criterion) {
+    let pdf = PDF::load("MAP22_grids_FF_Km_N3LL.neopdf.lz4", 0);
+    let ids: Vec<i32> = (-4..=4).filter(|&x| x != 0).collect();
+    let point = [1e-2, 5e-1, 10.0];
+    let mut out = vec![0.0; ids.len()];
+
+    let mut group = c.benchmark_group("xfxq2_allpids_cheby");
+
+    group.bench_function("fast_path", |b| {
+        b.iter(|| {
+            pdf.xfxq2_allpids(
+                std::hint::black_box(&ids),
+                std::hint::black_box(&point),
+                std::hint::black_box(&mut out),
+            )
+        })
+    });
+
+    group.bench_function("slow_path_loop", |b| {
+        b.iter(|| {
+            for (i, &pid) in ids.iter().enumerate() {
+                out[i] = pdf.xfxq2(std::hint::black_box(pid), std::hint::black_box(&point));
+            }
+        })
+    });
+
+    group.finish();
+}
+
+fn xfxq2s_cheby(c: &mut Criterion) {
+    let pdf = PDF::load("MAP22_grids_FF_Km_N3LL.neopdf.lz4", 0);
+    let ids: Vec<i32> = (-4..=4).filter(|&x| x != 0).collect();
+    let pts = [
+        [1e-4, 1e-2, 2.0],
+        [1e-3, 1e-2, 2.0],
+        [1e-2, 1e-2, 2.0],
+        [1e-1, 1e-2, 2.0],
+        [1e-4, 1e-1, 10.0],
+        [1e-3, 1e-1, 10.0],
+        [1e-2, 1e-1, 10.0],
+        [1e-1, 1e-1, 10.0],
+        [1e-4, 0.5, 100.0],
+        [1e-3, 0.5, 100.0],
+        [1e-2, 0.5, 100.0],
+        [1e-1, 0.5, 100.0],
+        [1e-4, 0.9, 1000.0],
+        [1e-3, 0.9, 1000.0],
+        [1e-2, 0.9, 1000.0],
+        [1e-1, 0.9, 1000.0],
+    ];
+    let pts_slices: Vec<&[f64]> = pts.iter().map(|p| &p[..]).collect();
+
+    let mut group = c.benchmark_group("xfxq2_matrix_cheby");
+
+    group.bench_function("flavor_batched_xfxq2s", |b| {
+        b.iter(|| {
+            pdf.xfxq2s(
+                std::hint::black_box(ids.clone()),
+                std::hint::black_box(&pts_slices),
+            )
+        })
+    });
+
+    group.bench_function("point_batched_loop", |b| {
+        b.iter(|| {
+            for &pid in &ids {
+                let _ = pdf.xfxq2_cheby_batch(
+                    std::hint::black_box(pid),
+                    std::hint::black_box(&pts_slices),
+                );
+            }
+        })
+    });
+
+    group.finish();
+}
+
 criterion_group!(
     benches,
     xfxq2,
     xfxq2s,
     xfxq2_members,
     xfxq2_cheby,
-    xfxq2_cheby_batch
+    xfxq2_cheby_batch,
+    xfxq2_allpids_cheby,
+    xfxq2s_cheby
 );
 criterion_main!(benches);