From 5488291fe98c8be99eed842d4e8a557a083c1254 Mon Sep 17 00:00:00 2001
From: Cursor Agent <cursoragent@cursor.com>
Date: Mon, 5 Jan 2026 06:30:44 +0000
Subject: [PATCH 1/2] Pre-allocate output buffers in render_impl

Co-authored-by: gemberg <gemberg@canva.com>
---
 sparse_strips/vello_common/src/strip.rs | 12 ++++++++++++
 1 file changed, 12 insertions(+)

diff --git a/sparse_strips/vello_common/src/strip.rs b/sparse_strips/vello_common/src/strip.rs
index 731b58d38..c1c84df9d 100644
--- a/sparse_strips/vello_common/src/strip.rs
+++ b/sparse_strips/vello_common/src/strip.rs
@@ -116,6 +116,18 @@ fn render_impl<S: Simd>(
         return;
     }
 
+    // Pre-allocate output buffers.
+    //
+    // `alpha_buf` gets one 16-byte (u8x16) column written each time we advance to a new tile
+    // *location* (x,y). The number of unique locations is <= tiles.len(), so this is a safe upper
+    // bound (we may over-reserve if many tiles share the same location).
+    //
+    // `strip_buf` grows roughly with the number of locations/row transitions; reserving ~tiles.len()
+    // keeps reallocations out of the hot loop.
+    let tiles_len = tiles.len() as usize;
+    alpha_buf.reserve(tiles_len.saturating_mul(16));
+    strip_buf.reserve(tiles_len.saturating_add(8));
+
     let should_fill = |winding: i32| match fill_rule {
         Fill::NonZero => winding != 0,
         Fill::EvenOdd => winding % 2 != 0,

From af95a7e041f6b672b97e7820fe45ca262bb8f669 Mon Sep 17 00:00:00 2001
From: Cursor Agent <cursoragent@cursor.com>
Date: Mon, 5 Jan 2026 06:43:00 +0000
Subject: [PATCH 2/2] Add benchmark for first render of strips

Co-authored-by: gemberg <gemberg@canva.com>
---
 sparse_strips/vello_bench/benches/main.rs |  2 +
 sparse_strips/vello_bench/src/strip.rs    | 48 ++++++++++++++++++++++-
 2 files changed, 49 insertions(+), 1 deletion(-)

diff --git a/sparse_strips/vello_bench/benches/main.rs b/sparse_strips/vello_bench/benches/main.rs
index b9f50d4e5..b9603b797 100644
--- a/sparse_strips/vello_bench/benches/main.rs
+++ b/sparse_strips/vello_bench/benches/main.rs
@@ -18,11 +18,13 @@ criterion_group!(tile, tile::tile);
 criterion_group!(flatten, flatten::flatten);
 criterion_group!(strokes, flatten::strokes);
 criterion_group!(render_strips, strip::render_strips);
+criterion_group!(render_strips_first_render, strip::render_strips_first_render);
 criterion_group!(glyph, glyph::glyph);
 criterion_group!(integration_bench, integration::images);
 criterion_main!(
     tile,
     render_strips,
+    render_strips_first_render,
     flatten,
     strokes,
     glyph,
diff --git a/sparse_strips/vello_bench/src/strip.rs b/sparse_strips/vello_bench/src/strip.rs
index 86909419f..14b97e16d 100644
--- a/sparse_strips/vello_bench/src/strip.rs
+++ b/sparse_strips/vello_bench/src/strip.rs
@@ -2,7 +2,7 @@
 // SPDX-License-Identifier: Apache-2.0 OR MIT
 
 use crate::data::get_data_items;
-use criterion::Criterion;
+use criterion::{BatchSize, Criterion};
 use vello_common::fearless_simd::Level;
 use vello_common::peniko::Fill;
 
@@ -47,3 +47,49 @@ pub fn render_strips(c: &mut Criterion) {
         }
     }
 }
+
+/// Benchmark strip rendering in a "first render" configuration.
+///
+/// This intentionally starts with empty output buffers (capacity == 0) each iteration, so we
+/// measure the cost of growing `strip_buf`/`alpha_buf` during `strip::render_impl` (and thus any
+/// effect from internal `reserve()` calls).
+pub fn render_strips_first_render(c: &mut Criterion) {
+    let mut g = c.benchmark_group("render_strips_first_render");
+    g.sample_size(50);
+
+    macro_rules! strip_single {
+        ($item:expr, $level:expr, $suffix:expr) => {
+            let lines = $item.lines();
+            let tiles = $item.sorted_tiles();
+
+            g.bench_function(format!("{}_{}", $item.name.clone(), $suffix), |b| {
+                b.iter_batched(
+                    || (Vec::new(), Vec::new()),
+                    |(mut strip_buf, mut alpha_buf)| {
+                        vello_common::strip::render(
+                            $level,
+                            &tiles,
+                            &mut strip_buf,
+                            &mut alpha_buf,
+                            Fill::NonZero,
+                            None,
+                            &lines,
+                        );
+                        std::hint::black_box((&strip_buf, &alpha_buf));
+                        (strip_buf, alpha_buf)
+                    },
+                    BatchSize::SmallInput,
+                )
+            });
+        };
+    }
+
+    for item in get_data_items() {
+        // Commenting this out by default since SIMD is what we care about most.
+        // strip_single!(item, Level::fallback(), "fallback");
+        let simd_level = Level::new();
+        if !matches!(simd_level, Level::Fallback(_)) {
+            strip_single!(item, simd_level, "simd");
+        }
+    }
+}