update comments and readme

README.md

@@ -6,9 +6,3 @@ The reference solution will be compared to a [Rust](https://github.com/rust-lang
 
 This repository contains the benchmark program and source code of all `step`-function implementations.
 A human-readable explanation of the Rust implementations can be found on [this page](https://parallel-rust-cpp.github.io/).
-
-## Example benchmark on Intel Xeon E3-1230 v5
-
-![Single thread benchmark results comparing the amount of single precision floating point instructions per second for 8 different implementations of the same program written in both Rust and C++.](reports/Xeon-E3-1230-v5/single_core/plot2.png "Single threaded performance")
-
-![Multi thread benchmark results comparing the amount of single precision floating point instructions per second for 8 different implementations of the same program written in both Rust and C++.](reports/Xeon-E3-1230-v5/multi_core/plot2.png "Multi threaded performance")

src/rust/v7_cache_reuse/src/lib.rs

@@ -36,6 +36,7 @@ fn _step(r: &mut [f32], d: &[f32], n: usize) {
         .par_chunks_mut(vecs_per_col)
         .enumerate()
         .for_each(interleave_row);
+    // We don't need stable sort since there are no duplicate keys
     row_pairs.par_sort_unstable();
     // ANCHOR_END: interleave_apply
     }
@@ -125,7 +126,7 @@ fn _step(r: &mut [f32], d: &[f32], n: usize) {
         // ANCHOR_END: stripe_loop_step_partial_block
         #[cfg(not(feature = "no-multi-thread"))]
         // ANCHOR: stripe_loop_step_partial_block_apply
-        // Process current stripe
+        // Process current stripe in parallel, each thread filling one `tmp` block
         partial_results
             .par_chunks_mut(simd::f32x8_LENGTH)
             .zip(row_pairs.par_iter())