updated structure of how we document the optimizations

2026-05-20 09:26:22 +02:00
parent 1718a2122f
commit 3ce1c8cc2e
4 changed files with 257 additions and 166 deletions
@@ -0,0 +1,76 @@
+<!--
+TEMPLATE - copy this to docs/optimization_<your_name>.md before editing.
+
+Workflow:
+  1. cp docs/optimization_template.md docs/optimization_<your_name>.md
+  2. Fill in the fields below.
+  3. Add a row to the summary tables in docs/OPTIMIZATIONS.md that links to your file.
+  4. Delete this HTML comment block before committing.
+-->
+
+# Optimization: <name>
+
+- **Author**: <your name>
+- **Predecessor**: variant_baseline.cpp
+- **File(s)**: variants/variant_<your_name>.cpp
+- **Status**: <not benchmarked / negative / partial / positive>
+
+## What this changes
+
+<2-3 sentences. What did you change and what is the high-level idea?
+No full code paste here - just a description.>
+
+## Code (relevant snippet)
+
+```cpp
+// the part that differs from the predecessor
+```
+
+## Benchmark
+
+Conditions: min of <N> runs, N_ITER=1000, -O3 -march=native -flto.
+
+### x86
+
+**Machine**:
+
+```
+<paste machine info here, e.g. CPU model, core count, RAM, manufacturer, system version>
+```
+
+**Frequency**: <e.g., 2 GHz>
+
+| N    | W(n) [Flops] | T(n) [Cycles] | Performance [Flops/Cycle] | Diff vs Predecessor |
+| ---- | ------------ | ------------- | ------------------------- | ------------------- |
+| 50   |              |               |                           |                     |
+| 100  |              |               |                           |                     |
+| 200  |              |               |                           |                     |
+| 500  |              |               |                           |                     |
+| 1000 |              |               |                           |                     |
+
+### ARM (Apple)
+
+**Machine**:
+
+```
+<paste machine info here, e.g. Apple M1/M2/M3/M4, chip cores, RAM>
+```
+
+**Frequency**: <e.g., 3.2 GHz>
+
+| N    | W(n) [Flops] | T(n) [Cycles] | Performance [Flops/Cycle] | Diff vs Predecessor |
+| ---- | ------------ | ------------- | ------------------------- | ------------------- |
+| 50   |              |               |                           |                     |
+| 100  |              |               |                           |                     |
+| 200  |              |               |                           |                     |
+| 500  |              |               |                           |                     |
+| 1000 |              |               |                           |                     |
+
+## Analysis
+
+<Why it worked or did not. What surprised you. What bottleneck this hits
+or moves. Mention generated assembly or a profile if relevant.>
+
+## Takeaway
+
+<One short paragraph that could be lifted straight into the report.>