Coup de grâce

README.md

@@ -10,6 +10,10 @@ This repository contains the C++ simulation code and Python visualization code u
 
 ---
 
+## Results
+
+The logs of our benchmarks are stored inside the [`measurements`](./measurements/) folder. However, the log files contain incorrect flop counts since our benchmarking code assumed the same baseline flop count for all versions. Thus, we manually recalculated the flop count for each version and created helper functions for the correct flop counts inside the [`flopcounts`](./flopcounts/) folder. **The final, correct results are stored inside [`analysis/results_x86.csv`](./analysis/results_x86.csv).** [`variant_inlining_8.cpp`](./variants/variant_inlining_8.cpp) is our best non-vectorized version, while [`variant_float_inlining_SoA_AVX512_polyexp_moreavx_ilp_2_rsqrt_morersqrt_improved_moreilp.cpp`](./variants/variant_float_inlining_SoA_AVX512_polyexp_moreavx_ilp_2_rsqrt_morersqrt_improved_moreilp.cpp) is our best vectorized version. The measurements for the best non-vectorized version compiled with flags specifically disabling vectorization are stored at [`4.log`](./measurements/variant_inlining_8/x86/4.log), [`5.log`](./measurements/variant_inlining_8/x86/5.log), and [`6.log`](measurements/variant_inlining_8/x86/6.log). Additionally, the markdown files we have inside [`docs`](./docs/) are outdated and incomplete. Thus, **the final report [`70_report.pdf`](./70_report.pdf) should be used as the sole reference for all of our optimizations**.
+
 ## Requirements
 
 The project uses a `Makefile` to simplify building, running, benchmarking, testing, and visualization.