Measurements arm

2026-05-31 19:42:21 +02:00
parent e76a850b31
commit f668e3c5bf
24 changed files with 1824 additions and 1 deletions
@@ -0,0 +1,52 @@
+rm -f main_*  run_tests *.o *.s variants/*.s variants/*.o headers/*.s headers/*.o
+sudo /usr/bin/g++ main.cpp ./variants/variant_baseline.cpp -o main_baseline -O3 -march=native -std=c++23 -Wall -Wextra -flto -DFLOPCOUNT=0 -DRANDOM_SEED=0 -DROOFLINE=0 -DBENCHMARK=1
+=== N_P=50 N_ITER=200 ===
+W(n) per iteration: 183700.00 flops
+Q(n) per iteration: 6512 bytes
+I(n): 28.21 flops/byte
+W(n): 36740000.00 flops
+T(n): 18154426.88 cycles
+P(n): 2.02 flops/cycle
+
+=== N_P=100 N_ITER=200 ===
+W(n) per iteration: 727400.00 flops
+Q(n) per iteration: 12912 bytes
+I(n): 56.34 flops/byte
+W(n): 145480000.00 flops
+T(n): 74812980.50 cycles
+P(n): 1.94 flops/cycle
+
+=== N_P=200 N_ITER=200 ===
+W(n) per iteration: 2894800.00 flops
+Q(n) per iteration: 25712 bytes
+I(n): 112.59 flops/byte
+W(n): 578960000.00 flops
+T(n): 322910201.00 cycles
+P(n): 1.79 flops/cycle
+
+=== N_P=500 N_ITER=200 ===
+
+=== N_P=1000 N_ITER=200 ===
+W(n) per iteration: 72074000.00 flops
+Q(n) per iteration: 128112 bytes
+I(n): 562.59 flops/byte
+W(n): 14414800000.00 flops
+T(n): 9012317356.00 cycles
+P(n): 1.60 flops/cycle
+
+=== N_P=2000 N_ITER=200 ===
+W(n) per iteration: 288148000.00 flops
+Q(n) per iteration: 256112 bytes
+I(n): 1125.09 flops/byte
+W(n): 57629600000.00 flops
+T(n): 37090810409.00 cycles
+P(n): 1.55 flops/cycle
+
+=== N_P=5000 N_ITER=200 ===
+W(n) per iteration: 1800370000.00 flops
+Q(n) per iteration: 640112 bytes
+I(n): 2812.59 flops/byte
+W(n): 360074000000.00 flops
+T(n): 234807270918.00 cycles
+P(n): 1.53 flops/cycle
+
@@ -0,0 +1,58 @@
+rm -f main_*  run_tests *.o *.s variants/*.s variants/*.o headers/*.s headers/*.o
+sudo /usr/bin/g++ main.cpp ./variants/variant_branchless.cpp -o main_branchless -O3 -march=native -std=c++23 -Wall -Wextra -flto -DFLOPCOUNT=0 -DRANDOM_SEED=0 -DROOFLINE=0 -DBENCHMARK=1
+=== N_P=50 N_ITER=200 ===
+W(n) per iteration: 183700.00 flops
+Q(n) per iteration: 6512 bytes
+I(n): 28.21 flops/byte
+W(n): 36740000.00 flops
+T(n): 16997520.38 cycles
+P(n): 2.16 flops/cycle
+
+=== N_P=100 N_ITER=200 ===
+W(n) per iteration: 727400.00 flops
+Q(n) per iteration: 12912 bytes
+I(n): 56.34 flops/byte
+W(n): 145480000.00 flops
+T(n): 67764539.00 cycles
+P(n): 2.15 flops/cycle
+
+=== N_P=200 N_ITER=200 ===
+W(n) per iteration: 2894800.00 flops
+Q(n) per iteration: 25712 bytes
+I(n): 112.59 flops/byte
+W(n): 578960000.00 flops
+T(n): 269251877.00 cycles
+P(n): 2.15 flops/cycle
+
+=== N_P=500 N_ITER=200 ===
+W(n) per iteration: 18037000.00 flops
+Q(n) per iteration: 64112 bytes
+I(n): 281.34 flops/byte
+W(n): 3607400000.00 flops
+T(n): 1667878996.00 cycles
+P(n): 2.16 flops/cycle
+
+=== N_P=1000 N_ITER=200 ===
+W(n) per iteration: 72074000.00 flops
+Q(n) per iteration: 128112 bytes
+I(n): 562.59 flops/byte
+W(n): 14414800000.00 flops
+T(n): 6785785070.00 cycles
+P(n): 2.12 flops/cycle
+
+=== N_P=2000 N_ITER=200 ===
+W(n) per iteration: 288148000.00 flops
+Q(n) per iteration: 256112 bytes
+I(n): 1125.09 flops/byte
+W(n): 57629600000.00 flops
+T(n): 26925708355.00 cycles
+P(n): 2.14 flops/cycle
+
+=== N_P=5000 N_ITER=200 ===
+W(n) per iteration: 1800370000.00 flops
+Q(n) per iteration: 640112 bytes
+I(n): 2812.59 flops/byte
+W(n): 360074000000.00 flops
+T(n): 167430181147.00 cycles
+P(n): 2.15 flops/cycle
+
@@ -0,0 +1,58 @@
+rm -f main_*  run_tests *.o *.s variants/*.s variants/*.o headers/*.s headers/*.o
+sudo /usr/bin/g++ main.cpp ./variants/variant_branchless_sqrtfree.cpp -o main_branchless_sqrtfree -O3 -march=native -std=c++23 -Wall -Wextra -flto -DFLOPCOUNT=0 -DRANDOM_SEED=0 -DROOFLINE=0 -DBENCHMARK=1
+=== N_P=50 N_ITER=200 ===
+W(n) per iteration: 183700.00 flops
+Q(n) per iteration: 6512 bytes
+I(n): 28.21 flops/byte
+W(n): 36740000.00 flops
+T(n): 17385705.00 cycles
+P(n): 2.11 flops/cycle
+
+=== N_P=100 N_ITER=200 ===
+W(n) per iteration: 727400.00 flops
+Q(n) per iteration: 12912 bytes
+I(n): 56.34 flops/byte
+W(n): 145480000.00 flops
+T(n): 66976913.50 cycles
+P(n): 2.17 flops/cycle
+
+=== N_P=200 N_ITER=200 ===
+W(n) per iteration: 2894800.00 flops
+Q(n) per iteration: 25712 bytes
+I(n): 112.59 flops/byte
+W(n): 578960000.00 flops
+T(n): 266594191.00 cycles
+P(n): 2.17 flops/cycle
+
+=== N_P=500 N_ITER=200 ===
+W(n) per iteration: 18037000.00 flops
+Q(n) per iteration: 64112 bytes
+I(n): 281.34 flops/byte
+W(n): 3607400000.00 flops
+T(n): 1659465829.00 cycles
+P(n): 2.17 flops/cycle
+
+=== N_P=1000 N_ITER=200 ===
+W(n) per iteration: 72074000.00 flops
+Q(n) per iteration: 128112 bytes
+I(n): 562.59 flops/byte
+W(n): 14414800000.00 flops
+T(n): 6723368454.00 cycles
+P(n): 2.14 flops/cycle
+
+=== N_P=2000 N_ITER=200 ===
+W(n) per iteration: 288148000.00 flops
+Q(n) per iteration: 256112 bytes
+I(n): 1125.09 flops/byte
+W(n): 57629600000.00 flops
+T(n): 26939978394.00 cycles
+P(n): 2.14 flops/cycle
+
+=== N_P=5000 N_ITER=200 ===
+W(n) per iteration: 1800370000.00 flops
+Q(n) per iteration: 640112 bytes
+I(n): 2812.59 flops/byte
+W(n): 360074000000.00 flops
+T(n): 166305320990.00 cycles
+P(n): 2.17 flops/cycle
+
@@ -0,0 +1,58 @@
+rm -f main_*  run_tests *.o *.s variants/*.s variants/*.o headers/*.s headers/*.o
+sudo /usr/bin/g++ main.cpp ./variants/variant_cutoff.cpp -o main_cutoff -O3 -march=native -std=c++23 -Wall -Wextra -flto -DFLOPCOUNT=0 -DRANDOM_SEED=0 -DROOFLINE=0 -DBENCHMARK=1
+=== N_P=50 N_ITER=200 ===
+W(n) per iteration: 183700.00 flops
+Q(n) per iteration: 6512 bytes
+I(n): 28.21 flops/byte
+W(n): 36740000.00 flops
+T(n): 8713477.81 cycles
+P(n): 4.22 flops/cycle
+
+=== N_P=100 N_ITER=200 ===
+W(n) per iteration: 727400.00 flops
+Q(n) per iteration: 12912 bytes
+I(n): 56.34 flops/byte
+W(n): 145480000.00 flops
+T(n): 40792521.50 cycles
+P(n): 3.57 flops/cycle
+
+=== N_P=200 N_ITER=200 ===
+W(n) per iteration: 2894800.00 flops
+Q(n) per iteration: 25712 bytes
+I(n): 112.59 flops/byte
+W(n): 578960000.00 flops
+T(n): 132995997.00 cycles
+P(n): 4.35 flops/cycle
+
+=== N_P=500 N_ITER=200 ===
+W(n) per iteration: 18037000.00 flops
+Q(n) per iteration: 64112 bytes
+I(n): 281.34 flops/byte
+W(n): 3607400000.00 flops
+T(n): 813430901.00 cycles
+P(n): 4.43 flops/cycle
+
+=== N_P=1000 N_ITER=200 ===
+W(n) per iteration: 72074000.00 flops
+Q(n) per iteration: 128112 bytes
+I(n): 562.59 flops/byte
+W(n): 14414800000.00 flops
+T(n): 3155176718.00 cycles
+P(n): 4.57 flops/cycle
+
+=== N_P=2000 N_ITER=200 ===
+W(n) per iteration: 288148000.00 flops
+Q(n) per iteration: 256112 bytes
+I(n): 1125.09 flops/byte
+W(n): 57629600000.00 flops
+T(n): 12638893219.00 cycles
+P(n): 4.56 flops/cycle
+
+=== N_P=5000 N_ITER=200 ===
+W(n) per iteration: 1800370000.00 flops
+Q(n) per iteration: 640112 bytes
+I(n): 2812.59 flops/byte
+W(n): 360074000000.00 flops
+T(n): 76387953469.00 cycles
+P(n): 4.71 flops/cycle
+
@@ -0,0 +1,58 @@
+rm -f main_*  run_tests *.o *.s variants/*.s variants/*.o headers/*.s headers/*.o
+sudo /usr/bin/g++ main.cpp ./variants/variant_cutoff_callsite.cpp -o main_cutoff_callsite -O3 -march=native -std=c++23 -Wall -Wextra -flto -DFLOPCOUNT=0 -DRANDOM_SEED=0 -DROOFLINE=0 -DBENCHMARK=1
+=== N_P=50 N_ITER=200 ===
+W(n) per iteration: 183700.00 flops
+Q(n) per iteration: 6512 bytes
+I(n): 28.21 flops/byte
+W(n): 36740000.00 flops
+T(n): 9210738.56 cycles
+P(n): 3.99 flops/cycle
+
+=== N_P=100 N_ITER=200 ===
+W(n) per iteration: 727400.00 flops
+Q(n) per iteration: 12912 bytes
+I(n): 56.34 flops/byte
+W(n): 145480000.00 flops
+T(n): 43525263.50 cycles
+P(n): 3.34 flops/cycle
+
+=== N_P=200 N_ITER=200 ===
+W(n) per iteration: 2894800.00 flops
+Q(n) per iteration: 25712 bytes
+I(n): 112.59 flops/byte
+W(n): 578960000.00 flops
+T(n): 141467337.00 cycles
+P(n): 4.09 flops/cycle
+
+=== N_P=500 N_ITER=200 ===
+W(n) per iteration: 18037000.00 flops
+Q(n) per iteration: 64112 bytes
+I(n): 281.34 flops/byte
+W(n): 3607400000.00 flops
+T(n): 870479259.00 cycles
+P(n): 4.14 flops/cycle
+
+=== N_P=1000 N_ITER=200 ===
+W(n) per iteration: 72074000.00 flops
+Q(n) per iteration: 128112 bytes
+I(n): 562.59 flops/byte
+W(n): 14414800000.00 flops
+T(n): 3340600189.00 cycles
+P(n): 4.32 flops/cycle
+
+=== N_P=2000 N_ITER=200 ===
+W(n) per iteration: 288148000.00 flops
+Q(n) per iteration: 256112 bytes
+I(n): 1125.09 flops/byte
+W(n): 57629600000.00 flops
+T(n): 13325615778.00 cycles
+P(n): 4.32 flops/cycle
+
+=== N_P=5000 N_ITER=200 ===
+W(n) per iteration: 1800370000.00 flops
+Q(n) per iteration: 640112 bytes
+I(n): 2812.59 flops/byte
+W(n): 360074000000.00 flops
+T(n): 79881699571.00 cycles
+P(n): 4.51 flops/cycle
+
@@ -0,0 +1,58 @@
+rm -f main_*  run_tests *.o *.s variants/*.s variants/*.o headers/*.s headers/*.o
+sudo /usr/bin/g++ main-float.cpp ./variants/variant_float_baseline.cpp -o main_float_baseline -O3 -march=native -std=c++23 -Wall -Wextra -flto -DFLOPCOUNT=0 -DRANDOM_SEED=0 -DROOFLINE=0 -DBENCHMARK=1
+=== N_P=50 N_ITER=200 ===
+W(n) per iteration: 183700.00 flops
+Q(n) per iteration: 3256 bytes
+I(n): 56.42 flops/byte
+W(n): 36740000.00 flops
+T(n): 12278939.00 cycles
+P(n): 2.99 flops/cycle
+
+=== N_P=100 N_ITER=200 ===
+W(n) per iteration: 727400.00 flops
+Q(n) per iteration: 6456 bytes
+I(n): 112.67 flops/byte
+W(n): 145480000.00 flops
+T(n): 49596532.00 cycles
+P(n): 2.93 flops/cycle
+
+=== N_P=200 N_ITER=200 ===
+W(n) per iteration: 2894800.00 flops
+Q(n) per iteration: 12856 bytes
+I(n): 225.17 flops/byte
+W(n): 578960000.00 flops
+T(n): 203275376.00 cycles
+P(n): 2.85 flops/cycle
+
+=== N_P=500 N_ITER=200 ===
+W(n) per iteration: 18037000.00 flops
+Q(n) per iteration: 32056 bytes
+I(n): 562.67 flops/byte
+W(n): 3607399936.00 flops
+T(n): 1414936832.00 cycles
+P(n): 2.55 flops/cycle
+
+=== N_P=1000 N_ITER=200 ===
+W(n) per iteration: 72074000.00 flops
+Q(n) per iteration: 64056 bytes
+I(n): 1125.17 flops/byte
+W(n): 14414799872.00 flops
+T(n): 6075253248.00 cycles
+P(n): 2.37 flops/cycle
+
+=== N_P=2000 N_ITER=200 ===
+W(n) per iteration: 288148000.00 flops
+Q(n) per iteration: 128056 bytes
+I(n): 2250.17 flops/byte
+W(n): 57629601792.00 flops
+T(n): 26570563584.00 cycles
+P(n): 2.17 flops/cycle
+
+=== N_P=5000 N_ITER=200 ===
+W(n) per iteration: 1800369920.00 flops
+Q(n) per iteration: 320056 bytes
+I(n): 5625.17 flops/byte
+W(n): 360073986048.00 flops
+T(n): 194256240640.00 cycles
+P(n): 1.85 flops/cycle
+
@@ -0,0 +1,58 @@
+rm -f main_*  run_tests *.o *.s variants/*.s variants/*.o headers/*.s headers/*.o
+sudo /usr/bin/g++ main-float.cpp ./variants/variant_float_inlining.cpp -o main_float_inlining -O3 -march=native -std=c++23 -Wall -Wextra -flto -DFLOPCOUNT=0 -DRANDOM_SEED=0 -DROOFLINE=0 -DBENCHMARK=1
+=== N_P=50 N_ITER=200 ===
+W(n) per iteration: 183700.00 flops
+Q(n) per iteration: 3256 bytes
+I(n): 56.42 flops/byte
+W(n): 36740000.00 flops
+T(n): 10775829.00 cycles
+P(n): 3.41 flops/cycle
+
+=== N_P=100 N_ITER=200 ===
+W(n) per iteration: 727400.00 flops
+Q(n) per iteration: 6456 bytes
+I(n): 112.67 flops/byte
+W(n): 145480000.00 flops
+T(n): 47963000.00 cycles
+P(n): 3.03 flops/cycle
+
+=== N_P=200 N_ITER=200 ===
+W(n) per iteration: 2894800.00 flops
+Q(n) per iteration: 12856 bytes
+I(n): 225.17 flops/byte
+W(n): 578960000.00 flops
+T(n): 190335120.00 cycles
+P(n): 3.04 flops/cycle
+
+=== N_P=500 N_ITER=200 ===
+W(n) per iteration: 18037000.00 flops
+Q(n) per iteration: 32056 bytes
+I(n): 562.67 flops/byte
+W(n): 3607399936.00 flops
+T(n): 1412006400.00 cycles
+P(n): 2.55 flops/cycle
+
+=== N_P=1000 N_ITER=200 ===
+W(n) per iteration: 72074000.00 flops
+Q(n) per iteration: 64056 bytes
+I(n): 1125.17 flops/byte
+W(n): 14414799872.00 flops
+T(n): 6137234944.00 cycles
+P(n): 2.35 flops/cycle
+
+=== N_P=2000 N_ITER=200 ===
+W(n) per iteration: 288148000.00 flops
+Q(n) per iteration: 128056 bytes
+I(n): 2250.17 flops/byte
+W(n): 57629601792.00 flops
+T(n): 27256184832.00 cycles
+P(n): 2.11 flops/cycle
+
+=== N_P=5000 N_ITER=200 ===
+W(n) per iteration: 1800369920.00 flops
+Q(n) per iteration: 320056 bytes
+I(n): 5625.17 flops/byte
+W(n): 360073986048.00 flops
+T(n): 194280062976.00 cycles
+P(n): 1.85 flops/cycle
+
@@ -0,0 +1,58 @@
+rm -f main_*  run_tests *.o *.s variants/*.s variants/*.o headers/*.s headers/*.o
+sudo /usr/bin/g++ main-float.cpp ./variants/variant_float_inlining_SoA.cpp -o main_float_inlining_SoA -O3 -march=native -std=c++23 -Wall -Wextra -flto -DFLOPCOUNT=0 -DRANDOM_SEED=0 -DROOFLINE=0 -DBENCHMARK=1
+=== N_P=50 N_ITER=200 ===
+W(n) per iteration: 183700.00 flops
+Q(n) per iteration: 3256 bytes
+I(n): 56.42 flops/byte
+W(n): 36740000.00 flops
+T(n): 11407106.00 cycles
+P(n): 3.22 flops/cycle
+
+=== N_P=100 N_ITER=200 ===
+W(n) per iteration: 727400.00 flops
+Q(n) per iteration: 6456 bytes
+I(n): 112.67 flops/byte
+W(n): 145480000.00 flops
+T(n): 45941640.00 cycles
+P(n): 3.17 flops/cycle
+
+=== N_P=200 N_ITER=200 ===
+W(n) per iteration: 2894800.00 flops
+Q(n) per iteration: 12856 bytes
+I(n): 225.17 flops/byte
+W(n): 578960000.00 flops
+T(n): 175844320.00 cycles
+P(n): 3.29 flops/cycle
+
+=== N_P=500 N_ITER=200 ===
+W(n) per iteration: 18037000.00 flops
+Q(n) per iteration: 32056 bytes
+I(n): 562.67 flops/byte
+W(n): 3607399936.00 flops
+T(n): 1161542784.00 cycles
+P(n): 3.11 flops/cycle
+
+=== N_P=1000 N_ITER=200 ===
+W(n) per iteration: 72074000.00 flops
+Q(n) per iteration: 64056 bytes
+I(n): 1125.17 flops/byte
+W(n): 14414799872.00 flops
+T(n): 4913180160.00 cycles
+P(n): 2.93 flops/cycle
+
+=== N_P=2000 N_ITER=200 ===
+W(n) per iteration: 288148000.00 flops
+Q(n) per iteration: 128056 bytes
+I(n): 2250.17 flops/byte
+W(n): 57629601792.00 flops
+T(n): 21352656896.00 cycles
+P(n): 2.70 flops/cycle
+
+=== N_P=5000 N_ITER=200 ===
+W(n) per iteration: 1800369920.00 flops
+Q(n) per iteration: 320056 bytes
+I(n): 5625.17 flops/byte
+W(n): 360073986048.00 flops
+T(n): 153264947200.00 cycles
+P(n): 2.35 flops/cycle
+
@@ -0,0 +1,58 @@
+rm -f main_*  run_tests *.o *.s variants/*.s variants/*.o headers/*.s headers/*.o
+sudo /usr/bin/g++ main-float.cpp ./variants/variant_float_inlining_SoA_NEON_polyexp_moreavx_ilp_2_rsqrt_morrsqrt.cpp -o main_float_inlining_SoA_NEON_polyexp_moreavx_ilp_2_rsqrt_morrsqrt -O3 -march=native -std=c++23 -Wall -Wextra -flto -DFLOPCOUNT=0 -DRANDOM_SEED=0 -DROOFLINE=0 -DBENCHMARK=1
+=== N_P=50 N_ITER=200 ===
+W(n) per iteration: 183700.00 flops
+Q(n) per iteration: 3256 bytes
+I(n): 56.42 flops/byte
+W(n): 36740000.00 flops
+T(n): 4492093.50 cycles
+P(n): 8.18 flops/cycle
+
+=== N_P=100 N_ITER=200 ===
+W(n) per iteration: 727400.00 flops
+Q(n) per iteration: 6456 bytes
+I(n): 112.67 flops/byte
+W(n): 145480000.00 flops
+T(n): 15717616.00 cycles
+P(n): 9.26 flops/cycle
+
+=== N_P=200 N_ITER=200 ===
+W(n) per iteration: 2894800.00 flops
+Q(n) per iteration: 12856 bytes
+I(n): 225.17 flops/byte
+W(n): 578960000.00 flops
+T(n): 58648184.00 cycles
+P(n): 9.87 flops/cycle
+
+=== N_P=500 N_ITER=200 ===
+W(n) per iteration: 18037000.00 flops
+Q(n) per iteration: 32056 bytes
+I(n): 562.67 flops/byte
+W(n): 3607399936.00 flops
+T(n): 359248768.00 cycles
+P(n): 10.04 flops/cycle
+
+=== N_P=1000 N_ITER=200 ===
+W(n) per iteration: 72074000.00 flops
+Q(n) per iteration: 64056 bytes
+I(n): 1125.17 flops/byte
+W(n): 14414799872.00 flops
+T(n): 1415209856.00 cycles
+P(n): 10.19 flops/cycle
+
+=== N_P=2000 N_ITER=200 ===
+W(n) per iteration: 288148000.00 flops
+Q(n) per iteration: 128056 bytes
+I(n): 2250.17 flops/byte
+W(n): 57629601792.00 flops
+T(n): 5657639936.00 cycles
+P(n): 10.19 flops/cycle
+
+=== N_P=5000 N_ITER=200 ===
+W(n) per iteration: 1800369920.00 flops
+Q(n) per iteration: 320056 bytes
+I(n): 5625.17 flops/byte
+W(n): 360073986048.00 flops
+T(n): 35147825152.00 cycles
+P(n): 10.24 flops/cycle
+
@@ -0,0 +1,58 @@
+rm -f main_*  run_tests *.o *.s variants/*.s variants/*.o headers/*.s headers/*.o
+sudo /usr/bin/g++ main-float_symmetry.cpp ./variants/variant_float_inlining_SoA_symmetry.cpp -o main_float_inlining_SoA_symmetry -O3 -march=native -std=c++23 -Wall -Wextra -flto -DFLOPCOUNT=0 -DRANDOM_SEED=0 -DROOFLINE=0 -DBENCHMARK=1
+=== N_P=50 N_ITER=200 ===
+W(n) per iteration: 95500.00 flops
+Q(n) per iteration: 3256 bytes
+I(n): 29.33 flops/byte
+W(n): 19100000.00 flops
+T(n): 12284355.00 cycles
+P(n): 1.55 flops/cycle
+
+=== N_P=100 N_ITER=200 ===
+W(n) per iteration: 371000.00 flops
+Q(n) per iteration: 6456 bytes
+I(n): 57.47 flops/byte
+W(n): 74200000.00 flops
+T(n): 50450768.00 cycles
+P(n): 1.47 flops/cycle
+
+=== N_P=200 N_ITER=200 ===
+W(n) per iteration: 1462000.00 flops
+Q(n) per iteration: 12856 bytes
+I(n): 113.72 flops/byte
+W(n): 292400000.00 flops
+T(n): 206580544.00 cycles
+P(n): 1.42 flops/cycle
+
+=== N_P=500 N_ITER=200 ===
+W(n) per iteration: 9055000.00 flops
+Q(n) per iteration: 32056 bytes
+I(n): 282.47 flops/byte
+W(n): 1811000064.00 flops
+T(n): 1437377152.00 cycles
+P(n): 1.26 flops/cycle
+
+=== N_P=1000 N_ITER=200 ===
+W(n) per iteration: 36110000.00 flops
+Q(n) per iteration: 64056 bytes
+I(n): 563.73 flops/byte
+W(n): 7222000128.00 flops
+T(n): 6982579200.00 cycles
+P(n): 1.03 flops/cycle
+
+=== N_P=2000 N_ITER=200 ===
+W(n) per iteration: 144220000.00 flops
+Q(n) per iteration: 128056 bytes
+I(n): 1126.23 flops/byte
+W(n): 28843999232.00 flops
+T(n): 32683665408.00 cycles
+P(n): 0.88 flops/cycle
+
+=== N_P=5000 N_ITER=200 ===
+W(n) per iteration: 900550016.00 flops
+Q(n) per iteration: 320056 bytes
+I(n): 2813.73 flops/byte
+W(n): 180110000128.00 flops
+T(n): 232878047232.00 cycles
+P(n): 0.77 flops/cycle
+
@@ -0,0 +1,58 @@
+rm -f main_*  run_tests *.o *.s variants/*.s variants/*.o headers/*.s headers/*.o
+sudo /usr/bin/g++ main-float_symmetry.cpp ./variants/variant_float_inlining_SoA_symmetry_ILP_16.cpp -o main_float_inlining_SoA_symmetry_ILP_16 -O3 -march=native -std=c++23 -Wall -Wextra -flto -DFLOPCOUNT=0 -DRANDOM_SEED=0 -DROOFLINE=0 -DBENCHMARK=1
+=== N_P=50 N_ITER=200 ===
+W(n) per iteration: 95500.00 flops
+Q(n) per iteration: 3256 bytes
+I(n): 29.33 flops/byte
+W(n): 19100000.00 flops
+T(n): 9097177.00 cycles
+P(n): 2.10 flops/cycle
+
+=== N_P=100 N_ITER=200 ===
+W(n) per iteration: 371000.00 flops
+Q(n) per iteration: 6456 bytes
+I(n): 57.47 flops/byte
+W(n): 74200000.00 flops
+T(n): 33503178.00 cycles
+P(n): 2.21 flops/cycle
+
+=== N_P=200 N_ITER=200 ===
+W(n) per iteration: 1462000.00 flops
+Q(n) per iteration: 12856 bytes
+I(n): 113.72 flops/byte
+W(n): 292400000.00 flops
+T(n): 118530416.00 cycles
+P(n): 2.47 flops/cycle
+
+=== N_P=500 N_ITER=200 ===
+W(n) per iteration: 9055000.00 flops
+Q(n) per iteration: 32056 bytes
+I(n): 282.47 flops/byte
+W(n): 1811000064.00 flops
+T(n): 684755776.00 cycles
+P(n): 2.64 flops/cycle
+
+=== N_P=1000 N_ITER=200 ===
+W(n) per iteration: 36110000.00 flops
+Q(n) per iteration: 64056 bytes
+I(n): 563.73 flops/byte
+W(n): 7222000128.00 flops
+T(n): 2936483584.00 cycles
+P(n): 2.46 flops/cycle
+
+=== N_P=2000 N_ITER=200 ===
+W(n) per iteration: 144220000.00 flops
+Q(n) per iteration: 128056 bytes
+I(n): 1126.23 flops/byte
+W(n): 28843999232.00 flops
+T(n): 12939599872.00 cycles
+P(n): 2.23 flops/cycle
+
+=== N_P=5000 N_ITER=200 ===
+W(n) per iteration: 900550016.00 flops
+Q(n) per iteration: 320056 bytes
+I(n): 2813.73 flops/byte
+W(n): 180110000128.00 flops
+T(n): 156191883264.00 cycles
+P(n): 1.15 flops/cycle
+
@@ -0,0 +1,58 @@
+rm -f main_*  run_tests *.o *.s variants/*.s variants/*.o headers/*.s headers/*.o
+sudo /usr/bin/g++ main-float_symmetry.cpp ./variants/variant_float_inlining_SoA_symmetry_ILP_2.cpp -o main_float_inlining_SoA_symmetry_ILP_2 -O3 -march=native -std=c++23 -Wall -Wextra -flto -DFLOPCOUNT=0 -DRANDOM_SEED=0 -DROOFLINE=0 -DBENCHMARK=1
+=== N_P=50 N_ITER=200 ===
+W(n) per iteration: 95500.00 flops
+Q(n) per iteration: 3256 bytes
+I(n): 29.33 flops/byte
+W(n): 19100000.00 flops
+T(n): 9170654.00 cycles
+P(n): 2.08 flops/cycle
+
+=== N_P=100 N_ITER=200 ===
+W(n) per iteration: 371000.00 flops
+Q(n) per iteration: 6456 bytes
+I(n): 57.47 flops/byte
+W(n): 74200000.00 flops
+T(n): 33103448.00 cycles
+P(n): 2.24 flops/cycle
+
+=== N_P=200 N_ITER=200 ===
+W(n) per iteration: 1462000.00 flops
+Q(n) per iteration: 12856 bytes
+I(n): 113.72 flops/byte
+W(n): 292400000.00 flops
+T(n): 123532256.00 cycles
+P(n): 2.37 flops/cycle
+
+=== N_P=500 N_ITER=200 ===
+W(n) per iteration: 9055000.00 flops
+Q(n) per iteration: 32056 bytes
+I(n): 282.47 flops/byte
+W(n): 1811000064.00 flops
+T(n): 768825216.00 cycles
+P(n): 2.36 flops/cycle
+
+=== N_P=1000 N_ITER=200 ===
+W(n) per iteration: 36110000.00 flops
+Q(n) per iteration: 64056 bytes
+I(n): 563.73 flops/byte
+W(n): 7222000128.00 flops
+T(n): 3933374208.00 cycles
+P(n): 1.84 flops/cycle
+
+=== N_P=2000 N_ITER=200 ===
+W(n) per iteration: 144220000.00 flops
+Q(n) per iteration: 128056 bytes
+I(n): 1126.23 flops/byte
+W(n): 28843999232.00 flops
+T(n): 18220894208.00 cycles
+P(n): 1.58 flops/cycle
+
+=== N_P=5000 N_ITER=200 ===
+W(n) per iteration: 900550016.00 flops
+Q(n) per iteration: 320056 bytes
+I(n): 2813.73 flops/byte
+W(n): 180110000128.00 flops
+T(n): 154008551424.00 cycles
+P(n): 1.17 flops/cycle
+
@@ -0,0 +1,58 @@
+rm -f main_*  run_tests *.o *.s variants/*.s variants/*.o headers/*.s headers/*.o
+sudo /usr/bin/g++ main-float_symmetry.cpp ./variants/variant_float_inlining_SoA_symmetry_ILP_4.cpp -o main_float_inlining_SoA_symmetry_ILP_4 -O3 -march=native -std=c++23 -Wall -Wextra -flto -DFLOPCOUNT=0 -DRANDOM_SEED=0 -DROOFLINE=0 -DBENCHMARK=1
+=== N_P=50 N_ITER=200 ===
+W(n) per iteration: 95500.00 flops
+Q(n) per iteration: 3256 bytes
+I(n): 29.33 flops/byte
+W(n): 19100000.00 flops
+T(n): 9431654.00 cycles
+P(n): 2.03 flops/cycle
+
+=== N_P=100 N_ITER=200 ===
+W(n) per iteration: 371000.00 flops
+Q(n) per iteration: 6456 bytes
+I(n): 57.47 flops/byte
+W(n): 74200000.00 flops
+T(n): 33563136.00 cycles
+P(n): 2.21 flops/cycle
+
+=== N_P=200 N_ITER=200 ===
+W(n) per iteration: 1462000.00 flops
+Q(n) per iteration: 12856 bytes
+I(n): 113.72 flops/byte
+W(n): 292400000.00 flops
+T(n): 118019376.00 cycles
+P(n): 2.48 flops/cycle
+
+=== N_P=500 N_ITER=200 ===
+W(n) per iteration: 9055000.00 flops
+Q(n) per iteration: 32056 bytes
+I(n): 282.47 flops/byte
+W(n): 1811000064.00 flops
+T(n): 678422208.00 cycles
+P(n): 2.67 flops/cycle
+
+=== N_P=1000 N_ITER=200 ===
+W(n) per iteration: 36110000.00 flops
+Q(n) per iteration: 64056 bytes
+I(n): 563.73 flops/byte
+W(n): 7222000128.00 flops
+T(n): 2959266048.00 cycles
+P(n): 2.44 flops/cycle
+
+=== N_P=2000 N_ITER=200 ===
+W(n) per iteration: 144220000.00 flops
+Q(n) per iteration: 128056 bytes
+I(n): 1126.23 flops/byte
+W(n): 28843999232.00 flops
+T(n): 12601305088.00 cycles
+P(n): 2.29 flops/cycle
+
+=== N_P=5000 N_ITER=200 ===
+W(n) per iteration: 900550016.00 flops
+Q(n) per iteration: 320056 bytes
+I(n): 2813.73 flops/byte
+W(n): 180110000128.00 flops
+T(n): 144715956224.00 cycles
+P(n): 1.24 flops/cycle
+
@@ -0,0 +1,58 @@
+rm -f main_*  run_tests *.o *.s variants/*.s variants/*.o headers/*.s headers/*.o
+sudo /usr/bin/g++ main-float_symmetry.cpp ./variants/variant_float_inlining_SoA_symmetry_ILP_8.cpp -o main_float_inlining_SoA_symmetry_ILP_8 -O3 -march=native -std=c++23 -Wall -Wextra -flto -DFLOPCOUNT=0 -DRANDOM_SEED=0 -DROOFLINE=0 -DBENCHMARK=1
+=== N_P=50 N_ITER=200 ===
+W(n) per iteration: 95500.00 flops
+Q(n) per iteration: 3256 bytes
+I(n): 29.33 flops/byte
+W(n): 19100000.00 flops
+T(n): 8858727.00 cycles
+P(n): 2.16 flops/cycle
+
+=== N_P=100 N_ITER=200 ===
+W(n) per iteration: 371000.00 flops
+Q(n) per iteration: 6456 bytes
+I(n): 57.47 flops/byte
+W(n): 74200000.00 flops
+T(n): 35191192.00 cycles
+P(n): 2.11 flops/cycle
+
+=== N_P=200 N_ITER=200 ===
+W(n) per iteration: 1462000.00 flops
+Q(n) per iteration: 12856 bytes
+I(n): 113.72 flops/byte
+W(n): 292400000.00 flops
+T(n): 136348512.00 cycles
+P(n): 2.14 flops/cycle
+
+=== N_P=500 N_ITER=200 ===
+W(n) per iteration: 9055000.00 flops
+Q(n) per iteration: 32056 bytes
+I(n): 282.47 flops/byte
+W(n): 1811000064.00 flops
+T(n): 881212544.00 cycles
+P(n): 2.06 flops/cycle
+
+=== N_P=1000 N_ITER=200 ===
+W(n) per iteration: 36110000.00 flops
+Q(n) per iteration: 64056 bytes
+I(n): 563.73 flops/byte
+W(n): 7222000128.00 flops
+T(n): 4358949376.00 cycles
+P(n): 1.66 flops/cycle
+
+=== N_P=2000 N_ITER=200 ===
+W(n) per iteration: 144220000.00 flops
+Q(n) per iteration: 128056 bytes
+I(n): 1126.23 flops/byte
+W(n): 28843999232.00 flops
+T(n): 19275378688.00 cycles
+P(n): 1.50 flops/cycle
+
+=== N_P=5000 N_ITER=200 ===
+W(n) per iteration: 900550016.00 flops
+Q(n) per iteration: 320056 bytes
+I(n): 2813.73 flops/byte
+W(n): 180110000128.00 flops
+T(n): 158142201856.00 cycles
+P(n): 1.14 flops/cycle
+
@@ -0,0 +1,58 @@
+rm -f main_*  run_tests *.o *.s variants/*.s variants/*.o headers/*.s headers/*.o
+sudo /usr/bin/g++ main.cpp ./variants/variant_inlining_7.cpp -o main_inlining_7 -O3 -march=native -std=c++23 -Wall -Wextra -flto -DFLOPCOUNT=0 -DRANDOM_SEED=0 -DROOFLINE=0 -DBENCHMARK=1
+=== N_P=50 N_ITER=200 ===
+W(n) per iteration: 183700.00 flops
+Q(n) per iteration: 6512 bytes
+I(n): 28.21 flops/byte
+W(n): 36740000.00 flops
+T(n): 16465719.25 cycles
+P(n): 2.23 flops/cycle
+
+=== N_P=100 N_ITER=200 ===
+W(n) per iteration: 727400.00 flops
+Q(n) per iteration: 12912 bytes
+I(n): 56.34 flops/byte
+W(n): 145480000.00 flops
+T(n): 68986819.50 cycles
+P(n): 2.11 flops/cycle
+
+=== N_P=200 N_ITER=200 ===
+W(n) per iteration: 2894800.00 flops
+Q(n) per iteration: 25712 bytes
+I(n): 112.59 flops/byte
+W(n): 578960000.00 flops
+T(n): 296100633.00 cycles
+P(n): 1.96 flops/cycle
+
+=== N_P=500 N_ITER=200 ===
+W(n) per iteration: 18037000.00 flops
+Q(n) per iteration: 64112 bytes
+I(n): 281.34 flops/byte
+W(n): 3607400000.00 flops
+T(n): 2015322383.00 cycles
+P(n): 1.79 flops/cycle
+
+=== N_P=1000 N_ITER=200 ===
+W(n) per iteration: 72074000.00 flops
+Q(n) per iteration: 128112 bytes
+I(n): 562.59 flops/byte
+W(n): 14414800000.00 flops
+T(n): 8425401169.00 cycles
+P(n): 1.71 flops/cycle
+
+=== N_P=2000 N_ITER=200 ===
+W(n) per iteration: 288148000.00 flops
+Q(n) per iteration: 256112 bytes
+I(n): 1125.09 flops/byte
+W(n): 57629600000.00 flops
+T(n): 35087107382.00 cycles
+P(n): 1.64 flops/cycle
+
+=== N_P=5000 N_ITER=200 ===
+W(n) per iteration: 1800370000.00 flops
+Q(n) per iteration: 640112 bytes
+I(n): 2812.59 flops/byte
+W(n): 360074000000.00 flops
+T(n): 225285744034.00 cycles
+P(n): 1.60 flops/cycle
+
@@ -0,0 +1,58 @@
+rm -f main_*  run_tests *.o *.s variants/*.s variants/*.o headers/*.s headers/*.o
+sudo /usr/bin/g++ main.cpp ./variants/variant_inlining_7_SoA.cpp -o main_inlining_7_SoA -O3 -march=native -std=c++23 -Wall -Wextra -flto -DFLOPCOUNT=0 -DRANDOM_SEED=0 -DROOFLINE=0 -DBENCHMARK=1
+=== N_P=50 N_ITER=200 ===
+W(n) per iteration: 183700.00 flops
+Q(n) per iteration: 6512 bytes
+I(n): 28.21 flops/byte
+W(n): 36740000.00 flops
+T(n): 15734093.62 cycles
+P(n): 2.34 flops/cycle
+
+=== N_P=100 N_ITER=200 ===
+W(n) per iteration: 727400.00 flops
+Q(n) per iteration: 12912 bytes
+I(n): 56.34 flops/byte
+W(n): 145480000.00 flops
+T(n): 63384057.50 cycles
+P(n): 2.30 flops/cycle
+
+=== N_P=200 N_ITER=200 ===
+W(n) per iteration: 2894800.00 flops
+Q(n) per iteration: 25712 bytes
+I(n): 112.59 flops/byte
+W(n): 578960000.00 flops
+T(n): 258485217.00 cycles
+P(n): 2.24 flops/cycle
+
+=== N_P=500 N_ITER=200 ===
+W(n) per iteration: 18037000.00 flops
+Q(n) per iteration: 64112 bytes
+I(n): 281.34 flops/byte
+W(n): 3607400000.00 flops
+T(n): 1806293743.00 cycles
+P(n): 2.00 flops/cycle
+
+=== N_P=1000 N_ITER=200 ===
+W(n) per iteration: 72074000.00 flops
+Q(n) per iteration: 128112 bytes
+I(n): 562.59 flops/byte
+W(n): 14414800000.00 flops
+T(n): 7331209997.00 cycles
+P(n): 1.97 flops/cycle
+
+=== N_P=2000 N_ITER=200 ===
+W(n) per iteration: 288148000.00 flops
+Q(n) per iteration: 256112 bytes
+I(n): 1125.09 flops/byte
+W(n): 57629600000.00 flops
+T(n): 31741034397.00 cycles
+P(n): 1.82 flops/cycle
+
+=== N_P=5000 N_ITER=200 ===
+W(n) per iteration: 1800370000.00 flops
+Q(n) per iteration: 640112 bytes
+I(n): 2812.59 flops/byte
+W(n): 360074000000.00 flops
+T(n): 228332189160.00 cycles
+P(n): 1.58 flops/cycle
+
@@ -0,0 +1,58 @@
+rm -f main_*  run_tests *.o *.s variants/*.s variants/*.o headers/*.s headers/*.o
+sudo /usr/bin/g++ main-symmetry.cpp ./variants/variant_inlining_7_symmetry_1.cpp -o main_inlining_7_symmetry_1 -O3 -march=native -std=c++23 -Wall -Wextra -flto -DFLOPCOUNT=0 -DRANDOM_SEED=0 -DROOFLINE=0 -DBENCHMARK=1
+=== N_P=50 N_ITER=200 ===
+W(n) per iteration: 95500.00 flops
+Q(n) per iteration: 6512 bytes
+I(n): 14.67 flops/byte
+W(n): 19100000.00 flops
+T(n): 16259793.62 cycles
+P(n): 1.17 flops/cycle
+
+=== N_P=100 N_ITER=200 ===
+W(n) per iteration: 371000.00 flops
+Q(n) per iteration: 12912 bytes
+I(n): 28.73 flops/byte
+W(n): 74200000.00 flops
+T(n): 65726236.50 cycles
+P(n): 1.13 flops/cycle
+
+=== N_P=200 N_ITER=200 ===
+W(n) per iteration: 1462000.00 flops
+Q(n) per iteration: 25712 bytes
+I(n): 56.86 flops/byte
+W(n): 292400000.00 flops
+T(n): 279432230.00 cycles
+P(n): 1.05 flops/cycle
+
+=== N_P=500 N_ITER=200 ===
+W(n) per iteration: 9055000.00 flops
+Q(n) per iteration: 64112 bytes
+I(n): 141.24 flops/byte
+W(n): 1811000000.00 flops
+T(n): 2173883627.00 cycles
+P(n): 0.83 flops/cycle
+
+=== N_P=1000 N_ITER=200 ===
+W(n) per iteration: 36110000.00 flops
+Q(n) per iteration: 128112 bytes
+I(n): 281.86 flops/byte
+W(n): 7222000000.00 flops
+T(n): 9352805794.00 cycles
+P(n): 0.77 flops/cycle
+
+=== N_P=2000 N_ITER=200 ===
+W(n) per iteration: 144220000.00 flops
+Q(n) per iteration: 256112 bytes
+I(n): 563.11 flops/byte
+W(n): 28844000000.00 flops
+T(n): 41809462783.00 cycles
+P(n): 0.69 flops/cycle
+
+=== N_P=5000 N_ITER=200 ===
+W(n) per iteration: 900550000.00 flops
+Q(n) per iteration: 640112 bytes
+I(n): 1406.86 flops/byte
+W(n): 180110000000.00 flops
+T(n): 284419980420.00 cycles
+P(n): 0.63 flops/cycle
+
@@ -0,0 +1,58 @@
+rm -f main_*  run_tests *.o *.s variants/*.s variants/*.o headers/*.s headers/*.o
+sudo /usr/bin/g++ main-symmetry.cpp ./variants/variant_inlining_7_symmetry_2.cpp -o main_inlining_7_symmetry_2 -O3 -march=native -std=c++23 -Wall -Wextra -flto -DFLOPCOUNT=0 -DRANDOM_SEED=0 -DROOFLINE=0 -DBENCHMARK=1
+=== N_P=50 N_ITER=200 ===
+W(n) per iteration: 95500.00 flops
+Q(n) per iteration: 6512 bytes
+I(n): 14.67 flops/byte
+W(n): 19100000.00 flops
+T(n): 16243406.75 cycles
+P(n): 1.18 flops/cycle
+
+=== N_P=100 N_ITER=200 ===
+W(n) per iteration: 371000.00 flops
+Q(n) per iteration: 12912 bytes
+I(n): 28.73 flops/byte
+W(n): 74200000.00 flops
+T(n): 66097465.00 cycles
+P(n): 1.12 flops/cycle
+
+=== N_P=200 N_ITER=200 ===
+W(n) per iteration: 1462000.00 flops
+Q(n) per iteration: 25712 bytes
+I(n): 56.86 flops/byte
+W(n): 292400000.00 flops
+T(n): 285031402.00 cycles
+P(n): 1.03 flops/cycle
+
+=== N_P=500 N_ITER=200 ===
+W(n) per iteration: 9055000.00 flops
+Q(n) per iteration: 64112 bytes
+I(n): 141.24 flops/byte
+W(n): 1811000000.00 flops
+T(n): 2157696662.00 cycles
+P(n): 0.84 flops/cycle
+
+=== N_P=1000 N_ITER=200 ===
+W(n) per iteration: 36110000.00 flops
+Q(n) per iteration: 128112 bytes
+I(n): 281.86 flops/byte
+W(n): 7222000000.00 flops
+T(n): 9630862508.00 cycles
+P(n): 0.75 flops/cycle
+
+=== N_P=2000 N_ITER=200 ===
+W(n) per iteration: 144220000.00 flops
+Q(n) per iteration: 256112 bytes
+I(n): 563.11 flops/byte
+W(n): 28844000000.00 flops
+T(n): 41773155034.00 cycles
+P(n): 0.69 flops/cycle
+
+=== N_P=5000 N_ITER=200 ===
+W(n) per iteration: 900550000.00 flops
+Q(n) per iteration: 640112 bytes
+I(n): 1406.86 flops/byte
+W(n): 180110000000.00 flops
+T(n): 276672447552.00 cycles
+P(n): 0.65 flops/cycle
+
@@ -0,0 +1,58 @@
+rm -f main_*  run_tests *.o *.s variants/*.s variants/*.o headers/*.s headers/*.o
+sudo /usr/bin/g++ main-symmetry.cpp ./variants/variant_inlining_7_symmetry_2_ilp_1.cpp -o main_inlining_7_symmetry_2_ilp_1 -O3 -march=native -std=c++23 -Wall -Wextra -flto -DFLOPCOUNT=0 -DRANDOM_SEED=0 -DROOFLINE=0 -DBENCHMARK=1
+=== N_P=50 N_ITER=200 ===
+W(n) per iteration: 95500.00 flops
+Q(n) per iteration: 6512 bytes
+I(n): 14.67 flops/byte
+W(n): 19100000.00 flops
+T(n): 16213663.88 cycles
+P(n): 1.18 flops/cycle
+
+=== N_P=100 N_ITER=200 ===
+W(n) per iteration: 371000.00 flops
+Q(n) per iteration: 12912 bytes
+I(n): 28.73 flops/byte
+W(n): 74200000.00 flops
+T(n): 64706500.00 cycles
+P(n): 1.15 flops/cycle
+
+=== N_P=200 N_ITER=200 ===
+W(n) per iteration: 1462000.00 flops
+Q(n) per iteration: 25712 bytes
+I(n): 56.86 flops/byte
+W(n): 292400000.00 flops
+T(n): 266479325.00 cycles
+P(n): 1.10 flops/cycle
+
+=== N_P=500 N_ITER=200 ===
+W(n) per iteration: 9055000.00 flops
+Q(n) per iteration: 64112 bytes
+I(n): 141.24 flops/byte
+W(n): 1811000000.00 flops
+T(n): 1860823343.00 cycles
+P(n): 0.97 flops/cycle
+
+=== N_P=1000 N_ITER=200 ===
+W(n) per iteration: 36110000.00 flops
+Q(n) per iteration: 128112 bytes
+I(n): 281.86 flops/byte
+W(n): 7222000000.00 flops
+T(n): 8047713694.00 cycles
+P(n): 0.90 flops/cycle
+
+=== N_P=2000 N_ITER=200 ===
+W(n) per iteration: 144220000.00 flops
+Q(n) per iteration: 256112 bytes
+I(n): 563.11 flops/byte
+W(n): 28844000000.00 flops
+T(n): 35415664148.00 cycles
+P(n): 0.81 flops/cycle
+
+=== N_P=5000 N_ITER=200 ===
+W(n) per iteration: 900550000.00 flops
+Q(n) per iteration: 640112 bytes
+I(n): 1406.86 flops/byte
+W(n): 180110000000.00 flops
+T(n): 188828167800.00 cycles
+P(n): 0.95 flops/cycle
+
@@ -0,0 +1,58 @@
+rm -f main_*  run_tests *.o *.s variants/*.s variants/*.o headers/*.s headers/*.o
+sudo /usr/bin/g++ main-symmetry.cpp ./variants/variant_inlining_7_symmetry_2_ilp_2.cpp -o main_inlining_7_symmetry_2_ilp_2 -O3 -march=native -std=c++23 -Wall -Wextra -flto -DFLOPCOUNT=0 -DRANDOM_SEED=0 -DROOFLINE=0 -DBENCHMARK=1
+=== N_P=50 N_ITER=200 ===
+W(n) per iteration: 95500.00 flops
+Q(n) per iteration: 6512 bytes
+I(n): 14.67 flops/byte
+W(n): 19100000.00 flops
+T(n): 16951060.50 cycles
+P(n): 1.13 flops/cycle
+
+=== N_P=100 N_ITER=200 ===
+W(n) per iteration: 371000.00 flops
+Q(n) per iteration: 12912 bytes
+I(n): 28.73 flops/byte
+W(n): 74200000.00 flops
+T(n): 65369333.50 cycles
+P(n): 1.14 flops/cycle
+
+=== N_P=200 N_ITER=200 ===
+W(n) per iteration: 1462000.00 flops
+Q(n) per iteration: 25712 bytes
+I(n): 56.86 flops/byte
+W(n): 292400000.00 flops
+T(n): 264705723.00 cycles
+P(n): 1.10 flops/cycle
+
+=== N_P=500 N_ITER=200 ===
+W(n) per iteration: 9055000.00 flops
+Q(n) per iteration: 64112 bytes
+I(n): 141.24 flops/byte
+W(n): 1811000000.00 flops
+T(n): 1782045217.00 cycles
+P(n): 1.02 flops/cycle
+
+=== N_P=1000 N_ITER=200 ===
+W(n) per iteration: 36110000.00 flops
+Q(n) per iteration: 128112 bytes
+I(n): 281.86 flops/byte
+W(n): 7222000000.00 flops
+T(n): 7925808103.00 cycles
+P(n): 0.91 flops/cycle
+
+=== N_P=2000 N_ITER=200 ===
+W(n) per iteration: 144220000.00 flops
+Q(n) per iteration: 256112 bytes
+I(n): 563.11 flops/byte
+W(n): 28844000000.00 flops
+T(n): 35824664520.00 cycles
+P(n): 0.81 flops/cycle
+
+=== N_P=5000 N_ITER=200 ===
+W(n) per iteration: 900550000.00 flops
+Q(n) per iteration: 640112 bytes
+I(n): 1406.86 flops/byte
+W(n): 180110000000.00 flops
+T(n): 247389191708.00 cycles
+P(n): 0.73 flops/cycle
+
@@ -0,0 +1,58 @@
+rm -f main_*  run_tests *.o *.s variants/*.s variants/*.o headers/*.s headers/*.o
+sudo /usr/bin/g++ main.cpp ./variants/variant_inlining_8.cpp -o main_inlining_8 -O3 -march=native -std=c++23 -Wall -Wextra -flto -DFLOPCOUNT=0 -DRANDOM_SEED=0 -DROOFLINE=0 -DBENCHMARK=1
+=== N_P=50 N_ITER=200 ===
+W(n) per iteration: 183700.00 flops
+Q(n) per iteration: 6512 bytes
+I(n): 28.21 flops/byte
+W(n): 36740000.00 flops
+T(n): 15054758.12 cycles
+P(n): 2.44 flops/cycle
+
+=== N_P=100 N_ITER=200 ===
+W(n) per iteration: 727400.00 flops
+Q(n) per iteration: 12912 bytes
+I(n): 56.34 flops/byte
+W(n): 145480000.00 flops
+T(n): 59392245.00 cycles
+P(n): 2.45 flops/cycle
+
+=== N_P=200 N_ITER=200 ===
+W(n) per iteration: 2894800.00 flops
+Q(n) per iteration: 25712 bytes
+I(n): 112.59 flops/byte
+W(n): 578960000.00 flops
+T(n): 238841666.00 cycles
+P(n): 2.42 flops/cycle
+
+=== N_P=500 N_ITER=200 ===
+W(n) per iteration: 18037000.00 flops
+Q(n) per iteration: 64112 bytes
+I(n): 281.34 flops/byte
+W(n): 3607400000.00 flops
+T(n): 1555607186.00 cycles
+P(n): 2.32 flops/cycle
+
+=== N_P=1000 N_ITER=200 ===
+W(n) per iteration: 72074000.00 flops
+Q(n) per iteration: 128112 bytes
+I(n): 562.59 flops/byte
+W(n): 14414800000.00 flops
+T(n): 6418988746.00 cycles
+P(n): 2.25 flops/cycle
+
+=== N_P=2000 N_ITER=200 ===
+W(n) per iteration: 288148000.00 flops
+Q(n) per iteration: 256112 bytes
+I(n): 1125.09 flops/byte
+W(n): 57629600000.00 flops
+T(n): 25996385057.00 cycles
+P(n): 2.22 flops/cycle
+
+=== N_P=5000 N_ITER=200 ===
+W(n) per iteration: 1800370000.00 flops
+Q(n) per iteration: 640112 bytes
+I(n): 2812.59 flops/byte
+W(n): 360074000000.00 flops
+T(n): 183377209437.00 cycles
+P(n): 1.96 flops/cycle
+
@@ -0,0 +1,58 @@
+rm -f main_*  run_tests *.o *.s variants/*.s variants/*.o headers/*.s headers/*.o
+sudo /usr/bin/g++ main.cpp ./variants/variant_lut_exp.cpp -o main_lut_exp -O3 -march=native -std=c++23 -Wall -Wextra -flto -DFLOPCOUNT=0 -DRANDOM_SEED=0 -DROOFLINE=0 -DBENCHMARK=1
+=== N_P=50 N_ITER=200 ===
+W(n) per iteration: 183700.00 flops
+Q(n) per iteration: 6512 bytes
+I(n): 28.21 flops/byte
+W(n): 36740000.00 flops
+T(n): 13721813.88 cycles
+P(n): 2.68 flops/cycle
+
+=== N_P=100 N_ITER=200 ===
+W(n) per iteration: 727400.00 flops
+Q(n) per iteration: 12912 bytes
+I(n): 56.34 flops/byte
+W(n): 145480000.00 flops
+T(n): 55725180.50 cycles
+P(n): 2.61 flops/cycle
+
+=== N_P=200 N_ITER=200 ===
+W(n) per iteration: 2894800.00 flops
+Q(n) per iteration: 25712 bytes
+I(n): 112.59 flops/byte
+W(n): 578960000.00 flops
+T(n): 213951168.00 cycles
+P(n): 2.71 flops/cycle
+
+=== N_P=500 N_ITER=200 ===
+W(n) per iteration: 18037000.00 flops
+Q(n) per iteration: 64112 bytes
+I(n): 281.34 flops/byte
+W(n): 3607400000.00 flops
+T(n): 1353984010.00 cycles
+P(n): 2.66 flops/cycle
+
+=== N_P=1000 N_ITER=200 ===
+W(n) per iteration: 72074000.00 flops
+Q(n) per iteration: 128112 bytes
+I(n): 562.59 flops/byte
+W(n): 14414800000.00 flops
+T(n): 5608638594.00 cycles
+P(n): 2.57 flops/cycle
+
+=== N_P=2000 N_ITER=200 ===
+W(n) per iteration: 288148000.00 flops
+Q(n) per iteration: 256112 bytes
+I(n): 1125.09 flops/byte
+W(n): 57629600000.00 flops
+T(n): 22755683575.00 cycles
+P(n): 2.53 flops/cycle
+
+=== N_P=5000 N_ITER=200 ===
+W(n) per iteration: 1800370000.00 flops
+Q(n) per iteration: 640112 bytes
+I(n): 2812.59 flops/byte
+W(n): 360074000000.00 flops
+T(n): 151113844624.00 cycles
+P(n): 2.38 flops/cycle
+
@@ -0,0 +1,553 @@
+// # Changes: NEON port of variant_float_inlining_SoA_AVX512_polyexp_moreavx_ilp_2_rsqrt.cpp
+//   - 16-wide AVX-512 → 4-wide NEON; ILP unroll by 2 preserved (inner loop processes 8 j's per iter)
+//   - rsqrt14_ps → vrsqrteq_f32 + 1 Newton iteration (single-precision accuracy)
+//   - reduce_add_ps → vaddvq_f32
+//   - Padding multiple changed from 32 to 8
+//   - Scalar obstacle loop unchanged
+
+#include "../headers/board_reduced_float.hpp"
+#include "../headers/constants_float.hpp"
+#include "../headers/pedestrian_reduced_float.hpp"
+
+#include <arm_neon.h>
+
+#include <cmath>
+#include <fstream>
+#include <random>
+#include <tuple>
+#include <vector>
+
+#define SEED 42
+#define BOARD_WIDTH 10.0f
+#define BOARD_LENGTH 50.0f
+#define WALL_EPS 1e-3f
+#define BOARD_WIDTH_MINUS_WALL_EPS 9.999f
+
+#ifndef RANDOM_SEED
+#define RANDOM_SEED 0
+#endif
+
+#ifndef BENCHMARK
+#define BENCHMARK 0
+#endif
+
+using namespace std;
+
+
+alignas(64) vector<float> pos_x, pos_y;
+alignas(64) vector<float> sb_sq;
+alignas(64) vector<float> sb_e_x, sb_e_y;
+alignas(64) vector<float> e_x, e_y;
+alignas(64) vector<float> curr_vel_x, curr_vel_y;
+alignas(64) vector<float> w_x, w_y;
+alignas(64) vector<float> desired_v0;
+alignas(64) vector<float> v_max;
+alignas(64) vector<float> destination_x, destination_y;
+alignas(64) vector<int> color;
+
+Pedestrian *init_pedestrians(int n_p) {
+#if RANDOM_SEED
+  random_device rd;
+  mt19937 gen(rd());
+#else
+  mt19937 gen(SEED);
+#endif
+
+  normal_distribution<float> velocity_gen(v0_mean, v0_std_dev);
+  uniform_real_distribution<float> y_gen(0.3, BOARD_WIDTH - 0.3);
+  uniform_real_distribution<float> x_gen_left(2.0, BOARD_LENGTH / 2.0);
+  uniform_real_distribution<float> x_gen_right(BOARD_LENGTH / 2.0, BOARD_LENGTH - 2.0);
+
+  
+  const int n_p_pad = (n_p + 7) & ~7;
+
+  pos_x.resize(n_p_pad);
+  pos_y.resize(n_p_pad);
+  sb_sq.resize(n_p_pad);
+  sb_e_x.resize(n_p_pad);
+  sb_e_y.resize(n_p_pad);
+  e_x.resize(n_p_pad);
+  e_y.resize(n_p_pad);
+  curr_vel_x.resize(n_p_pad);
+  curr_vel_y.resize(n_p_pad);
+  w_x.resize(n_p_pad);
+  w_y.resize(n_p_pad);
+  desired_v0.resize(n_p_pad);
+  v_max.resize(n_p_pad);
+  destination_x.resize(n_p_pad);
+  destination_y.resize(n_p_pad);
+  color.resize(n_p_pad);
+
+  int n_p_half = n_p / 2;
+
+  for (int i = 0; i < n_p_half; i++) {
+    float x = x_gen_left(gen);
+    float y = y_gen(gen);
+
+    float dest_x = BOARD_LENGTH + 30.0f;
+    float dest_y = y;
+
+    float init_e_x = dest_x - x;
+    float init_e_y = dest_y - y;
+    float init_e_norm = sqrt(init_e_x * init_e_x + init_e_y * init_e_y);
+    float init_e_unit_x = init_e_x / init_e_norm;
+    float init_e_unit_y = init_e_y / init_e_norm;
+
+    float d_v0;
+    do {
+      d_v0 = velocity_gen(gen);
+    } while (d_v0 <= 0.1f);
+
+    float v_m = 1.3f * d_v0;
+
+    color[i] = 0;
+    pos_x[i] = x;
+    pos_y[i] = y;
+    e_x[i] = init_e_unit_x;
+    e_y[i] = init_e_unit_y;
+    v_max[i] = v_m;
+    desired_v0[i] = d_v0;
+    destination_x[i] = dest_x;
+    destination_y[i] = dest_y;
+
+    w_x[i] = 0.0f;
+    w_y[i] = 0.0f;
+    curr_vel_x[i] = 0.0f;
+    curr_vel_y[i] = 0.0f;
+    sb_sq[i] = 0.0f;
+    sb_e_x[i] = 0.0f;
+    sb_e_y[i] = 0.0f;
+  }
+
+  for (int i = n_p_half; i < n_p; i++) {
+    float x = x_gen_right(gen);
+    float y = y_gen(gen);
+
+    float dest_x = -30.0f;
+    float dest_y = y;
+
+    float init_e_x = dest_x - x;
+    float init_e_y = dest_y - y;
+    float init_e_norm = sqrt(init_e_x * init_e_x + init_e_y * init_e_y);
+    float init_e_unit_x = init_e_x / init_e_norm;
+    float init_e_unit_y = init_e_y / init_e_norm;
+
+    float d_v0;
+    do {
+      d_v0 = velocity_gen(gen);
+    } while (d_v0 <= 0.1f);
+
+    float v_m = 1.3f * d_v0;
+
+    color[i] = 1;
+    pos_x[i] = x;
+    pos_y[i] = y;
+    e_x[i] = init_e_unit_x;
+    e_y[i] = init_e_unit_y;
+    v_max[i] = v_m;
+    desired_v0[i] = d_v0;
+    destination_x[i] = dest_x;
+    destination_y[i] = dest_y;
+
+    w_x[i] = 0.0f;
+    w_y[i] = 0.0f;
+    curr_vel_x[i] = 0.0f;
+    curr_vel_y[i] = 0.0f;
+    sb_sq[i] = 0.0f;
+    sb_e_x[i] = 0.0f;
+    sb_e_y[i] = 0.0f;
+  }
+
+  for (int j = n_p; j < n_p_pad; j++) {
+    pos_x[j] = 1e9f;
+    pos_y[j] = 1e9f;
+    sb_sq[j] = 0.0f;
+    sb_e_x[j] = 0.0f;
+    sb_e_y[j] = 0.0f;
+  }
+  return nullptr;
+}
+
+Board init_board_without_obstacles() { return Board(BOARD_LENGTH, BOARD_WIDTH, 0, nullptr); }
+
+
+static inline float32x4_t _neon_exp_ps(float32x4_t x) {
+  const float32x4_t log2e = vdupq_n_f32(1.4426950408889634f);
+  float32x4_t nf = vrndnq_f32(vmulq_f32(x, log2e));
+
+  const float32x4_t ln2 = vdupq_n_f32(0.6931471805599453f);
+  float32x4_t r = vfmsq_f32(x, nf, ln2);   // r = x - nf*ln2
+
+  const float32x4_t c5 = vdupq_n_f32(1.0f / 120.0f);
+  const float32x4_t c4 = vdupq_n_f32(1.0f / 24.0f);
+  const float32x4_t c3 = vdupq_n_f32(1.0f / 6.0f);
+  const float32x4_t c2 = vdupq_n_f32(0.5f);
+  const float32x4_t c1 = vdupq_n_f32(1.0f);
+  const float32x4_t c0 = vdupq_n_f32(1.0f);
+
+  float32x4_t p = c5;
+  p = vfmaq_f32(c4, p, r);
+  p = vfmaq_f32(c3, p, r);
+  p = vfmaq_f32(c2, p, r);
+  p = vfmaq_f32(c1, p, r);
+  p = vfmaq_f32(c0, p, r);
+
+  int32x4_t ni     = vcvtq_s32_f32(nf);
+  int32x4_t biased = vaddq_s32(ni, vdupq_n_s32(127));
+  int32x4_t scaled = vshlq_n_s32(biased, 23);
+  float32x4_t pow2n = vreinterpretq_f32_s32(scaled);
+
+  return vmulq_f32(p, pow2n);
+}
+
+// Newton-Raphson refinement formula.
+static inline float32x4_t _neon_rsqrt_ps(float32x4_t x) {
+  float32x4_t y = vrsqrteq_f32(x);
+  y = vmulq_f32(y, vrsqrtsq_f32(vmulq_f32(x, y), y));
+  return y;
+}
+
+int run_simulation(Pedestrian *pedestrians, Board &board, int n_p, int n_iter, const char *dump_path) {
+  (void)pedestrians;
+#if BENCHMARK
+  (void)dump_path;
+#endif
+
+#if !BENCHMARK
+  const int dump_interval = 1;
+  vector<tuple<int, int, float, float, float, float, int>> samples;
+
+  if (dump_path) {
+    samples.reserve(static_cast<size_t>(n_iter / dump_interval + 1) * n_p);
+    for (int i = 0; i < n_p; i++) {
+      samples.emplace_back(0, i, pos_x[i], pos_y[i], e_x[i], e_y[i], color[i]);
+    }
+  }
+#endif
+
+  const int n_p_pad = (n_p + 7) & ~7;
+
+  const float32x4_t lane_off = {0.0f, 1.0f, 2.0f, 3.0f};
+
+  const float32x4_t v_neg_v0_div_2s = vdupq_n_f32(NEG_V0_AB_div_TWO_SIGMA);
+  const float32x4_t v_neg_2s_inv    = vdupq_n_f32(NEG_TWO_SIGMA_INV);
+  const float32x4_t v_cos_phi_sq    = vdupq_n_f32(cos_phi_sq);
+  const float32x4_t v_dt_neg        = vdupq_n_f32(dt_neg);
+  const float32x4_t v_c_neg_dt      = vdupq_n_f32(c_neg_times_dt);
+  const float32x4_t v_zero          = vdupq_n_f32(0.0f);
+  const float32x4_t v_one           = vdupq_n_f32(1.0f);
+  const float32x4_t v_eps_sq        = vdupq_n_f32(EPS_SQ);
+  const float32x4_t v_np            = vdupq_n_f32((float)n_p);
+  const float32x4_t v_exp_min       = vdupq_n_f32(-87.0f);
+
+  for (int t = 0; t < n_iter; t++) {
+    for (int i = 0; i < n_p; i++) {
+      
+      float desired_velocity_x = e_x[i] * desired_v0[i];
+      float desired_velocity_y = e_y[i] * desired_v0[i];
+
+      float f_alpha_dt_x = (desired_velocity_x - curr_vel_x[i]) * dt_times_tau_alpha_inv;
+      float f_alpha_dt_y = (desired_velocity_y - curr_vel_y[i]) * dt_times_tau_alpha_inv;
+
+      const float32x4_t pix = vdupq_n_f32(pos_x[i]);
+      const float32x4_t piy = vdupq_n_f32(pos_y[i]);
+      const float32x4_t eix = vdupq_n_f32(e_x[i]);
+      const float32x4_t eiy = vdupq_n_f32(e_y[i]);
+      const float32x4_t v_i = vdupq_n_f32((float)i);
+
+      
+      float32x4_t acc_x_0 = vdupq_n_f32(0.0f);
+      float32x4_t acc_y_0 = vdupq_n_f32(0.0f);
+      float32x4_t acc_x_1 = vdupq_n_f32(0.0f);
+      float32x4_t acc_y_1 = vdupq_n_f32(0.0f);
+
+      for (int j = 0; j < n_p_pad; j += 8) {
+        
+        float32x4_t j_idx_0 = vaddq_f32(vdupq_n_f32((float)j),       lane_off);
+        float32x4_t j_idx_1 = vaddq_f32(vdupq_n_f32((float)(j + 4)), lane_off);
+
+        uint32x4_t m_neq_i_0    = vmvnq_u32(vceqq_f32(j_idx_0, v_i));
+        uint32x4_t m_neq_i_1    = vmvnq_u32(vceqq_f32(j_idx_1, v_i));
+        uint32x4_t m_in_range_0 = vcltq_f32(j_idx_0, v_np);
+        uint32x4_t m_in_range_1 = vcltq_f32(j_idx_1, v_np);
+        uint32x4_t active_0     = vandq_u32(m_neq_i_0, m_in_range_0);
+        uint32x4_t active_1     = vandq_u32(m_neq_i_1, m_in_range_1);
+
+        float32x4_t pjx_0    = vld1q_f32(&pos_x[j]);
+        float32x4_t pjx_1    = vld1q_f32(&pos_x[j + 4]);
+        float32x4_t pjy_0    = vld1q_f32(&pos_y[j]);
+        float32x4_t pjy_1    = vld1q_f32(&pos_y[j + 4]);
+        float32x4_t sbe_jx_0 = vld1q_f32(&sb_e_x[j]);
+        float32x4_t sbe_jx_1 = vld1q_f32(&sb_e_x[j + 4]);
+        float32x4_t sbe_jy_0 = vld1q_f32(&sb_e_y[j]);
+        float32x4_t sbe_jy_1 = vld1q_f32(&sb_e_y[j + 4]);
+        float32x4_t sb_jsq_0 = vld1q_f32(&sb_sq[j]);
+        float32x4_t sb_jsq_1 = vld1q_f32(&sb_sq[j + 4]);
+
+        
+        float32x4_t r_ab_x_0 = vsubq_f32(pix, pjx_0);
+        float32x4_t r_ab_x_1 = vsubq_f32(pix, pjx_1);
+        float32x4_t r_ab_y_0 = vsubq_f32(piy, pjy_0);
+        float32x4_t r_ab_y_1 = vsubq_f32(piy, pjy_1);
+
+        float32x4_t r_ab_n2_0 = vfmaq_f32(vmulq_f32(r_ab_y_0, r_ab_y_0), r_ab_x_0, r_ab_x_0);
+        float32x4_t r_ab_n2_1 = vfmaq_f32(vmulq_f32(r_ab_y_1, r_ab_y_1), r_ab_x_1, r_ab_x_1);
+
+        r_ab_n2_0 = vmaxq_f32(r_ab_n2_0, v_eps_sq);
+        r_ab_n2_1 = vmaxq_f32(r_ab_n2_1, v_eps_sq);
+
+        float32x4_t inv_r_ab_n_0 = _neon_rsqrt_ps(r_ab_n2_0);
+        float32x4_t inv_r_ab_n_1 = _neon_rsqrt_ps(r_ab_n2_1);
+
+        float32x4_t r_ab_n_0 = vmulq_f32(r_ab_n2_0, inv_r_ab_n_0);
+        float32x4_t r_ab_n_1 = vmulq_f32(r_ab_n2_1, inv_r_ab_n_1);
+
+        
+        float32x4_t r_ab2_x_0 = vsubq_f32(r_ab_x_0, sbe_jx_0);
+        float32x4_t r_ab2_x_1 = vsubq_f32(r_ab_x_1, sbe_jx_1);
+        float32x4_t r_ab2_y_0 = vsubq_f32(r_ab_y_0, sbe_jy_0);
+        float32x4_t r_ab2_y_1 = vsubq_f32(r_ab_y_1, sbe_jy_1);
+
+        float32x4_t r_ab2_n2_0 = vfmaq_f32(vmulq_f32(r_ab2_y_0, r_ab2_y_0), r_ab2_x_0, r_ab2_x_0);
+        float32x4_t r_ab2_n2_1 = vfmaq_f32(vmulq_f32(r_ab2_y_1, r_ab2_y_1), r_ab2_x_1, r_ab2_x_1);
+
+        r_ab2_n2_0 = vmaxq_f32(r_ab2_n2_0, v_eps_sq);
+        r_ab2_n2_1 = vmaxq_f32(r_ab2_n2_1, v_eps_sq);
+
+        float32x4_t inv_r_ab2_n_0 = _neon_rsqrt_ps(r_ab2_n2_0);
+        float32x4_t inv_r_ab2_n_1 = _neon_rsqrt_ps(r_ab2_n2_1);
+
+        float32x4_t r_ab2_n_0 = vmulq_f32(r_ab2_n2_0, inv_r_ab2_n_0);
+        float32x4_t r_ab2_n_1 = vmulq_f32(r_ab2_n2_1, inv_r_ab2_n_1);
+
+        float32x4_t r_sum_0 = vaddq_f32(r_ab_n_0, r_ab2_n_0);
+        float32x4_t r_sum_1 = vaddq_f32(r_ab_n_1, r_ab2_n_1);
+
+        
+        float32x4_t two_b_sq_0 = vsubq_f32(vmulq_f32(r_sum_0, r_sum_0), sb_jsq_0);
+        float32x4_t two_b_sq_1 = vsubq_f32(vmulq_f32(r_sum_1, r_sum_1), sb_jsq_1);
+
+        two_b_sq_0 = vmaxq_f32(two_b_sq_0, v_eps_sq);
+        two_b_sq_1 = vmaxq_f32(two_b_sq_1, v_eps_sq);
+
+        float32x4_t inv_two_b_0 = _neon_rsqrt_ps(two_b_sq_0);
+        float32x4_t inv_two_b_1 = _neon_rsqrt_ps(two_b_sq_1);
+
+        float32x4_t two_b_0 = vmulq_f32(two_b_sq_0, inv_two_b_0);
+        float32x4_t two_b_1 = vmulq_f32(two_b_sq_1, inv_two_b_1);
+
+        float32x4_t e_arg_0 = vmulq_f32(two_b_0, v_neg_2s_inv);
+        float32x4_t e_arg_1 = vmulq_f32(two_b_1, v_neg_2s_inv);
+
+        
+        e_arg_0 = vmaxq_f32(e_arg_0, v_exp_min);
+        e_arg_1 = vmaxq_f32(e_arg_1, v_exp_min);
+
+        float32x4_t e_val_0 = _neon_exp_ps(e_arg_0);
+        float32x4_t e_val_1 = _neon_exp_ps(e_arg_1);
+
+        
+        float32x4_t numer_inv_two_b_0 = vmulq_f32(
+            vmulq_f32(r_sum_0, v_neg_v0_div_2s),
+            vmulq_f32(e_val_0, inv_two_b_0));
+        float32x4_t numer_inv_two_b_1 = vmulq_f32(
+            vmulq_f32(r_sum_1, v_neg_v0_div_2s),
+            vmulq_f32(e_val_1, inv_two_b_1));
+
+        float32x4_t sm_r_ab_0  = vmulq_f32(numer_inv_two_b_0, inv_r_ab_n_0);
+        float32x4_t sm_r_ab_1  = vmulq_f32(numer_inv_two_b_1, inv_r_ab_n_1);
+        float32x4_t sm_r_ab2_0 = vmulq_f32(numer_inv_two_b_0, inv_r_ab2_n_0);
+        float32x4_t sm_r_ab2_1 = vmulq_f32(numer_inv_two_b_1, inv_r_ab2_n_1);
+
+        
+        float32x4_t f_x_0 = vfmaq_f32(vmulq_f32(r_ab_x_0, sm_r_ab_0), r_ab2_x_0, sm_r_ab2_0);
+        float32x4_t f_x_1 = vfmaq_f32(vmulq_f32(r_ab_x_1, sm_r_ab_1), r_ab2_x_1, sm_r_ab2_1);
+        float32x4_t f_y_0 = vfmaq_f32(vmulq_f32(r_ab_y_0, sm_r_ab_0), r_ab2_y_0, sm_r_ab2_0);
+        float32x4_t f_y_1 = vfmaq_f32(vmulq_f32(r_ab_y_1, sm_r_ab_1), r_ab2_y_1, sm_r_ab2_1);
+
+        
+        float32x4_t f_norm_sq_0 = vfmaq_f32(vmulq_f32(f_y_0, f_y_0), f_x_0, f_x_0);
+        float32x4_t f_norm_sq_1 = vfmaq_f32(vmulq_f32(f_y_1, f_y_1), f_x_1, f_x_1);
+
+        float32x4_t rhs_0 = vmulq_f32(f_norm_sq_0, v_cos_phi_sq);
+        float32x4_t rhs_1 = vmulq_f32(f_norm_sq_1, v_cos_phi_sq);
+
+        float32x4_t e_dot_f_0 = vfmaq_f32(vmulq_f32(eiy, f_y_0), eix, f_x_0);
+        float32x4_t e_dot_f_1 = vfmaq_f32(vmulq_f32(eiy, f_y_1), eix, f_x_1);
+
+        float32x4_t e_dot_f_sq_0 = vmulq_f32(e_dot_f_0, e_dot_f_0);
+        float32x4_t e_dot_f_sq_1 = vmulq_f32(e_dot_f_1, e_dot_f_1);
+
+        uint32x4_t c1_0 = vcltq_f32(e_dot_f_0, v_zero);
+        uint32x4_t c1_1 = vcltq_f32(e_dot_f_1, v_zero);
+        uint32x4_t c2_0 = vcgtq_f32(e_dot_f_sq_0, rhs_0);
+        uint32x4_t c2_1 = vcgtq_f32(e_dot_f_sq_1, rhs_1);
+
+        uint32x4_t cond_0 = vandq_u32(c1_0, c2_0);
+        uint32x4_t cond_1 = vandq_u32(c1_1, c2_1);
+
+        float32x4_t w_val_0 = vbslq_f32(cond_0, v_c_neg_dt, v_dt_neg);
+        float32x4_t w_val_1 = vbslq_f32(cond_1, v_c_neg_dt, v_dt_neg);
+
+        
+        float32x4_t f_x_safe_0 = vbslq_f32(active_0, f_x_0, v_zero);
+        float32x4_t f_x_safe_1 = vbslq_f32(active_1, f_x_1, v_zero);
+        float32x4_t f_y_safe_0 = vbslq_f32(active_0, f_y_0, v_zero);
+        float32x4_t f_y_safe_1 = vbslq_f32(active_1, f_y_1, v_zero);
+
+        acc_x_0 = vfmaq_f32(acc_x_0, f_x_safe_0, w_val_0);
+        acc_x_1 = vfmaq_f32(acc_x_1, f_x_safe_1, w_val_1);
+        acc_y_0 = vfmaq_f32(acc_y_0, f_y_safe_0, w_val_0);
+        acc_y_1 = vfmaq_f32(acc_y_1, f_y_safe_1, w_val_1);
+      }
+
+      
+      f_alpha_dt_x += vaddvq_f32(vaddq_f32(acc_x_0, acc_x_1));
+      f_alpha_dt_y += vaddvq_f32(vaddq_f32(acc_y_0, acc_y_1));
+
+      
+      for (int k = 0; k < board.n_obstacles; k++) {
+        float v1_x = pos_x[i] - board.obstacles[k].start_x;
+        float v1_y = pos_y[i] - board.obstacles[k].start_y;
+        float v2_x = board.obstacles[k].obstacle_v_x;
+        float v2_y = board.obstacles[k].obstacle_v_y;
+
+        float projection = (v1_x * v2_x + v1_y * v2_y) * board.obstacles[k].length_sq_inv;
+
+        if (projection < 0.0f)
+          projection = 0.0f;
+        else if (projection > 1.0f)
+          projection = 1.0f;
+
+        float r_aB_x = pos_x[i] - (board.obstacles[k].start_x + board.obstacles[k].obstacle_v_x * projection);
+        float r_aB_y = pos_y[i] - (board.obstacles[k].start_y + board.obstacles[k].obstacle_v_y * projection);
+
+        float r_aB_norm = sqrt(r_aB_x * r_aB_x + r_aB_y * r_aB_y);
+        r_aB_norm = r_aB_norm > EPS ? r_aB_norm : EPS;
+
+        float scalar_mult = (NEG_U0_AB_div_R * exp(r_aB_norm * NEG_R_INV)) / r_aB_norm;
+        float f_alpha_B_neg_x = r_aB_x * scalar_mult;
+        float f_alpha_B_neg_y = r_aB_y * scalar_mult;
+
+        float w_val = dt_neg;
+        float e_dot_f = (f_alpha_B_neg_y == 0) ? 0 : (f_alpha_B_neg_y < 0 ? -e_y[i] : e_y[i]);
+
+        if (e_dot_f < cos_phi)
+          w_val = c_neg_times_dt;
+
+        float f_alpha_B_dt_x = f_alpha_B_neg_x * w_val;
+        float f_alpha_B_dt_y = f_alpha_B_neg_y * w_val;
+
+        f_alpha_dt_x += f_alpha_B_dt_x;
+        f_alpha_dt_y += f_alpha_B_dt_y;
+      }
+
+      w_x[i] += f_alpha_dt_x;
+      w_y[i] += f_alpha_dt_y;
+    }
+
+    
+    for (int i = 0; i < n_p_pad; i += 4) {
+
+      float32x4_t w_x_i_vec           = vld1q_f32(&w_x[i]);
+      float32x4_t w_y_i_vec           = vld1q_f32(&w_y[i]);
+      float32x4_t v_max_i_vec         = vld1q_f32(&v_max[i]);
+      float32x4_t pos_x_i_vec         = vld1q_f32(&pos_x[i]);
+      float32x4_t pos_y_i_vec         = vld1q_f32(&pos_y[i]);
+      float32x4_t destination_y_i_vec = vld1q_f32(&destination_y[i]);
+      float32x4_t destination_x_i_vec = vld1q_f32(&destination_x[i]);
+      float32x4_t desired_v0_i_vec    = vld1q_f32(&desired_v0[i]);
+      int32x4_t   color_i_vec         = vld1q_s32(&color[i]);
+
+      
+      float32x4_t w_norm_sq  = vfmaq_f32(vmulq_f32(w_y_i_vec, w_y_i_vec), w_x_i_vec, w_x_i_vec);
+      float32x4_t w_norm_inv = _neon_rsqrt_ps(w_norm_sq);
+      float32x4_t correction_factor = vminq_f32(v_one, vmulq_f32(v_max_i_vec, w_norm_inv));
+
+      float32x4_t curr_vel_x_i_vec = vmulq_f32(w_x_i_vec, correction_factor);
+      float32x4_t curr_vel_y_i_vec = vmulq_f32(w_y_i_vec, correction_factor);
+
+      const float32x4_t dt_vec = vdupq_n_f32(dt);
+      pos_x_i_vec = vfmaq_f32(pos_x_i_vec, curr_vel_x_i_vec, dt_vec);
+      pos_y_i_vec = vfmaq_f32(pos_y_i_vec, curr_vel_y_i_vec, dt_vec);
+
+      
+      uint32x4_t cond_1_left  = vceqq_s32(color_i_vec, vdupq_n_s32(0));
+      uint32x4_t cond_1_right = vcgtq_f32(pos_x_i_vec, vdupq_n_f32(BOARD_LENGTH - 1.5f));
+      uint32x4_t cond_1       = vandq_u32(cond_1_left, cond_1_right);
+
+      
+      pos_x_i_vec         = vbslq_f32(cond_1, v_one,             pos_x_i_vec);
+      destination_y_i_vec = vbslq_f32(cond_1, pos_y_i_vec,       destination_y_i_vec);
+      w_x_i_vec           = vbslq_f32(cond_1, desired_v0_i_vec,  w_x_i_vec);
+      w_y_i_vec           = vbslq_f32(cond_1, v_zero,            w_y_i_vec);
+      curr_vel_x_i_vec    = vbslq_f32(cond_1, w_x_i_vec,         curr_vel_x_i_vec);
+      curr_vel_y_i_vec    = vbslq_f32(cond_1, w_y_i_vec,         curr_vel_y_i_vec);
+
+      
+      uint32x4_t cond_2_left  = vceqq_s32(color_i_vec, vdupq_n_s32(1));
+      uint32x4_t cond_2_right = vcltq_f32(pos_x_i_vec, vdupq_n_f32(1.5f));
+      uint32x4_t cond_2       = vandq_u32(cond_2_left, cond_2_right);
+
+      pos_x_i_vec         = vbslq_f32(cond_2, vdupq_n_f32(BOARD_LENGTH - 1.0f), pos_x_i_vec);
+      destination_y_i_vec = vbslq_f32(cond_2, pos_y_i_vec,                      destination_y_i_vec);
+
+      float32x4_t neg_desired_v0 = vnegq_f32(desired_v0_i_vec);
+      w_x_i_vec        = vbslq_f32(cond_2, neg_desired_v0, w_x_i_vec);
+      w_y_i_vec        = vbslq_f32(cond_2, v_zero,         w_y_i_vec);
+      curr_vel_x_i_vec = vbslq_f32(cond_2, w_x_i_vec,      curr_vel_x_i_vec);
+      curr_vel_y_i_vec = vbslq_f32(cond_2, w_y_i_vec,      curr_vel_y_i_vec);
+
+      
+      pos_y_i_vec = vmaxq_f32(pos_y_i_vec, vdupq_n_f32(WALL_EPS));
+      pos_y_i_vec = vminq_f32(pos_y_i_vec, vdupq_n_f32(BOARD_WIDTH_MINUS_WALL_EPS));
+
+      
+      float32x4_t ex_vec = vsubq_f32(destination_x_i_vec, pos_x_i_vec);
+      float32x4_t ey_vec = vsubq_f32(destination_y_i_vec, pos_y_i_vec);
+      float32x4_t e_norm_sq  = vfmaq_f32(vmulq_f32(ey_vec, ey_vec), ex_vec, ex_vec);
+      float32x4_t e_norm_inv = _neon_rsqrt_ps(e_norm_sq);
+      float32x4_t e_x_i_vec  = vmulq_f32(ex_vec, e_norm_inv);
+      float32x4_t e_y_i_vec  = vmulq_f32(ey_vec, e_norm_inv);
+
+      
+      float32x4_t vel_norm_sq = vfmaq_f32(vmulq_f32(curr_vel_y_i_vec, curr_vel_y_i_vec),
+                                          curr_vel_x_i_vec, curr_vel_x_i_vec);
+      float32x4_t sb_sq_i_vec = vmulq_f32(vel_norm_sq, vdupq_n_f32(delta_t_sq));
+      float32x4_t vel_norm_dt = vsqrtq_f32(sb_sq_i_vec);   // full-precision sqrt, matches AVX-512
+
+      vst1q_f32(&curr_vel_x[i],    curr_vel_x_i_vec);
+      vst1q_f32(&curr_vel_y[i],    curr_vel_y_i_vec);
+      vst1q_f32(&w_x[i],           w_x_i_vec);
+      vst1q_f32(&w_y[i],           w_y_i_vec);
+      vst1q_f32(&pos_x[i],         pos_x_i_vec);
+      vst1q_f32(&pos_y[i],         pos_y_i_vec);
+      vst1q_f32(&e_x[i],           e_x_i_vec);
+      vst1q_f32(&e_y[i],           e_y_i_vec);
+      vst1q_f32(&destination_y[i], destination_y_i_vec);
+      vst1q_f32(&sb_sq[i],         sb_sq_i_vec);
+      vst1q_f32(&sb_e_x[i],        vmulq_f32(e_x_i_vec, vel_norm_dt));
+      vst1q_f32(&sb_e_y[i],        vmulq_f32(e_y_i_vec, vel_norm_dt));
+    }
+
+#if !BENCHMARK
+    for (int i = 0; i < n_p; i++) {
+      if (dump_path && (t % dump_interval == 0)) {
+        samples.emplace_back(t + 1, i, pos_x[i], pos_y[i], e_x[i], e_y[i], color[i]);
+      }
+    }
+#endif
+  }
+
+#if !BENCHMARK
+  if (dump_path) {
+    ofstream out(dump_path);
+    out << "time,id,x,y,p.e.x,p.e.y,color\n";
+    for (auto &[t, id, x, y, e_x, e_y, c] : samples) {
+      out << t << ',' << id << ',' << x << ',' << y << ',' << e_x << ',' << e_y << ',' << c << '\n';
+    }
+  }
+#endif
+
+  return 0;
+}