x86 measurements

2026-05-30 21:26:19 +02:00
parent c16a67a1f8
commit 3aa84d6b8f
90 changed files with 3657 additions and 2561 deletions
@@ -0,0 +1,58 @@
+rm -f main_*  run_tests *.o *.s variants/*.s variants/*.o headers/*.s headers/*.o
+/usr/bin/g++ main.cpp ./variants/variant_inlining_7_SoA_AVX2_polyexp.cpp -o main_inlining_7_SoA_AVX2_polyexp -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): 9225415.00 cycles
+P(n): 3.98 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): 32320875.00 cycles
+P(n): 4.50 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): 122392560.00 cycles
+P(n): 4.73 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): 727457980.00 cycles
+P(n): 4.96 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): 2871839460.00 cycles
+P(n): 5.02 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): 11342093520.00 cycles
+P(n): 5.08 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): 70407692300.00 cycles
+P(n): 5.11 flops/cycle
+
@@ -0,0 +1,66 @@
+rm -f main_*  run_tests *.o *.s variants/*.s variants/*.o headers/*.s headers/*.o
+/usr/bin/g++ main.cpp ./variants/variant_inlining_7_SoA_AVX2_polyexp.cpp -o main_inlining_7_SoA_AVX2_polyexp -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): 9074662.50 cycles
+P(n): 4.05 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): 31994315.00 cycles
+P(n): 4.55 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): 120234740.00 cycles
+P(n): 4.82 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): 719504200.00 cycles
+P(n): 5.01 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): 2844257700.00 cycles
+P(n): 5.07 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): 11300324680.00 cycles
+P(n): 5.10 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): 70554568640.00 cycles
+P(n): 5.10 flops/cycle
+
+=== N_P=10000 N_ITER=200 ===
+W(n) per iteration: 2905772704.00 flops
+Q(n) per iteration: 1280112 bytes
+I(n): 2269.94 flops/byte
+W(n): 581154540800.00 flops
+T(n): 281416807980.00 cycles
+P(n): 2.07 flops/cycle
+