GPU: NVIDIA A40 L2 cache: 6.0 MB Quattra Tiled Benchmark (3 interleaved rounds per L; table shows median of the three per-round medians) L N NaiveMod(ms) NaiveCond(ms) Cond16x16(ms) SkewTiled(ms) Speedup(vs Best) Max Abs Diff Large ------------------------------------------------------------------------------------------------------------------ 16 256 0.005420 0.005420 0.005469 0.005395 1.005 2.76e-07 - 32 1024 0.003305 0.003311 0.003287 0.003266 1.006 4.17e-07 - 48 2304 0.003270 0.003274 0.003243 0.003235 1.002 5.36e-07 - 64 4096 0.003286 0.003262 0.003242 0.003221 1.007 5.66e-07 - 128 16384 0.003489 0.003271 0.003226 0.003242 0.995 6.56e-07 - 256 65536 0.005674 0.005278 0.004062 0.004037 1.006 5.96e-07 - 512 262144 0.014545 0.013988 0.008562 0.007558 1.133 7.15e-07 - 1024 1048576 0.053176 0.051966 0.027200 0.025282 1.076 8.34e-07 - 2048 4194304 0.214677 0.209526 0.098814 0.097934 1.009 9.39e-07 Y 4096 16777216 0.866671 0.817629 0.511694 0.466750 1.096 9.54e-07 Y 8192 67108864 4.498874 4.437103 2.300666 2.090187 1.101 9.83e-07 Y Raw per-round medians: L=16 runs(ms): NaiveMod=[0.005420, 0.005284, 0.008414] NaiveCond=[0.005496, 0.005420, 0.003324] Cond16x16=[0.005469, 0.005475, 0.005403] SkewTiled=[0.005395, 0.005494, 0.005284] best_control=NaiveCond L=32 runs(ms): NaiveMod=[0.003301, 0.003406, 0.003305] NaiveCond=[0.003314, 0.003308, 0.003311] Cond16x16=[0.003287, 0.003271, 0.003298] SkewTiled=[0.003266, 0.003338, 0.003260] best_control=Cond16x16 L=48 runs(ms): NaiveMod=[0.003307, 0.003270, 0.003263] NaiveCond=[0.003294, 0.003274, 0.003250] Cond16x16=[0.003274, 0.003243, 0.003236] SkewTiled=[0.003247, 0.003235, 0.003224] best_control=Cond16x16 L=64 runs(ms): NaiveMod=[0.003259, 0.003286, 0.003296] NaiveCond=[0.003262, 0.003251, 0.003279] Cond16x16=[0.003235, 0.003250, 0.003242] SkewTiled=[0.003203, 0.003238, 0.003221] best_control=Cond16x16 L=128 runs(ms): NaiveMod=[0.003482, 0.003489, 0.004558] NaiveCond=[0.003259, 0.003271, 0.004395] Cond16x16=[0.003226, 0.003221, 0.003239] SkewTiled=[0.003242, 0.003242, 0.003246] best_control=Cond16x16 L=256 runs(ms): NaiveMod=[0.005714, 0.005662, 0.005674] NaiveCond=[0.005278, 0.005272, 0.005349] Cond16x16=[0.004067, 0.004062, 0.004059] SkewTiled=[0.004038, 0.004029, 0.004037] best_control=Cond16x16 L=512 runs(ms): NaiveMod=[0.014584, 0.014545, 0.014542] NaiveCond=[0.013988, 0.013917, 0.014004] Cond16x16=[0.008564, 0.008562, 0.008554] SkewTiled=[0.007558, 0.007566, 0.007551] best_control=Cond16x16 L=1024 runs(ms): NaiveMod=[0.051507, 0.053176, 0.054345] NaiveCond=[0.051307, 0.051966, 0.052372] Cond16x16=[0.027200, 0.026943, 0.027254] SkewTiled=[0.025264, 0.025282, 0.025549] best_control=Cond16x16 L=2048 runs(ms): NaiveMod=[0.211138, 0.215005, 0.214677] NaiveCond=[0.209526, 0.208270, 0.210581] Cond16x16=[0.098814, 0.098089, 0.103856] SkewTiled=[0.097934, 0.097835, 0.098731] best_control=Cond16x16 L=4096 runs(ms): NaiveMod=[0.866687, 0.866576, 0.866671] NaiveCond=[0.806151, 0.817629, 0.817655] Cond16x16=[0.511510, 0.511694, 0.513382] SkewTiled=[0.466750, 0.466631, 0.466832] best_control=Cond16x16 L=8192 runs(ms): NaiveMod=[4.496256, 4.498874, 4.500846] NaiveCond=[4.437049, 4.437103, 4.441286] Cond16x16=[2.301416, 2.299871, 2.300666] SkewTiled=[2.090092, 2.090187, 2.091756] best_control=Cond16x16 Large-size set used: [2048, 4096, 8192] Pre-Registered Pass/Fail Evaluation: RESULT: PASS — 2/3 large-size wins meeting rules 1 and 3 (requires >= 2).