benchmark_linear.py

gistfile1.txt

shape_x         shape_w             lin_ms     qlin_ms    trit_qlin_ms    qlin_speedup    trit_qlin_speedup    matmul_ms    trit_matmul_ms    trit_matmul_speedup
--------------  --------------  ----------  ----------  --------------  --------------  -------------------  -----------  ----------------  ---------------------
(512, 512)      (512, 512)         29.4059     474.733         293.763       0.061942             0.100101       112.424           138.208               0.813438
(512, 512)      (512, 2048)        28.5993     461.46          290.286       0.0619758            0.0985211      111.883           139.265               0.80338
(512, 512)      (512, 16384)       50.4579     781.971         292.295       0.0645266            0.172627       496.533           139.407               3.56174
(512, 2048)     (2048, 512)        28.4453     472.304         294.685       0.0602266            0.0965278      145.547           138.449               1.05127
(512, 2048)     (2048, 2048)       37.0158     476.439         293.468       0.0776926            0.126132       188.486           138.061               1.36524
(512, 2048)     (2048, 16384)     184.05      1402.46          500.208       0.131234             0.367947      1102.43            370.81                2.97302
(512, 16384)    (16384, 512)       61.5232    1579.62          342.221       0.0389481            0.179776       883.973           146.563               6.03134
(512, 16384)    (16384, 2048)     195.982     1893.12          494.01        0.103523             0.396716      1196.22            319.283               3.7466
(512, 16384)    (16384, 16384)   1143.18      7951.81         1432.92        0.143764             0.797799      7173.38           1242.9                 5.77147
(2048, 512)     (512, 512)         31.2884     470.816         293.73        0.0664557            0.106521       113.171           141.561               0.799446
(2048, 512)     (512, 2048)        29.5524     532.72          299.295       0.0554745            0.0987398      245.347           140.869               1.74167
(2048, 512)     (512, 16384)      174.753     2001.87         1125.73        0.0872949            0.155235      1569               982.964               1.5962
(2048, 2048)    (2048, 512)        34.2346     667.216         309.257       0.0513096            0.110699       161.156           140.487               1.14713
(2048, 2048)    (2048, 2048)       87.2535     961.908         380.485       0.0907087            0.229321       483.714           204.291               2.36777
(2048, 2048)    (2048, 16384)     600.917     3795.3          1347.78        0.158332             0.445858      3169.03           1114.66                2.84305
(2048, 16384)   (16384, 512)      170.546     2957.25          669.443       0.0576704            0.254758       899.164           250.17                3.59421
(2048, 16384)   (16384, 2048)     595.638     4872.2          1038.69        0.122253             0.573451      2808.46            626.956               4.47952
(2048, 16384)   (16384, 16384)   4317.46     20808            3669.44        0.207491             1.1766       18541.7            3244.42                5.71494
(16384, 512)    (512, 512)         50.374     1160.4           487.555       0.0434109            0.10332        417.783           281.45                1.4844
(16384, 512)    (512, 2048)       174.348     2356.11          755.647       0.0739983            0.230727      1499.56            489.848               3.06128
(16384, 512)    (512, 16384)     1493.83     13662.1          7514.93        0.109341             0.198782     11692.9            6687.46                1.74849
(16384, 2048)   (2048, 512)       179.109     2876.52          799.866       0.062266             0.223924       776.69            349.362               2.22316
(16384, 2048)   (2048, 2048)      634.543     5089.18         1604           0.124685             0.395601      2873.26           1060.66                2.70893
(16384, 2048)   (2048, 16384)    4647.16     26083.7          9129.69        0.178164             0.509016     22745.8            8055.5                 2.82363
(16384, 16384)  (16384, 512)     1187.73     18797.2          2939.39        0.0631866            0.404075      4168.51           1115.86                3.7357
(16384, 16384)  (16384, 2048)    4318.39     30570.9          5033.91        0.141258             0.85786      15824.7            3246.31                4.87469
(16384, 16384)  (16384, 16384)  36084.6     142042           24682.6         0.254042             1.46195     126176             22349.8                 5.64552