
mtl5 5.9.7 — build flags: native_fast_gemm, highway_simd
x86_64, 4 thread(s), baseline f64

== dot ===============================================================
dtype              n         s/op    GFLOP/s     vs f64
-------------------------------------------------------
f64             1000    2.360e-07      8.474 >     1.0x
f64            10000    1.555e-06     12.864 >     1.0x
f64           100000    1.720e-05     11.625       1.0x
f32             1000    1.303e-07     15.347 >     0.6x
f32            10000    8.016e-07     24.951 >     0.5x
f32           100000    6.997e-06     28.585       0.4x
cfloat32        1000    3.083e-04      0.006 >  1306.3x
cfloat32       10000    3.182e-03      0.006 >  2046.8x
cfloat32      100000    3.163e-02      0.006    1838.4x
posit32         1000    1.197e-03      0.002 >  5070.5x
posit32        10000    1.219e-02      0.002 >  7840.3x
posit32       100000    1.207e-01      0.002    7017.9x
lns32           1000    9.938e-04      0.002 >  4210.7x
lns32          10000    1.002e-02      0.002 >  6446.8x
lns32         100000    1.009e-01      0.002    5863.4x
takum32         1000    1.394e-04      0.014 >   590.4x
takum32        10000    1.379e-03      0.015 >   886.7x
takum32       100000    1.395e-02      0.014     810.6x
dd_cascade      1000    9.163e-05      0.022 >   388.2x
dd_cascade     10000    9.420e-04      0.021 >   605.9x
dd_cascade    100000    9.549e-03      0.021     555.1x
td_cascade      1000    1.488e-04      0.013 >   630.2x
td_cascade     10000    1.471e-03      0.014 >   946.4x
td_cascade    100000    1.478e-02      0.014     859.0x
qd_cascade      1000    2.103e-04      0.010 >   891.1x
qd_cascade     10000    2.048e-03      0.010 >  1317.4x
qd_cascade    100000    2.071e-02      0.010    1203.7x
  > baseline is within 20x of per-call overhead — ratio is a LOWER bound; prefer the larger sizes

== gemv ==============================================================
dtype              n         s/op    GFLOP/s     vs f64
-------------------------------------------------------
f64               64    3.614e-06      2.267       1.0x
f64              256    4.773e-05      2.746       1.0x
f64              512    2.149e-04      2.439       1.0x
f32               64    3.504e-06      2.338       1.0x
f32              256    4.455e-05      2.942       0.9x
f32              512    1.865e-04      2.811       0.9x
cfloat32          64    1.291e-03      0.006     357.3x
cfloat32         256    2.083e-02      0.006     436.5x
cfloat32         512    8.343e-02      0.006     388.1x
posit32           64    5.009e-03      0.002    1386.1x
posit32          256    7.994e-02      0.002    1674.9x
posit32          512    3.204e-01      0.002    1490.6x
lns32             64    4.153e-03      0.002    1149.1x
lns32            256    6.644e-02      0.002    1392.0x
lns32            512    2.659e-01      0.002    1237.0x
takum32           64    6.344e-04      0.013     175.6x
takum32          256    1.001e-02      0.013     209.8x
takum32          512    3.888e-02      0.013     180.9x
dd_cascade        64    3.489e-04      0.023      96.6x
dd_cascade       256    5.689e-03      0.023     119.2x
dd_cascade       512    2.235e-02      0.023     104.0x
td_cascade        64    6.682e-04      0.012     184.9x
td_cascade       256    1.040e-02      0.013     217.9x
td_cascade       512    4.106e-02      0.013     191.0x
qd_cascade        64    9.051e-04      0.009     250.4x
qd_cascade       256    1.415e-02      0.009     296.5x
qd_cascade       512    5.585e-02      0.009     259.8x

== gemm ==============================================================
dtype              n         s/op    GFLOP/s     vs f64
-------------------------------------------------------
f64               32    9.097e-06      7.204       1.0x
f64               64    4.030e-05     13.010       1.0x
f64              128    1.380e-04     30.396       1.0x
f32               32    6.209e-06     10.555       0.7x
f32               64    2.912e-05     18.004       0.7x
f32              128    1.152e-04     36.408       0.8x
cfloat32          32    8.794e-03      0.007     966.7x
cfloat32          64    7.195e-02      0.007    1785.4x
cfloat32         128    6.059e-01      0.007    4391.2x
posit32           32    3.886e-02      0.002    4271.4x
posit32           64    3.158e-01      0.002    7837.5x
posit32          128     too slow          -          -
lns32             32    3.319e-02      0.002    3648.3x
lns32             64    2.663e-01      0.002    6608.0x
lns32            128     too slow          -          -
takum32           32    5.131e-03      0.013     564.0x
takum32           64    4.087e-02      0.013    1014.1x
takum32          128    3.237e-01      0.013    2346.0x
dd_cascade        32    2.865e-03      0.023     315.0x
dd_cascade        64    2.342e-02      0.022     581.3x
dd_cascade       128    1.855e-01      0.023    1344.2x
td_cascade        32    5.561e-03      0.012     611.3x
td_cascade        64    4.381e-02      0.012    1087.2x
td_cascade       128    3.452e-01      0.012    2501.6x
qd_cascade        32    7.488e-03      0.009     823.1x
qd_cascade        64    5.864e-02      0.009    1455.0x
qd_cascade       128    4.614e-01      0.009    3343.5x

== lu ================================================================
dtype              n         s/op    GFLOP/s     vs f64
-------------------------------------------------------
f64               32    6.812e-06      3.207       1.0x
f64               64    4.083e-05      4.280       1.0x
f64              128    2.808e-04      4.979       1.0x
f32               32    6.806e-06      3.210       1.0x
f32               64    4.402e-05      3.970       1.1x
f32              128    2.778e-04      5.033       1.0x
cfloat32          32    3.686e-03      0.006     541.1x
cfloat32          64    2.541e-02      0.007     622.3x
cfloat32         128    1.889e-01      0.007     672.9x
posit32           32    1.464e-02      0.001    2148.6x
posit32           64    1.106e-01      0.002    2708.9x
posit32          128    8.543e-01      0.002    3042.3x
lns32             32    1.099e-02      0.002    1612.7x
lns32             64    8.953e-02      0.002    2192.8x
lns32            128    7.269e-01      0.002    2588.5x
takum32           32    1.575e-03      0.014     231.2x
takum32           64    1.263e-02      0.014     309.4x
takum32          128    1.007e-01      0.014     358.5x
dd_cascade        32    8.562e-04      0.026     125.7x
dd_cascade        64    6.653e-03      0.026     162.9x
dd_cascade       128    5.215e-02      0.027     185.7x
td_cascade        32    1.148e-02      0.002    1685.1x
td_cascade        64    9.482e-02      0.002    2322.3x
td_cascade       128    7.704e-01      0.002    2743.5x
qd_cascade        32    2.896e-02      0.001    4252.0x
qd_cascade        64    2.417e-01      0.001    5919.7x
qd_cascade       128    1.989e+00      0.001    7084.0x

== qr ================================================================
dtype              n         s/op    GFLOP/s     vs f64
-------------------------------------------------------
f64               32    2.574e-05      1.698       1.0x
f64               64    1.652e-04      2.116       1.0x
f64              128    1.453e-03      1.925       1.0x
f32               32    2.415e-05      1.809       0.9x
f32               64    1.631e-04      2.143       1.0x
f32              128    1.320e-03      2.118       0.9x
cfloat32          32    1.106e-02      0.004     429.7x
cfloat32          64    7.671e-02      0.005     464.3x
cfloat32         128    5.752e-01      0.005     395.9x
posit32           32    4.386e-02      0.001    1704.3x
posit32           64    3.150e-01      0.001    1906.7x
posit32          128     too slow          -          -
lns32             32    2.405e-02      0.002     934.3x
lns32             64    1.865e-01      0.002    1128.7x
lns32            128    1.465e+00      0.002    1008.4x
takum32           32    4.452e-03      0.010     173.0x
takum32           64    3.335e-02      0.010     201.9x
takum32          128    2.579e-01      0.011     177.5x
dd_cascade        32    2.922e-03      0.015     113.5x
dd_cascade        64    2.176e-02      0.016     131.7x
dd_cascade       128    1.658e-01      0.017     114.2x
td_cascade        32    3.754e-02      0.001    1458.8x
td_cascade        64    2.964e-01      0.001    1793.8x
td_cascade       128     too slow          -          -
qd_cascade        32    9.617e-02      0.000    3736.6x
qd_cascade        64    7.566e-01      0.000    4579.5x
qd_cascade       128     too slow          -          -
