
mtl5 5.9.7 — build flags: native_fast_gemm, highway_simd
x86_64, 8 thread(s), baseline f64

== dot ===============================================================
dtype              n         s/op    GFLOP/s     vs f64
-------------------------------------------------------
f64             1000    2.352e-07      8.504 >     1.0x
f64            10000    1.551e-06     12.893 >     1.0x
f64           100000    1.715e-05     11.663       1.0x
f32             1000    1.302e-07     15.360 >     0.6x
f32            10000    8.044e-07     24.862 >     0.5x
f32           100000    7.385e-06     27.083       0.4x
cfloat32        1000    3.095e-04      0.006 >  1316.0x
cfloat32       10000    3.215e-03      0.006 >  2072.3x
cfloat32      100000    3.170e-02      0.006    1848.7x
posit32         1000    1.197e-03      0.002 >  5087.7x
posit32        10000    1.216e-02      0.002 >  7836.9x
posit32       100000    1.210e-01      0.002    7054.6x
lns32           1000    9.944e-04      0.002 >  4228.0x
lns32          10000    1.002e-02      0.002 >  6457.7x
lns32         100000    1.004e-01      0.002    5852.2x
takum32         1000    1.393e-04      0.014 >   592.4x
takum32        10000    1.380e-03      0.014 >   889.4x
takum32       100000    1.396e-02      0.014     814.3x
dd_cascade      1000    9.176e-05      0.022 >   390.2x
dd_cascade     10000    9.335e-04      0.021 >   601.8x
dd_cascade    100000    9.564e-03      0.021     557.7x
td_cascade      1000    1.488e-04      0.013 >   632.6x
td_cascade     10000    1.477e-03      0.014 >   952.3x
td_cascade    100000    1.478e-02      0.014     861.9x
qd_cascade      1000    2.141e-04      0.009 >   910.2x
qd_cascade     10000    2.082e-03      0.010 >  1342.0x
qd_cascade    100000    2.090e-02      0.010    1218.9x
  > 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    3.509e-06      2.335       1.0x
f64              256    4.765e-05      2.751       1.0x
f64              512    2.185e-04      2.400       1.0x
f32               64    3.479e-06      2.355       1.0x
f32              256    4.437e-05      2.954       0.9x
f32              512    1.983e-04      2.644       0.9x
cfloat32          64    1.299e-03      0.006     370.2x
cfloat32         256    2.082e-02      0.006     437.0x
cfloat32         512    8.318e-02      0.006     380.8x
posit32           64    4.980e-03      0.002    1419.4x
posit32          256    7.973e-02      0.002    1673.1x
posit32          512    3.193e-01      0.002    1461.7x
lns32             64    4.149e-03      0.002    1182.6x
lns32            256    6.649e-02      0.002    1395.2x
lns32            512    2.654e-01      0.002    1214.7x
takum32           64    6.361e-04      0.013     181.3x
takum32          256    1.003e-02      0.013     210.4x
takum32          512    3.902e-02      0.013     178.6x
dd_cascade        64    3.502e-04      0.023      99.8x
dd_cascade       256    5.640e-03      0.023     118.4x
dd_cascade       512    2.231e-02      0.024     102.1x
td_cascade        64    6.809e-04      0.012     194.0x
td_cascade       256    1.073e-02      0.012     225.2x
td_cascade       512    4.182e-02      0.013     191.4x
qd_cascade        64    9.049e-04      0.009     257.9x
qd_cascade       256    1.415e-02      0.009     296.9x
qd_cascade       512    5.601e-02      0.009     256.4x

== gemm ==============================================================
dtype              n         s/op    GFLOP/s     vs f64
-------------------------------------------------------
f64               32    8.930e-06      7.339       1.0x
f64               64    4.629e-05     11.326       1.0x
f64              128    1.389e-04     30.201       1.0x
f32               32    6.196e-06     10.577       0.7x
f32               64    2.892e-05     18.128       0.6x
f32              128    1.187e-04     35.342       0.9x
cfloat32          32    8.794e-03      0.007     984.7x
cfloat32          64    7.189e-02      0.007    1553.0x
cfloat32         128    6.031e-01      0.007    4342.6x
posit32           32    3.890e-02      0.002    4356.0x
posit32           64    3.141e-01      0.002    6786.4x
posit32          128     too slow          -          -
lns32             32    3.318e-02      0.002    3715.0x
lns32             64    2.670e-01      0.002    5768.2x
lns32            128     too slow          -          -
takum32           32    5.161e-03      0.013     578.0x
takum32           64    4.070e-02      0.013     879.1x
takum32          128    3.238e-01      0.013    2331.2x
dd_cascade        32    2.868e-03      0.023     321.2x
dd_cascade        64    2.320e-02      0.023     501.3x
dd_cascade       128    1.852e-01      0.023    1333.8x
td_cascade        32    5.744e-03      0.011     643.2x
td_cascade        64    4.539e-02      0.012     980.5x
td_cascade       128    3.550e-01      0.012    2556.5x
qd_cascade        32    7.381e-03      0.009     826.5x
qd_cascade        64    5.841e-02      0.009    1261.8x
qd_cascade       128    4.610e-01      0.009    3319.3x

== lu ================================================================
dtype              n         s/op    GFLOP/s     vs f64
-------------------------------------------------------
f64               32    6.860e-06      3.184       1.0x
f64               64    4.077e-05      4.286       1.0x
f64              128    2.819e-04      4.960       1.0x
f32               32    6.844e-06      3.192       1.0x
f32               64    4.384e-05      3.986       1.1x
f32              128    2.773e-04      5.042       1.0x
cfloat32          32    3.648e-03      0.006     531.8x
cfloat32          64    2.543e-02      0.007     623.6x
cfloat32         128    1.881e-01      0.007     667.2x
posit32           32    1.458e-02      0.001    2125.4x
posit32           64    1.106e-01      0.002    2712.3x
posit32          128    8.528e-01      0.002    3025.1x
lns32             32    1.094e-02      0.002    1595.4x
lns32             64    8.965e-02      0.002    2198.8x
lns32            128    7.275e-01      0.002    2580.8x
takum32           32    1.567e-03      0.014     228.4x
takum32           64    1.262e-02      0.014     309.5x
takum32          128    1.011e-01      0.014     358.7x
dd_cascade        32    8.553e-04      0.026     124.7x
dd_cascade        64    6.658e-03      0.026     163.3x
dd_cascade       128    5.204e-02      0.027     184.6x
td_cascade        32    1.153e-02      0.002    1680.2x
td_cascade        64    9.480e-02      0.002    2325.1x
td_cascade       128    7.713e-01      0.002    2736.1x
qd_cascade        32    2.898e-02      0.001    4225.1x
qd_cascade        64    2.416e-01      0.001    5925.7x
qd_cascade       128    1.993e+00      0.001    7071.0x

== qr ================================================================
dtype              n         s/op    GFLOP/s     vs f64
-------------------------------------------------------
f64               32    2.569e-05      1.701       1.0x
f64               64    1.655e-04      2.112       1.0x
f64              128    1.500e-03      1.864       1.0x
f32               32    2.428e-05      1.799       0.9x
f32               64    1.628e-04      2.146       1.0x
f32              128    1.360e-03      2.055       0.9x
cfloat32          32    1.106e-02      0.004     430.6x
cfloat32          64    7.676e-02      0.005     463.9x
cfloat32         128    5.748e-01      0.005     383.1x
posit32           32    4.391e-02      0.001    1709.3x
posit32           64    3.144e-01      0.001    1900.0x
posit32          128     too slow          -          -
lns32             32    2.402e-02      0.002     935.1x
lns32             64    1.859e-01      0.002    1123.6x
lns32            128    1.462e+00      0.002     974.2x
takum32           32    4.481e-03      0.010     174.4x
takum32           64    3.331e-02      0.010     201.3x
takum32          128    2.574e-01      0.011     171.6x
dd_cascade        32    2.919e-03      0.015     113.6x
dd_cascade        64    2.165e-02      0.016     130.8x
dd_cascade       128    1.659e-01      0.017     110.6x
td_cascade        32    3.763e-02      0.001    1464.7x
td_cascade        64    2.949e-01      0.001    1782.0x
td_cascade       128     too slow          -          -
qd_cascade        32    9.622e-02      0.000    3745.5x
qd_cascade        64    7.540e-01      0.000    4556.5x
qd_cascade       128     too slow          -          -
