tlinalg-competition on zen5-cpu

What was measured

Hardware

Every timed cell matches a successful verification census, including operation variants. Each cell is median of process medians ± sample SD of those medians; W is the median within-process sample SD. All values are milliseconds. These are descriptive spreads, not confidence intervals. The between-process spread includes the rotating-order schedule. Each process retains 5 chronological calls; 4 complete process sets are recorded. Spread guide (computed from this run's own rows): per-engine between-start range median 1.29%, worst 33.28% among cells of 1 us or more (a sub-microsecond row is timer and scheduler jitter, not kernel time); 40 of 396 engine cells exceed 5% and carry !; median within-process relative SD 0.33%. A cell whose ±, ! or p is as large as the difference you are reading is not evidence of a regression. The vs tlinalg-pool column is the rotation-averaged ratio exp(mean over starts of log(engine median / tlinalg-pool median)), with the per-start ratios beside it: R > 1 means the engine was slower than that first-declared row in that start. It removes a row-position effect only under one position effect shared by the engines and stable across starts, and only where every declared engine has a measured row (* marks a cell where one has no row, so no ratio there is position-balanced). ! marks a cell where some engine's own per-start median range exceeds 5% of its median, and p a balanced cell whose estimated position-effect ratio exceeds 1.05. Flagged cells are retained, not discarded; the between-start variation can exceed the difference being compared. The vs columns mark only differences larger than the spread this run recorded for those two rows (combined between-start SD, floored at 5%), always with the measured library as the subject: 🔵 <row> X faster than <row> / 🔴 <row> X slower than <row> (the subject is the first column of the table, the compared row is named in the cell, not only in the column header), · inside the band, - that row is not measured in this arm. notes carries the cell warnings (* a declared row has no measurement, ! between-start range over 5%, p position effect over 5%, dispatch-dominated means the row is under 20 us so 4T/8T pay thread-region start-up). Per-start ratios and every chronological sample are in the run directory named at the top of this page (run*.csv). These describe the recorded revisions; they are not a regression claim.

batch-small, f64, 8T (CPU 4-11)

case tlinalg-pool (ms) lapack-mkl (ms) vs lapack-mkl notes
cholesky/n16/b1024 0.119366 ± 0.000695398 (W 0.00243473) 1.06652 ± 0.0012858 (W 0.00236541) 🔵 tlinalg-pool 8.93× faster than lapack-mkl
cholesky/n32/b1024 0.478287 ± 0.00162469 (W 0.00805826) 3.80665 ± 0.0286122 (W 0.0272017) 🔵 tlinalg-pool 7.96× faster than lapack-mkl
cholesky/n4/b1024 0.180484 ± 0.000492269 (W 0.00118859) 0.122973 ± 0.00348857 (W 0.000154185) 🔴 tlinalg-pool 1.47× slower than lapack-mkl ! between-start range over 5%
cholesky/n8/b1024 0.049577 ± 0.000192094 (W 0.00013725) 0.324446 ± 0.00294615 (W 0.00208806) 🔵 tlinalg-pool 6.54× faster than lapack-mkl
eig/n16/b1024 11.9975 ± 0.0650107 (W 0.0168664) 52.9614 ± 0.224429 (W 0.0880673) 🔵 tlinalg-pool 4.41× faster than lapack-mkl
eig/n32/b1024 54.104 ± 0.303204 (W 0.053995) 284.458 ± 0.569281 (W 0.394215) 🔵 tlinalg-pool 5.26× faster than lapack-mkl
eig/n4/b1024 0.809944 ± 0.00616379 (W 0.00230116) 2.86368 ± 0.00988159 (W 0.00386905) 🔵 tlinalg-pool 3.54× faster than lapack-mkl
eig/n8/b1024 3.01189 ± 0.0171852 (W 0.00313188) 11.8942 ± 0.0941696 (W 0.0771669) 🔵 tlinalg-pool 3.95× faster than lapack-mkl
eigh/n16/b1024 3.76006 ± 0.032309 (W 0.00302925) 23.7982 ± 0.093972 (W 0.0233712) 🔵 tlinalg-pool 6.33× faster than lapack-mkl
eigh/n32/b1024 14.0282 ± 0.379079 (W 0.0152491) 98.2109 ± 0.30753 (W 0.0830395) 🔵 tlinalg-pool 7.00× faster than lapack-mkl ! between-start range over 5%
eigh/n4/b1024 0.327388 ± 0.00378897 (W 0.00181745) 2.97912 ± 0.00642394 (W 0.00693686) 🔵 tlinalg-pool 9.10× faster than lapack-mkl
eigh/n8/b1024 1.09479 ± 0.00690597 (W 0.00213734) 7.90817 ± 0.0439254 (W 0.0251364) 🔵 tlinalg-pool 7.22× faster than lapack-mkl
eigvals/n16/b1024 8.31216 ± 0.0279002 (W 0.00776498) 36.6592 ± 0.0651843 (W 0.0579515) 🔵 tlinalg-pool 4.41× faster than lapack-mkl
eigvals/n32/b1024 36.5899 ± 1.00648 (W 0.14705) 172.208 ± 0.241021 (W 0.125402) 🔵 tlinalg-pool 4.71× faster than lapack-mkl ! between-start range over 5%
eigvals/n4/b1024 0.435388 ± 0.00193674 (W 0.00371657) 2.11081 ± 0.00254759 (W 0.00253394) 🔵 tlinalg-pool 4.85× faster than lapack-mkl
eigvals/n8/b1024 1.98517 ± 0.0229189 (W 0.00450907) 9.00817 ± 0.0458474 (W 0.0326745) 🔵 tlinalg-pool 4.54× faster than lapack-mkl
eigvalsh/n16/b1024 2.83436 ± 0.0160631 (W 0.00378289) 12.8109 ± 0.0534213 (W 0.0350109) 🔵 tlinalg-pool 4.52× faster than lapack-mkl
eigvalsh/n32/b1024 10.6891 ± 0.067658 (W 0.00980008) 41.7591 ± 0.227219 (W 0.0356976) 🔵 tlinalg-pool 3.91× faster than lapack-mkl
eigvalsh/n4/b1024 0.258344 ± 0.00341627 (W 0.00177503) 1.56355 ± 0.00466164 (W 0.00398407) 🔵 tlinalg-pool 6.05× faster than lapack-mkl
eigvalsh/n8/b1024 0.867673 ± 0.00232566 (W 0.00165564) 4.47864 ± 0.0183198 (W 0.00901993) 🔵 tlinalg-pool 5.16× faster than lapack-mkl
full_piv_lu/n16/b1024 0.511474 ± 0.00531828 (W 0.00410327) 3.90206 ± 0.0198445 (W 0.0150677) 🔵 tlinalg-pool 7.63× faster than lapack-mkl
full_piv_lu/n32/b1024 2.04786 ± 0.0164883 (W 0.0125611) 26.4313 ± 0.0914399 (W 0.0342001) 🔵 tlinalg-pool 12.91× faster than lapack-mkl
full_piv_lu/n4/b1024 0.499476 ± 0.00463083 (W 0.00303123) 0.410552 ± 0.00221707 (W 0.00366255) 🔴 tlinalg-pool 1.22× slower than lapack-mkl
full_piv_lu/n8/b1024 0.18033 ± 0.013566 (W 0.00215613) 0.990912 ± 0.00673613 (W 0.00362077) 🔵 tlinalg-pool 5.50× faster than lapack-mkl ! between-start range over 5%
full_piv_lu_solve/n16/b1024 0.494542 ± 0.00640566 (W 0.00229787) 5.7094 ± 0.0498696 (W 0.0186833) 🔵 tlinalg-pool 11.54× faster than lapack-mkl
full_piv_lu_solve/n32/b1024 1.64316 ± 0.0583896 (W 0.011049) 29.4225 ± 0.0987485 (W 0.0256403) 🔵 tlinalg-pool 17.91× faster than lapack-mkl ! between-start range over 5%
full_piv_lu_solve/n4/b1024 0.083129 ± 0.00109193 (W 0.00233362) 1.00305 ± 0.010374 (W 0.00246386) 🔵 tlinalg-pool 12.07× faster than lapack-mkl
full_piv_lu_solve/n8/b1024 0.196053 ± 0.00181085 (W 0.00202102) 1.84464 ± 0.00846074 (W 0.00556875) 🔵 tlinalg-pool 9.41× faster than lapack-mkl
householder_apply/n16/b1024 0.593711 ± 0.0276941 (W 0.00316985) 2.66365 ± 0.0121582 (W 0.0173898) 🔵 tlinalg-pool 4.49× faster than lapack-mkl ! between-start range over 5%
householder_apply/n32/b1024 2.38334 ± 0.103052 (W 0.00841008) 13.1578 ± 0.069558 (W 0.0297086) 🔵 tlinalg-pool 5.52× faster than lapack-mkl ! between-start range over 5%
householder_apply/n4/b1024 0.053154 ± 0.00214224 (W 9.14955e-05) 0.324001 ± 0.00936724 (W 0.00175717) 🔵 tlinalg-pool 6.10× faster than lapack-mkl ! between-start range over 5%
householder_apply/n8/b1024 0.162937 ± 0.00547241 (W 0.00139798) 0.790552 ± 0.0236046 (W 0.00727305) 🔵 tlinalg-pool 4.85× faster than lapack-mkl ! between-start range over 5%
householder_factor/n16/b1024 0.509431 ± 0.0100065 (W 0.00160179) 2.97236 ± 0.0148809 (W 0.00399112) 🔵 tlinalg-pool 5.83× faster than lapack-mkl
householder_factor/n32/b1024 1.6921 ± 0.0399759 (W 0.00448705) 11.9834 ± 0.0576054 (W 0.0169103) 🔵 tlinalg-pool 7.08× faster than lapack-mkl
householder_factor/n4/b1024 0.0641495 ± 0.000420595 (W 0.000219177) 0.338652 ± 0.000353067 (W 0.00226803) 🔵 tlinalg-pool 5.28× faster than lapack-mkl
householder_factor/n8/b1024 0.168623 ± 0.00276937 (W 0.00184911) 0.870051 ± 0.00435685 (W 0.00247303) 🔵 tlinalg-pool 5.16× faster than lapack-mkl
lu/n16/b1024 0.37944 ± 0.00892459 (W 0.00265691) 1.77274 ± 0.0679692 (W 0.00614462) 🔵 tlinalg-pool 4.67× faster than lapack-mkl ! between-start range over 5%
lu/n32/b1024 1.30948 ± 0.0124985 (W 0.0090838) 12.8866 ± 0.0955553 (W 0.0535561) 🔵 tlinalg-pool 9.84× faster than lapack-mkl
lu/n4/b1024 0.320168 ± 0.00710418 (W 0.00198132) 0.225382 ± 0.00133864 (W 0.00299944) 🔴 tlinalg-pool 1.42× slower than lapack-mkl
lu/n8/b1024 0.135136 ± 0.00337252 (W 0.00164006) 0.575311 ± 0.0270779 (W 0.00310002) 🔵 tlinalg-pool 4.26× faster than lapack-mkl ! between-start range over 5%
lu_factor/n16/b1024 0.257242 ± 0.00316305 (W 0.00207202) 1.28429 ± 0.0250401 (W 0.00398581) 🔵 tlinalg-pool 4.99× faster than lapack-mkl
lu_factor/n32/b1024 0.72575 ± 0.0149605 (W 0.00483698) 10.1861 ± 0.0506232 (W 0.020593) 🔵 tlinalg-pool 14.04× faster than lapack-mkl
lu_factor/n4/b1024 0.241818 ± 0.00105785 (W 0.0030186) 0.177428 ± 0.00477551 (W 0.00251216) 🔴 tlinalg-pool 1.36× slower than lapack-mkl ! between-start range over 5%
lu_factor/n8/b1024 0.097922 ± 0.00251031 (W 0.00152307) 0.360488 ± 0.0063067 (W 0.00270498) 🔵 tlinalg-pool 3.68× faster than lapack-mkl ! between-start range over 5%
lu_factor_solve/n16/b1024 0.330608 ± 0.00592306 (W 0.00184971) 2.34275 ± 0.0272617 (W 0.00825972) 🔵 tlinalg-pool 7.09× faster than lapack-mkl
lu_factor_solve/n32/b1024 0.895887 ± 0.0147266 (W 0.00238273) 14.9719 ± 0.0330008 (W 0.0269733) 🔵 tlinalg-pool 16.71× faster than lapack-mkl
lu_factor_solve/n4/b1024 0.058208 ± 0.000583417 (W 0.00078612) 0.273182 ± 0.00255291 (W 0.00198573) 🔵 tlinalg-pool 4.69× faster than lapack-mkl
lu_factor_solve/n8/b1024 0.122723 ± 0.0033357 (W 0.00188726) 0.907332 ± 0.00297284 (W 0.00286306) 🔵 tlinalg-pool 7.39× faster than lapack-mkl ! between-start range over 5%
lu_solve_prepared/n16/b1024 0.098523 ± 0.000294766 (W 0.00151826) 1.06835 ± 0.00260445 (W 0.00337279) 🔵 tlinalg-pool 10.84× faster than lapack-mkl
lu_solve_prepared/n32/b1024 0.230759 ± 0.00309364 (W 0.00263335) 4.54385 ± 0.0464633 (W 0.0132085) 🔵 tlinalg-pool 19.69× faster than lapack-mkl
lu_solve_prepared/n4/b1024 0.0210085 ± 8.12214e-05 (W 7.75035e-05) 0.133111 ± 0.00170169 (W 0.000771642) 🔵 tlinalg-pool 6.34× faster than lapack-mkl
lu_solve_prepared/n8/b1024 0.0445475 ± 0.000460545 (W 0.000795233) 0.604369 ± 0.00179725 (W 0.00278207) 🔵 tlinalg-pool 13.57× faster than lapack-mkl
qr/n16/b1024 1.14216 ± 0.0219224 (W 0.00267594) 5.2779 ± 0.0317296 (W 0.0111846) 🔵 tlinalg-pool 4.62× faster than lapack-mkl
qr/n32/b1024 4.32393 ± 0.0289844 (W 0.00824499) 20.7514 ± 0.248037 (W 0.0530126) 🔵 tlinalg-pool 4.80× faster than lapack-mkl
qr/n4/b1024 0.100737 ± 0.00181879 (W 0.00191474) 0.703844 ± 0.00255676 (W 0.0034727) 🔵 tlinalg-pool 6.99× faster than lapack-mkl
qr/n8/b1024 0.31785 ± 0.00380356 (W 0.00159441) 1.67646 ± 0.0131834 (W 0.00626323) 🔵 tlinalg-pool 5.27× faster than lapack-mkl
rank_revealing_qr/n16/b1024 1.35124 ± 0.0236431 (W 0.00277756) 8.91119 ± 0.0216359 (W 0.0197253) 🔵 tlinalg-pool 6.59× faster than lapack-mkl
rank_revealing_qr/n32/b1024 4.44424 ± 0.0372938 (W 0.0110531) 30.7919 ± 0.136283 (W 0.0453715) 🔵 tlinalg-pool 6.93× faster than lapack-mkl
rank_revealing_qr/n4/b1024 0.16989 ± 0.00226019 (W 0.00202227) 1.4229 ± 0.00989097 (W 0.00907752) 🔵 tlinalg-pool 8.38× faster than lapack-mkl
rank_revealing_qr/n8/b1024 0.434831 ± 0.00572977 (W 0.00233604) 3.35526 ± 0.0105529 (W 0.0085574) 🔵 tlinalg-pool 7.72× faster than lapack-mkl
solve/n16/b1024 0.349388 ± 0.0107417 (W 0.00267286) 2.54989 ± 0.0221538 (W 0.00753675) 🔵 tlinalg-pool 7.30× faster than lapack-mkl ! between-start range over 5%
solve/n32/b1024 0.966313 ± 0.0215521 (W 0.00396535) 14.72 ± 0.057166 (W 0.0343269) 🔵 tlinalg-pool 15.23× faster than lapack-mkl ! between-start range over 5%
solve/n4/b1024 0.054035 ± 0.00111283 (W 0.000405406) 0.340878 ± 0.00199523 (W 0.00251761) 🔵 tlinalg-pool 6.31× faster than lapack-mkl
solve/n8/b1024 0.122893 ± 0.00093204 (W 0.00149233) 1.01487 ± 0.0162562 (W 0.0052772) 🔵 tlinalg-pool 8.26× faster than lapack-mkl
svd_full/n16/b1024 6.87036 ± 0.443727 (W 0.00675432) 52.4679 ± 0.0902141 (W 0.0785504) 🔵 tlinalg-pool 7.64× faster than lapack-mkl ! between-start range over 5%
svd_full/n32/b1024 24.1871 ± 0.967404 (W 0.0183637) 191.723 ± 1.12906 (W 0.303226) 🔵 tlinalg-pool 7.93× faster than lapack-mkl ! between-start range over 5%
svd_full/n4/b1024 0.572588 ± 0.0215205 (W 0.00322539) 6.3225 ± 0.028221 (W 0.0144448) 🔵 tlinalg-pool 11.04× faster than lapack-mkl ! between-start range over 5%
svd_full/n8/b1024 1.88997 ± 0.0566594 (W 0.0022933) 19.099 ± 0.0623562 (W 0.0204268) 🔵 tlinalg-pool 10.11× faster than lapack-mkl ! between-start range over 5%
svd_thin/n16/b1024 6.89804 ± 0.622163 (W 0.00457178) 52.6191 ± 0.114144 (W 0.0510928) 🔵 tlinalg-pool 7.63× faster than lapack-mkl ! between-start range over 5%
svd_thin/n32/b1024 24.2551 ± 0.975725 (W 0.0279806) 191.814 ± 0.93527 (W 0.227386) 🔵 tlinalg-pool 7.91× faster than lapack-mkl ! between-start range over 5%
svd_thin/n4/b1024 0.571729 ± 0.0222319 (W 0.00214241) 6.33864 ± 0.0120435 (W 0.0086895) 🔵 tlinalg-pool 11.09× faster than lapack-mkl ! between-start range over 5%
svd_thin/n8/b1024 1.88562 ± 0.0137377 (W 0.00266192) 19.1577 ± 0.0832235 (W 0.0351668) 🔵 tlinalg-pool 10.16× faster than lapack-mkl
svd_values/n16/b1024 2.81557 ± 0.0121674 (W 0.00403825) 19.5588 ± 0.0986441 (W 0.0113506) 🔵 tlinalg-pool 6.95× faster than lapack-mkl
svd_values/n32/b1024 9.31691 ± 0.0452969 (W 0.00866318) 63.595 ± 0.216324 (W 0.0432841) 🔵 tlinalg-pool 6.83× faster than lapack-mkl
svd_values/n4/b1024 0.36844 ± 0.00110696 (W 0.00243827) 2.90059 ± 0.0112345 (W 0.00432119) 🔵 tlinalg-pool 7.87× faster than lapack-mkl
svd_values/n8/b1024 1.00315 ± 0.00488521 (W 0.00178261) 7.0708 ± 0.0217544 (W 0.00704129) 🔵 tlinalg-pool 7.05× faster than lapack-mkl
triangular_solve/n16/b1024 0.069133 ± 0.0012737 (W 0.00163497) 0.550728 ± 0.00387386 (W 0.00385602) 🔵 tlinalg-pool 7.97× faster than lapack-mkl
triangular_solve/n32/b1024 0.155453 ± 0.000959696 (W 0.00208212) 2.4389 ± 0.021163 (W 0.00768739) 🔵 tlinalg-pool 15.69× faster than lapack-mkl
triangular_solve/n4/b1024 0.0142065 ± 0.000211692 (W 9.52228e-05) 0.273933 ± 0.00152698 (W 0.00208705) 🔵 tlinalg-pool 19.28× faster than lapack-mkl dispatch-dominated (row under 20 us)
triangular_solve/n8/b1024 0.030878 ± 0.000124125 (W 8.37215e-05) 0.353351 ± 0.000878036 (W 0.0031586) 🔵 tlinalg-pool 11.44× faster than lapack-mkl

batch-small, c64, 8T (CPU 4-11)

case tlinalg-pool (ms) lapack-mkl (ms) vs lapack-mkl notes
cholesky/n16/b1024 0.167972 ± 0.000867682 (W 0.00171297) 1.82851 ± 0.0175074 (W 0.0064643) 🔵 tlinalg-pool 10.89× faster than lapack-mkl
cholesky/n32/b1024 0.813679 ± 0.0047973 (W 0.0101863) 8.80461 ± 0.0223368 (W 0.0221494) 🔵 tlinalg-pool 10.82× faster than lapack-mkl
cholesky/n4/b1024 0.027947 ± 0.000108128 (W 0.000104704) 0.176668 ± 0.00104121 (W 0.00122062) 🔵 tlinalg-pool 6.32× faster than lapack-mkl
cholesky/n8/b1024 0.056149 ± 0.000176448 (W 0.000166025) 0.485301 ± 0.00585491 (W 0.00300216) 🔵 tlinalg-pool 8.64× faster than lapack-mkl
eig/n16/b1024 16.4298 ± 0.0668662 (W 0.0202465) 178.909 ± 0.637622 (W 0.325128) 🔵 tlinalg-pool 10.89× faster than lapack-mkl
eig/n32/b1024 67.8524 ± 0.215202 (W 0.0388257) 829.526 ± 2.70966 (W 0.706293) 🔵 tlinalg-pool 12.23× faster than lapack-mkl
eig/n4/b1024 1.35031 ± 0.0218482 (W 0.00276297) 12.2118 ± 0.0202064 (W 0.012747) 🔵 tlinalg-pool 9.04× faster than lapack-mkl
eig/n8/b1024 4.53662 ± 0.268479 (W 0.00795708) 42.6675 ± 0.0423436 (W 0.0307788) 🔵 tlinalg-pool 9.41× faster than lapack-mkl ! between-start range over 5%
eigh/n16/b1024 4.01873 ± 0.0226977 (W 0.00570928) 32.4064 ± 0.0725361 (W 0.0375604) 🔵 tlinalg-pool 8.06× faster than lapack-mkl
eigh/n32/b1024 17.5467 ± 0.0314309 (W 0.0243668) 148.67 ± 0.41902 (W 0.114779) 🔵 tlinalg-pool 8.47× faster than lapack-mkl
eigh/n4/b1024 0.386557 ± 0.0036253 (W 0.00220016) 2.90388 ± 0.00319824 (W 0.00594525) 🔵 tlinalg-pool 7.51× faster than lapack-mkl
eigh/n8/b1024 1.17594 ± 0.0076735 (W 0.00600897) 8.97761 ± 0.0266706 (W 0.0145838) 🔵 tlinalg-pool 7.63× faster than lapack-mkl
eigvals/n16/b1024 9.88176 ± 0.0476928 (W 0.00998751) 125.635 ± 0.715946 (W 0.0859134) 🔵 tlinalg-pool 12.71× faster than lapack-mkl
eigvals/n32/b1024 43.0824 ± 0.157906 (W 0.203105) 545.63 ± 1.86374 (W 0.737776) 🔵 tlinalg-pool 12.66× faster than lapack-mkl
eigvals/n4/b1024 0.656215 ± 0.00295758 (W 0.00214927) 9.4867 ± 0.0139339 (W 0.00805664) 🔵 tlinalg-pool 14.46× faster than lapack-mkl
eigvals/n8/b1024 2.55768 ± 0.0330171 (W 0.00562527) 33.8712 ± 0.0120229 (W 0.0339689) 🔵 tlinalg-pool 13.24× faster than lapack-mkl
eigvalsh/n16/b1024 3.02979 ± 0.0122945 (W 0.00456902) 17.4238 ± 0.0715822 (W 0.0275924) 🔵 tlinalg-pool 5.75× faster than lapack-mkl
eigvalsh/n32/b1024 11.5511 ± 0.0753726 (W 0.00957483) 72.1641 ± 0.143205 (W 0.0680414) 🔵 tlinalg-pool 6.25× faster than lapack-mkl
eigvalsh/n4/b1024 0.313126 ± 0.000419917 (W 0.00435368) 1.75258 ± 0.00404464 (W 0.002383) 🔵 tlinalg-pool 5.60× faster than lapack-mkl
eigvalsh/n8/b1024 0.950023 ± 0.00341168 (W 0.001126) 5.2963 ± 0.0227017 (W 0.00669682) 🔵 tlinalg-pool 5.57× faster than lapack-mkl
full_piv_lu/n16/b1024 0.663536 ± 0.00663237 (W 0.00719773) 5.52821 ± 0.0541605 (W 0.0318754) 🔵 tlinalg-pool 8.33× faster than lapack-mkl
full_piv_lu/n32/b1024 3.05292 ± 0.0162968 (W 0.0237323) 27.4138 ± 0.0478198 (W 0.0312138) 🔵 tlinalg-pool 8.98× faster than lapack-mkl
full_piv_lu/n4/b1024 0.089807 ± 0.00140417 (W 0.00195054) 0.524309 ± 0.00251695 (W 0.0026538) 🔵 tlinalg-pool 5.84× faster than lapack-mkl
full_piv_lu/n8/b1024 0.215661 ± 0.00358962 (W 0.0036334) 1.2991 ± 0.0173998 (W 0.00431858) 🔵 tlinalg-pool 6.02× faster than lapack-mkl
full_piv_lu_solve/n16/b1024 0.772098 ± 0.0760571 (W 0.00592058) 8.6789 ± 0.057581 (W 0.0147617) 🔵 tlinalg-pool 11.24× faster than lapack-mkl ! between-start range over 5%
full_piv_lu_solve/n32/b1024 2.78859 ± 0.162026 (W 0.00810435) 35.7819 ± 0.127571 (W 0.0277378) 🔵 tlinalg-pool 12.83× faster than lapack-mkl ! between-start range over 5%
full_piv_lu_solve/n4/b1024 0.124721 ± 0.00158905 (W 0.000346273) 1.46958 ± 0.00514433 (W 0.0057464) 🔵 tlinalg-pool 11.78× faster than lapack-mkl
full_piv_lu_solve/n8/b1024 0.290895 ± 0.0238717 (W 0.00237389) 3.08634 ± 0.019169 (W 0.00460287) 🔵 tlinalg-pool 10.61× faster than lapack-mkl ! between-start range over 5%
householder_apply/n16/b1024 0.628286 ± 0.0104661 (W 0.0037559) 6.42107 ± 0.0383028 (W 0.0196395) 🔵 tlinalg-pool 10.22× faster than lapack-mkl
householder_apply/n32/b1024 3.41821 ± 0.0256013 (W 0.014677) 38.2476 ± 0.0993242 (W 0.0717989) 🔵 tlinalg-pool 11.19× faster than lapack-mkl
householder_apply/n4/b1024 0.053284 ± 0.000899909 (W 0.000173094) 0.474705 ± 0.00351651 (W 0.00408517) 🔵 tlinalg-pool 8.91× faster than lapack-mkl
householder_apply/n8/b1024 0.152568 ± 0.00222402 (W 0.00268478) 1.4457 ± 0.0154141 (W 0.0243934) 🔵 tlinalg-pool 9.48× faster than lapack-mkl
householder_factor/n16/b1024 0.606064 ± 0.00475241 (W 0.00303154) 6.3175 ± 0.0307895 (W 0.017034) 🔵 tlinalg-pool 10.42× faster than lapack-mkl
householder_factor/n32/b1024 2.55174 ± 0.0266503 (W 0.00995623) 26.8795 ± 0.252602 (W 0.0471571) 🔵 tlinalg-pool 10.53× faster than lapack-mkl
householder_factor/n4/b1024 0.078856 ± 0.00034618 (W 0.000386736) 0.53092 ± 0.00113349 (W 0.00213582) 🔵 tlinalg-pool 6.73× faster than lapack-mkl
householder_factor/n8/b1024 0.201734 ± 0.00152752 (W 0.0019306) 1.498 ± 0.00350066 (W 0.0029168) 🔵 tlinalg-pool 7.43× faster than lapack-mkl
lu/n16/b1024 0.451277 ± 0.00830773 (W 0.010145) 3.55922 ± 0.0382891 (W 0.0213834) 🔵 tlinalg-pool 7.89× faster than lapack-mkl
lu/n32/b1024 2.15567 ± 0.0042777 (W 0.0120985) 19.4545 ± 0.0975885 (W 0.0304779) 🔵 tlinalg-pool 9.02× faster than lapack-mkl
lu/n4/b1024 0.065216 ± 0.000203446 (W 0.000447652) 0.29643 ± 0.00774873 (W 0.00223171) 🔵 tlinalg-pool 4.55× faster than lapack-mkl ! between-start range over 5%
lu/n8/b1024 0.146483 ± 0.00150834 (W 0.00386085) 0.893957 ± 0.014012 (W 0.00337117) 🔵 tlinalg-pool 6.10× faster than lapack-mkl
lu_factor/n16/b1024 0.29932 ± 0.0453424 (W 0.00960387) 2.40883 ± 0.0296736 (W 0.00892815) 🔵 tlinalg-pool 8.05× faster than lapack-mkl ! between-start range over 5%; p position effect over 5%
lu_factor/n32/b1024 1.23868 ± 0.00441619 (W 0.00897191) 15.8422 ± 0.108862 (W 0.06842) 🔵 tlinalg-pool 12.79× faster than lapack-mkl
lu_factor/n4/b1024 0.0621 ± 0.00138809 (W 0.00104195) 0.202215 ± 0.00329816 (W 0.00164566) 🔵 tlinalg-pool 3.26× faster than lapack-mkl
lu_factor/n8/b1024 0.110675 ± 0.00137764 (W 0.00144311) 0.598351 ± 0.00684294 (W 0.00284748) 🔵 tlinalg-pool 5.41× faster than lapack-mkl
lu_factor_solve/n16/b1024 0.457174 ± 0.00712969 (W 0.00320492) 5.0818 ± 0.0248943 (W 0.00572779) 🔵 tlinalg-pool 11.12× faster than lapack-mkl
lu_factor_solve/n32/b1024 1.71004 ± 0.00855749 (W 0.00773808) 21.3466 ± 0.0667313 (W 0.0328272) 🔵 tlinalg-pool 12.48× faster than lapack-mkl
lu_factor_solve/n4/b1024 0.073722 ± 0.000702314 (W 0.000763728) 0.5202 ± 0.00481519 (W 0.00316847) 🔵 tlinalg-pool 7.06× faster than lapack-mkl
lu_factor_solve/n8/b1024 0.163483 ± 0.000518144 (W 0.00184367) 1.53273 ± 0.00815629 (W 0.00278981) 🔵 tlinalg-pool 9.38× faster than lapack-mkl
lu_solve_prepared/n16/b1024 0.185109 ± 0.00573822 (W 0.00292238) 2.57222 ± 0.00415097 (W 0.00802481) 🔵 tlinalg-pool 13.90× faster than lapack-mkl ! between-start range over 5%
lu_solve_prepared/n32/b1024 0.857903 ± 0.0116911 (W 0.0157464) 5.80603 ± 0.0737261 (W 0.0213193) 🔵 tlinalg-pool 6.77× faster than lapack-mkl
lu_solve_prepared/n4/b1024 0.0336225 ± 0.000189435 (W 0.000110729) 0.324848 ± 0.000329841 (W 0.0024088) 🔵 tlinalg-pool 9.66× faster than lapack-mkl
lu_solve_prepared/n8/b1024 0.076452 ± 0.000498405 (W 0.000718886) 0.948265 ± 0.00578778 (W 0.00101495) 🔵 tlinalg-pool 12.40× faster than lapack-mkl
qr/n16/b1024 1.39302 ± 0.0131984 (W 0.00383314) 10.8516 ± 0.0238972 (W 0.0207752) 🔵 tlinalg-pool 7.79× faster than lapack-mkl
qr/n32/b1024 9.05035 ± 0.0621253 (W 0.0533833) 49.8696 ± 0.339317 (W 0.0481521) 🔵 tlinalg-pool 5.51× faster than lapack-mkl
qr/n4/b1024 0.116982 ± 0.000568191 (W 0.00068379) 0.992902 ± 0.01388 (W 0.00281623) 🔵 tlinalg-pool 8.49× faster than lapack-mkl
qr/n8/b1024 0.343191 ± 0.000496805 (W 0.00223239) 2.81914 ± 0.00697732 (W 0.00382445) 🔵 tlinalg-pool 8.21× faster than lapack-mkl
rank_revealing_qr/n16/b1024 1.60772 ± 0.0109874 (W 0.00518074) 15.9342 ± 0.0497725 (W 0.0428035) 🔵 tlinalg-pool 9.91× faster than lapack-mkl
rank_revealing_qr/n32/b1024 8.46454 ± 0.0226968 (W 0.0173802) 68.0391 ± 0.103188 (W 0.0579193) 🔵 tlinalg-pool 8.04× faster than lapack-mkl
rank_revealing_qr/n4/b1024 0.196785 ± 0.00117459 (W 0.00161189) 1.92676 ± 0.00901126 (W 0.00492602) 🔵 tlinalg-pool 9.79× faster than lapack-mkl
rank_revealing_qr/n8/b1024 0.500244 ± 0.00416756 (W 0.00230648) 4.98903 ± 0.0125952 (W 0.0158562) 🔵 tlinalg-pool 9.97× faster than lapack-mkl
solve/n16/b1024 0.504792 ± 0.00500344 (W 0.00328202) 5.32623 ± 0.0316632 (W 0.00702174) 🔵 tlinalg-pool 10.55× faster than lapack-mkl
solve/n32/b1024 1.84128 ± 0.00559207 (W 0.0135305) 21.2901 ± 0.0837487 (W 0.042374) 🔵 tlinalg-pool 11.56× faster than lapack-mkl
solve/n4/b1024 0.0768775 ± 0.000971446 (W 0.00138394) 0.596172 ± 0.00161865 (W 0.00253086) 🔵 tlinalg-pool 7.75× faster than lapack-mkl
solve/n8/b1024 0.192462 ± 0.00167068 (W 0.00186658) 1.64203 ± 0.0110557 (W 0.00392524) 🔵 tlinalg-pool 8.53× faster than lapack-mkl
svd_full/n16/b1024 7.23494 ± 0.215149 (W 0.0104705) 68.1057 ± 0.111342 (W 0.121762) 🔵 tlinalg-pool 9.41× faster than lapack-mkl ! between-start range over 5%
svd_full/n32/b1024 32.2954 ± 1.4704 (W 0.0349428) 280.013 ± 1.73455 (W 0.403932) 🔵 tlinalg-pool 8.67× faster than lapack-mkl ! between-start range over 5%
svd_full/n4/b1024 0.656308 ± 0.00199331 (W 0.00287165) 7.57509 ± 0.0114215 (W 0.0108629) 🔵 tlinalg-pool 11.54× faster than lapack-mkl
svd_full/n8/b1024 2.09477 ± 0.0075509 (W 0.00234563) 22.4333 ± 0.0577012 (W 0.018546) 🔵 tlinalg-pool 10.71× faster than lapack-mkl
svd_thin/n16/b1024 7.30161 ± 0.109845 (W 0.0121055) 68.226 ± 0.211607 (W 0.0480674) 🔵 tlinalg-pool 9.34× faster than lapack-mkl
svd_thin/n32/b1024 31.6144 ± 1.29042 (W 0.0302418) 280.755 ± 1.5689 (W 0.563023) 🔵 tlinalg-pool 8.88× faster than lapack-mkl ! between-start range over 5%
svd_thin/n4/b1024 0.660842 ± 0.0277084 (W 0.00235839) 7.56623 ± 0.00689966 (W 0.0105329) 🔵 tlinalg-pool 11.45× faster than lapack-mkl ! between-start range over 5%
svd_thin/n8/b1024 2.0973 ± 0.0211672 (W 0.00438508) 22.4731 ± 0.0759655 (W 0.0210141) 🔵 tlinalg-pool 10.72× faster than lapack-mkl
svd_values/n16/b1024 3.2133 ± 0.0111974 (W 0.00398733) 27.4942 ± 0.105445 (W 0.0239469) 🔵 tlinalg-pool 8.56× faster than lapack-mkl
svd_values/n32/b1024 11.5924 ± 0.0669832 (W 0.00715205) 102.873 ± 0.085737 (W 0.0410978) 🔵 tlinalg-pool 8.87× faster than lapack-mkl
svd_values/n4/b1024 0.43886 ± 0.000991242 (W 0.00225378) 3.54738 ± 0.00812642 (W 0.00500346) 🔵 tlinalg-pool 8.08× faster than lapack-mkl
svd_values/n8/b1024 1.15005 ± 0.0058432 (W 0.00195437) 8.93908 ± 0.0158868 (W 0.0143353) 🔵 tlinalg-pool 7.77× faster than lapack-mkl
triangular_solve/n16/b1024 0.128112 ± 0.000771222 (W 0.00152988) 1.4305 ± 0.0376085 (W 0.00238329) 🔵 tlinalg-pool 11.17× faster than lapack-mkl
triangular_solve/n32/b1024 0.44712 ± 0.00819434 (W 0.0089471) 3.61862 ± 0.00578569 (W 0.00955593) 🔵 tlinalg-pool 8.09× faster than lapack-mkl
triangular_solve/n4/b1024 0.023764 ± 0.000171367 (W 7.94309e-05) 0.30992 ± 0.000655568 (W 0.00238498) 🔵 tlinalg-pool 13.04× faster than lapack-mkl
triangular_solve/n8/b1024 0.0542105 ± 0.00116934 (W 0.000185104) 0.6501 ± 0.00342499 (W 0.00419132) 🔵 tlinalg-pool 11.99× faster than lapack-mkl

canary-factor, f64, 8T (CPU 4-11)

case tlinalg-pool (ms) lapack-mkl (ms) vs lapack-mkl notes
cholesky/n128/b1 0.0735465 ± 0.000687156 (W 0.000853346) 0.0933735 ± 0.000476988 (W 0.00126529) 🔵 tlinalg-pool 1.27× faster than lapack-mkl
lu/n128/b1 0.167216 ± 0.00106527 (W 0.00165127) 0.149533 ± 0.000908165 (W 0.00131143) 🔴 tlinalg-pool 1.12× slower than lapack-mkl
lu_factor/n128/b1 0.092332 ± 0.000776729 (W 0.000346523) 0.135632 ± 0.000514717 (W 0.00180098) 🔵 tlinalg-pool 1.47× faster than lapack-mkl
lu_factor_solve/n128/b1 0.103141 ± 0.000524736 (W 0.000676954) 0.158649 ± 0.00101824 (W 0.00198641) 🔵 tlinalg-pool 1.54× faster than lapack-mkl
lu_solve_prepared/n128/b1 0.011251 ± 0.00018038 (W 3.82569e-05) 0.016716 ± 6.18466e-05 (W 0.000193425) 🔵 tlinalg-pool 1.49× faster than lapack-mkl dispatch-dominated (row under 20 us)
solve/n128/b1 0.122693 ± 0.00133637 (W 0.000828576) 0.165031 ± 0.000725834 (W 0.00177351) 🔵 tlinalg-pool 1.35× faster than lapack-mkl
svd_values/n128/b1 1.14748 ± 0.00443571 (W 0.00325793) 1.11546 ± 0.00569049 (W 0.00357881) ·
triangular_solve/n128/b1 0.00794 ± 2.94392e-05 (W 4.18594e-05) 0.009773 ± 0.00020563 (W 0.000198949) 🔵 tlinalg-pool 1.23× faster than lapack-mkl dispatch-dominated (row under 20 us)

canary-factor, c64, 8T (CPU 4-11)

case tlinalg-pool (ms) lapack-mkl (ms) vs lapack-mkl notes
cholesky/n128/b1 0.160518 ± 0.0011965 (W 0.00182112) 0.405117 ± 0.00300999 (W 0.00285939) 🔵 tlinalg-pool 2.52× faster than lapack-mkl
lu/n128/b1 0.321968 ± 0.00733412 (W 0.00227619) 0.436812 ± 0.00224565 (W 0.0032091) 🔵 tlinalg-pool 1.36× faster than lapack-mkl ! between-start range over 5%
lu_factor/n128/b1 0.219212 ± 0.00177031 (W 0.00195326) 0.405553 ± 0.00236996 (W 0.00340396) 🔵 tlinalg-pool 1.85× faster than lapack-mkl
lu_factor_solve/n128/b1 0.253315 ± 0.000942247 (W 0.00208183) 0.434641 ± 0.00172474 (W 0.00205019) 🔵 tlinalg-pool 1.72× faster than lapack-mkl
lu_solve_prepared/n128/b1 0.033838 ± 0.000161348 (W 0.000159423) 0.025257 ± 0.000181292 (W 0.000241655) 🔴 tlinalg-pool 1.34× slower than lapack-mkl
solve/n128/b1 0.277134 ± 0.00173591 (W 0.00158139) 0.448612 ± 0.00209856 (W 0.00334157) 🔵 tlinalg-pool 1.62× faster than lapack-mkl
svd_values/n128/b1 1.47441 ± 0.00310378 (W 0.00210003) 1.87705 ± 0.00559179 (W 0.00509176) 🔵 tlinalg-pool 1.27× faster than lapack-mkl
triangular_solve/n128/b1 0.0202525 ± 0.000170177 (W 0.000135064) 0.0128685 ± 0.000132414 (W 9.70445e-05) 🔴 tlinalg-pool 1.57× slower than lapack-mkl

canary-decompose, f64, 8T (CPU 4-11)

case tlinalg-pool (ms) lapack-mkl (ms) vs lapack-mkl notes
eigh/n512/b1 27.4738 ± 0.0563774 (W 0.0339181) 14.7972 ± 0.0855144 (W 0.0394975) 🔴 tlinalg-pool 1.86× slower than lapack-mkl
eigvalsh/n512/b1 16.4032 ± 0.058584 (W 0.0195112) 12.1463 ± 0.0371633 (W 0.0164584) 🔴 tlinalg-pool 1.35× slower than lapack-mkl
householder_apply/n512/b1 1.59774 ± 0.00381379 (W 0.00369377) 3.15687 ± 0.0173248 (W 0.0113455) 🔵 tlinalg-pool 1.98× faster than lapack-mkl
householder_factor/n512/b1 1.84929 ± 0.00988124 (W 0.00247655) 2.12667 ± 0.00747398 (W 0.00489313) 🔵 tlinalg-pool 1.15× faster than lapack-mkl
qr/n512/b1 5.53346 ± 0.025967 (W 0.00820808) 8.72886 ± 0.0579293 (W 0.0151971) 🔵 tlinalg-pool 1.58× faster than lapack-mkl
rank_revealing_qr/n512/b1 6.75077 ± 0.0306744 (W 0.0168998) 24.4183 ± 0.120216 (W 0.0416585) 🔵 tlinalg-pool 3.62× faster than lapack-mkl
svd_thin/n512/b1 38.5014 ± 0.0894592 (W 0.0483642) 48.629 ± 0.547267 (W 0.107363) 🔵 tlinalg-pool 1.26× faster than lapack-mkl

canary-decompose, c64, 8T (CPU 4-11)

case tlinalg-pool (ms) lapack-mkl (ms) vs lapack-mkl notes
eigh/n512/b1 37.239 ± 0.0486318 (W 0.0445833) 65.6223 ± 0.166345 (W 0.0382979) 🔵 tlinalg-pool 1.76× faster than lapack-mkl
eigvalsh/n512/b1 20.4405 ± 0.0299368 (W 0.0257485) 28.0051 ± 0.0697041 (W 0.0320551) 🔵 tlinalg-pool 1.37× faster than lapack-mkl
householder_apply/n512/b1 5.64474 ± 0.0182834 (W 0.0105379) 8.22309 ± 0.0216757 (W 0.0132105) 🔵 tlinalg-pool 1.46× faster than lapack-mkl
householder_factor/n512/b1 5.26644 ± 0.0221716 (W 0.0113588) 5.50302 ± 0.019781 (W 0.00920332) ·
qr/n512/b1 13.2016 ± 0.0448342 (W 0.0159747) 15.4331 ± 0.0478059 (W 0.0273575) 🔵 tlinalg-pool 1.17× faster than lapack-mkl
rank_revealing_qr/n512/b1 14.2815 ± 0.0232676 (W 0.0218404) 59.1693 ± 0.403852 (W 0.139036) 🔵 tlinalg-pool 4.14× faster than lapack-mkl
svd_thin/n512/b1 75.8311 ± 0.134335 (W 0.122135) 86.3747 ± 0.848109 (W 0.643391) 🔵 tlinalg-pool 1.14× faster than lapack-mkl

canary-rectangular, f64, 8T (CPU 4-11)

case tlinalg-pool (ms) lapack-mkl (ms) vs lapack-mkl notes
householder_factor/4096x256/b1 6.01493 ± 0.0439439 (W 0.0111476) 6.23455 ± 0.00943519 (W 0.043483) ·
qr/4096x256/b1 22.4006 ± 0.313193 (W 0.086658) 18.2284 ± 0.190078 (W 0.0831052) 🔴 tlinalg-pool 1.23× slower than lapack-mkl
rank_revealing_qr/4096x256/b1 21.7024 ± 0.0399421 (W 0.0529407) 56.27 ± 1.30057 (W 1.01239) 🔵 tlinalg-pool 2.59× faster than lapack-mkl ! between-start range over 5%
svd_thin/4096x256/b1 35.1677 ± 0.0746121 (W 0.0546528) 34.1378 ± 0.0932472 (W 0.059167) ·

canary-rectangular, c64, 8T (CPU 4-11)

case tlinalg-pool (ms) lapack-mkl (ms) vs lapack-mkl notes
householder_factor/4096x256/b1 17.7569 ± 0.0756208 (W 0.0373886) 16.8929 ± 0.116388 (W 0.0968582) ·
qr/4096x256/b1 58.7002 ± 0.16483 (W 0.120004) 40.6364 ± 0.208148 (W 0.23166) 🔴 tlinalg-pool 1.44× slower than lapack-mkl
rank_revealing_qr/4096x256/b1 58.6503 ± 0.500407 (W 0.185234) 155.816 ± 2.07293 (W 0.435231) 🔵 tlinalg-pool 2.66× faster than lapack-mkl
svd_thin/4096x256/b1 108.93 ± 0.267681 (W 0.203813) 75.3067 ± 0.481383 (W 0.535393) 🔴 tlinalg-pool 1.45× slower than lapack-mkl

Repeated process sets describe observed repeatability under the recorded order schedule; they are not a paired baseline/candidate experiment and do not establish equal performance between engines.