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 1.42%, worst 44.91% among cells of 1 us or more (a sub-microsecond row is timer and scheduler jitter, not kernel time); 31 of 396 engine cells exceed 5% and carry !; median within-process relative SD 0.30%. 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-openblas (ms) vs lapack-openblas notes
cholesky/n16/b1024 0.113637 ± 0.000702971 (W 0.00146877) 1.0144 ± 0.00941944 (W 0.00295457) 🔵 tlinalg-pool 8.93× faster than lapack-openblas
cholesky/n32/b1024 0.473548 ± 0.0204748 (W 0.00380976) 3.6075 ± 0.0349977 (W 0.0103144) 🔵 tlinalg-pool 7.62× faster than lapack-openblas ! between-start range over 5%
cholesky/n4/b1024 0.178849 ± 0.00113335 (W 0.00170774) 0.19552 ± 0.00122606 (W 0.00166425) 🔵 tlinalg-pool 1.09× faster than lapack-openblas
cholesky/n8/b1024 0.0473985 ± 0.000402236 (W 5.98449e-05) 0.399275 ± 0.0164214 (W 0.00296936) 🔵 tlinalg-pool 8.42× faster than lapack-openblas ! between-start range over 5%
eig/n16/b1024 12.1179 ± 0.060568 (W 0.0106715) 68.6138 ± 0.440776 (W 0.232486) 🔵 tlinalg-pool 5.66× faster than lapack-openblas
eig/n32/b1024 54.1907 ± 0.21907 (W 0.230475) 323.455 ± 1.9205 (W 0.460581) 🔵 tlinalg-pool 5.97× faster than lapack-openblas
eig/n4/b1024 0.81181 ± 0.00258999 (W 0.0024569) 5.32108 ± 0.0269894 (W 0.00854824) 🔵 tlinalg-pool 6.55× faster than lapack-openblas
eig/n8/b1024 3.03126 ± 0.0159111 (W 0.00647191) 17.4806 ± 0.101067 (W 0.0213814) 🔵 tlinalg-pool 5.77× faster than lapack-openblas
eigh/n16/b1024 3.78046 ± 0.0180345 (W 0.00511951) 40.2435 ± 0.240105 (W 0.0747414) 🔵 tlinalg-pool 10.65× faster than lapack-openblas
eigh/n32/b1024 13.9953 ± 0.0140153 (W 0.0109558) 117.371 ± 1.14203 (W 0.128688) 🔵 tlinalg-pool 8.39× faster than lapack-openblas
eigh/n4/b1024 0.322522 ± 0.00264501 (W 0.00262968) 3.02853 ± 0.047657 (W 0.005041) 🔵 tlinalg-pool 9.39× faster than lapack-openblas
eigh/n8/b1024 1.09772 ± 0.00696365 (W 0.00287568) 10.8607 ± 0.103804 (W 0.0167573) 🔵 tlinalg-pool 9.89× faster than lapack-openblas
eigvals/n16/b1024 8.37632 ± 0.0309983 (W 0.00801003) 49.6668 ± 0.34083 (W 0.0455797) 🔵 tlinalg-pool 5.93× faster than lapack-openblas
eigvals/n32/b1024 36.7128 ± 0.0834891 (W 0.0502087) 218.867 ± 0.529766 (W 0.461716) 🔵 tlinalg-pool 5.96× faster than lapack-openblas
eigvals/n4/b1024 0.439795 ± 0.00327809 (W 0.00224756) 3.42441 ± 0.0212902 (W 0.00535561) 🔵 tlinalg-pool 7.79× faster than lapack-openblas
eigvals/n8/b1024 2.0019 ± 0.00612204 (W 0.00517038) 13.23 ± 0.020519 (W 0.0126543) 🔵 tlinalg-pool 6.61× faster than lapack-openblas
eigvalsh/n16/b1024 2.85802 ± 0.0177015 (W 0.00493553) 14.566 ± 0.0691657 (W 0.0109715) 🔵 tlinalg-pool 5.10× faster than lapack-openblas
eigvalsh/n32/b1024 10.6853 ± 0.0401972 (W 0.00772998) 49.975 ± 0.146727 (W 0.0378518) 🔵 tlinalg-pool 4.68× faster than lapack-openblas
eigvalsh/n4/b1024 0.264394 ± 0.00479905 (W 0.0017066) 1.39248 ± 0.00893127 (W 0.00509385) 🔵 tlinalg-pool 5.27× faster than lapack-openblas
eigvalsh/n8/b1024 0.882705 ± 0.00762956 (W 0.00289559) 4.54011 ± 0.0375614 (W 0.00620435) 🔵 tlinalg-pool 5.14× faster than lapack-openblas
full_piv_lu/n16/b1024 0.511039 ± 0.00555535 (W 0.00438864) 4.74741 ± 0.131469 (W 0.0189384) 🔵 tlinalg-pool 9.29× faster than lapack-openblas ! between-start range over 5%
full_piv_lu/n32/b1024 2.01948 ± 0.0294496 (W 0.00633381) 28.4704 ± 0.444058 (W 0.122104) 🔵 tlinalg-pool 14.10× faster than lapack-openblas
full_piv_lu/n4/b1024 0.493667 ± 0.00360875 (W 0.00379742) 0.37206 ± 0.00432701 (W 0.00220794) 🔴 tlinalg-pool 1.33× slower than lapack-openblas
full_piv_lu/n8/b1024 0.181084 ± 0.00161244 (W 0.00213312) 1.03839 ± 0.0108366 (W 0.00366431) 🔵 tlinalg-pool 5.73× faster than lapack-openblas
full_piv_lu_solve/n16/b1024 0.517367 ± 0.00424877 (W 0.00302689) 6.39387 ± 0.166147 (W 0.0326794) 🔵 tlinalg-pool 12.36× faster than lapack-openblas ! between-start range over 5%
full_piv_lu_solve/n32/b1024 1.67728 ± 0.0264198 (W 0.0172894) 30.7264 ± 1.02755 (W 0.130431) 🔵 tlinalg-pool 18.32× faster than lapack-openblas ! between-start range over 5%
full_piv_lu_solve/n4/b1024 0.084643 ± 0.000454001 (W 0.000333903) 0.950705 ± 0.00543087 (W 0.00198699) 🔵 tlinalg-pool 11.23× faster than lapack-openblas
full_piv_lu_solve/n8/b1024 0.205689 ± 0.000799246 (W 0.00231976) 1.98263 ± 0.0193922 (W 0.00593324) 🔵 tlinalg-pool 9.64× faster than lapack-openblas
householder_apply/n16/b1024 0.62197 ± 0.0267052 (W 0.00191618) 4.4854 ± 0.0275885 (W 0.00823738) 🔵 tlinalg-pool 7.21× faster than lapack-openblas ! between-start range over 5%
householder_apply/n32/b1024 2.44853 ± 0.0664279 (W 0.00906694) 19.3162 ± 0.280287 (W 0.0624283) 🔵 tlinalg-pool 7.89× faster than lapack-openblas ! between-start range over 5%
householder_apply/n4/b1024 0.0543915 ± 0.00182311 (W 0.000904463) 0.414374 ± 0.00191379 (W 0.00253759) 🔵 tlinalg-pool 7.62× faster than lapack-openblas ! between-start range over 5%
householder_apply/n8/b1024 0.171665 ± 0.00735371 (W 0.00167355) 1.20516 ± 0.00941745 (W 0.00349716) 🔵 tlinalg-pool 7.02× faster than lapack-openblas ! between-start range over 5%
householder_factor/n16/b1024 0.509896 ± 0.00393539 (W 0.00199243) 4.3828 ± 0.016638 (W 0.00734207) 🔵 tlinalg-pool 8.60× faster than lapack-openblas
householder_factor/n32/b1024 1.68705 ± 0.0256475 (W 0.00521465) 14.7129 ± 0.0439735 (W 0.020967) 🔵 tlinalg-pool 8.72× faster than lapack-openblas
householder_factor/n4/b1024 0.063884 ± 0.00018518 (W 0.00137804) 0.542114 ± 0.00155842 (W 0.00266131) 🔵 tlinalg-pool 8.49× faster than lapack-openblas
householder_factor/n8/b1024 0.168395 ± 0.00228822 (W 0.00113771) 1.5137 ± 0.010886 (W 0.00236105) 🔵 tlinalg-pool 8.99× faster than lapack-openblas
lu/n16/b1024 0.378718 ± 0.00556474 (W 0.00298386) 2.3824 ± 0.0269309 (W 0.0156598) 🔵 tlinalg-pool 6.29× faster than lapack-openblas
lu/n32/b1024 1.34402 ± 0.00911716 (W 0.007284) 9.54719 ± 0.0877557 (W 0.0178615) 🔵 tlinalg-pool 7.10× faster than lapack-openblas
lu/n4/b1024 0.30596 ± 0.0150693 (W 0.00323127) 0.323244 ± 0.0024349 (W 0.00227444) 🔵 tlinalg-pool 1.06× faster than lapack-openblas ! between-start range over 5%
lu/n8/b1024 0.134802 ± 0.000832268 (W 0.00157552) 0.805808 ± 0.0123864 (W 0.00332517) 🔵 tlinalg-pool 5.98× faster than lapack-openblas
lu_factor/n16/b1024 0.253138 ± 0.00271177 (W 0.00230391) 1.81208 ± 0.0196057 (W 0.00375656) 🔵 tlinalg-pool 7.16× faster than lapack-openblas
lu_factor/n32/b1024 0.728163 ± 0.00548621 (W 0.0038916) 6.68616 ± 0.0622966 (W 0.0122029) 🔵 tlinalg-pool 9.18× faster than lapack-openblas
lu_factor/n4/b1024 0.235906 ± 0.00343856 (W 0.00237391) 0.218282 ± 0.00039731 (W 0.0017539) 🔴 tlinalg-pool 1.08× slower than lapack-openblas
lu_factor/n8/b1024 0.095724 ± 0.000790473 (W 0.00146437) 0.568021 ± 0.00171527 (W 0.00340386) 🔵 tlinalg-pool 5.93× faster than lapack-openblas
lu_factor_solve/n16/b1024 0.322182 ± 0.0035143 (W 0.00226528) 3.44301 ± 0.0155718 (W 0.0143953) 🔵 tlinalg-pool 10.69× faster than lapack-openblas
lu_factor_solve/n32/b1024 0.883377 ± 0.00402351 (W 0.00313532) 9.70563 ± 0.0528652 (W 0.0141085) 🔵 tlinalg-pool 10.99× faster than lapack-openblas
lu_factor_solve/n4/b1024 0.056932 ± 0.000312234 (W 0.000889787) 1.14556 ± 0.0273782 (W 0.00644519) 🔵 tlinalg-pool 20.12× faster than lapack-openblas
lu_factor_solve/n8/b1024 0.119087 ± 0.00147217 (W 0.0011653) 1.51461 ± 0.0191967 (W 0.0243727) 🔵 tlinalg-pool 12.72× faster than lapack-openblas
lu_solve_prepared/n16/b1024 0.0967655 ± 0.000634893 (W 0.00119696) 1.16731 ± 0.0232353 (W 0.0213954) 🔵 tlinalg-pool 12.06× faster than lapack-openblas
lu_solve_prepared/n32/b1024 0.231257 ± 0.00185917 (W 0.0025262) 2.13969 ± 0.0111898 (W 0.00864756) 🔵 tlinalg-pool 9.25× faster than lapack-openblas
lu_solve_prepared/n4/b1024 0.020799 ± 0.000412679 (W 0.000120644) 0.863685 ± 0.0104609 (W 0.00766977) 🔵 tlinalg-pool 41.53× faster than lapack-openblas
lu_solve_prepared/n8/b1024 0.0436565 ± 0.00037422 (W 0.000212915) 0.791886 ± 0.0338814 (W 0.0266088) 🔵 tlinalg-pool 18.14× faster than lapack-openblas ! between-start range over 5%
qr/n16/b1024 1.1455 ± 0.0120778 (W 0.00346222) 7.88776 ± 0.0457973 (W 0.0183901) 🔵 tlinalg-pool 6.89× faster than lapack-openblas
qr/n32/b1024 4.24405 ± 0.0312234 (W 0.00720853) 27.6089 ± 0.149946 (W 0.0310253) 🔵 tlinalg-pool 6.51× faster than lapack-openblas
qr/n4/b1024 0.10169 ± 0.0034508 (W 0.000463141) 0.966481 ± 0.0110653 (W 0.00363798) 🔵 tlinalg-pool 9.50× faster than lapack-openblas ! between-start range over 5%
qr/n8/b1024 0.31596 ± 0.0014777 (W 0.00186408) 2.64362 ± 0.0098279 (W 0.00571947) 🔵 tlinalg-pool 8.37× faster than lapack-openblas
rank_revealing_qr/n16/b1024 1.37069 ± 0.0269469 (W 0.00406587) 9.46642 ± 0.0545899 (W 0.0397452) 🔵 tlinalg-pool 6.91× faster than lapack-openblas
rank_revealing_qr/n32/b1024 4.37808 ± 0.0214128 (W 0.0205493) 34.1061 ± 0.179205 (W 0.0680055) 🔵 tlinalg-pool 7.79× faster than lapack-openblas
rank_revealing_qr/n4/b1024 0.176029 ± 0.00126602 (W 0.00171284) 1.17645 ± 0.00723306 (W 0.00303146) 🔵 tlinalg-pool 6.68× faster than lapack-openblas
rank_revealing_qr/n8/b1024 0.443988 ± 0.00908807 (W 0.00245319) 3.04034 ± 0.0125155 (W 0.00929313) 🔵 tlinalg-pool 6.85× faster than lapack-openblas
solve/n16/b1024 0.36112 ± 0.00397385 (W 0.00275424) 3.84673 ± 0.0633988 (W 0.0106512) 🔵 tlinalg-pool 10.65× faster than lapack-openblas
solve/n32/b1024 0.956388 ± 0.0158358 (W 0.00474281) 9.86737 ± 0.0458723 (W 0.0136092) 🔵 tlinalg-pool 10.32× faster than lapack-openblas
solve/n4/b1024 0.053495 ± 0.00095947 (W 0.000381996) 1.34446 ± 0.042994 (W 0.00527474) 🔵 tlinalg-pool 25.13× faster than lapack-openblas ! between-start range over 5%
solve/n8/b1024 0.126951 ± 0.00111959 (W 0.00121283) 1.97649 ± 0.0240962 (W 0.00430178) 🔵 tlinalg-pool 15.57× faster than lapack-openblas
svd_full/n16/b1024 6.86574 ± 0.203534 (W 0.00577274) 81.2154 ± 0.740273 (W 0.081829) 🔵 tlinalg-pool 11.83× faster than lapack-openblas ! between-start range over 5%
svd_full/n32/b1024 23.6293 ± 0.185813 (W 0.0206994) 239.513 ± 1.19443 (W 0.42837) 🔵 tlinalg-pool 10.14× faster than lapack-openblas
svd_full/n4/b1024 0.574718 ± 0.032691 (W 0.00226631) 6.39579 ± 0.100113 (W 0.012799) 🔵 tlinalg-pool 11.13× faster than lapack-openblas ! between-start range over 5%
svd_full/n8/b1024 1.91186 ± 0.0405172 (W 0.00165365) 21.3165 ± 0.21028 (W 0.026294) 🔵 tlinalg-pool 11.15× faster than lapack-openblas
svd_thin/n16/b1024 6.87327 ± 0.213082 (W 0.00587681) 80.8895 ± 0.74564 (W 0.0860693) 🔵 tlinalg-pool 11.77× faster than lapack-openblas ! between-start range over 5%
svd_thin/n32/b1024 23.6614 ± 0.136814 (W 0.0278183) 239.349 ± 0.636488 (W 0.207811) 🔵 tlinalg-pool 10.12× faster than lapack-openblas
svd_thin/n4/b1024 0.575444 ± 0.0196686 (W 0.00221472) 6.43313 ± 0.0737174 (W 0.00906488) 🔵 tlinalg-pool 11.18× faster than lapack-openblas ! between-start range over 5%
svd_thin/n8/b1024 1.87604 ± 0.0433958 (W 0.00299454) 21.3416 ± 0.177102 (W 0.0228253) 🔵 tlinalg-pool 11.38× faster than lapack-openblas ! between-start range over 5%
svd_values/n16/b1024 2.81168 ± 0.0309683 (W 0.00438005) 25.0764 ± 0.104787 (W 0.022889) 🔵 tlinalg-pool 8.92× faster than lapack-openblas
svd_values/n32/b1024 9.42195 ± 0.0232528 (W 0.0058235) 79.7255 ± 0.415219 (W 0.259808) 🔵 tlinalg-pool 8.46× faster than lapack-openblas
svd_values/n4/b1024 0.365042 ± 0.000838665 (W 0.00225174) 3.47329 ± 0.0133249 (W 0.0050386) 🔵 tlinalg-pool 9.51× faster than lapack-openblas
svd_values/n8/b1024 1.00597 ± 0.00299834 (W 0.0020489) 8.94876 ± 0.0370544 (W 0.0131582) 🔵 tlinalg-pool 8.90× faster than lapack-openblas
triangular_solve/n16/b1024 0.067717 ± 0.000161824 (W 0.000231471) 0.897073 ± 0.0143926 (W 0.00289873) 🔵 tlinalg-pool 13.25× faster than lapack-openblas
triangular_solve/n32/b1024 0.156678 ± 0.00095801 (W 0.00232962) 1.47367 ± 0.0141495 (W 0.00530808) 🔵 tlinalg-pool 9.41× faster than lapack-openblas
triangular_solve/n4/b1024 0.0143115 ± 9.32041e-05 (W 0.0002423) 0.716373 ± 0.0332544 (W 0.00687479) 🔵 tlinalg-pool 50.06× faster than lapack-openblas ! between-start range over 5%; dispatch-dominated (row under 20 us)
triangular_solve/n8/b1024 0.0306775 ± 0.000426646 (W 0.000113389) 0.75159 ± 0.0040158 (W 0.00425316) 🔵 tlinalg-pool 24.50× faster than lapack-openblas

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

case tlinalg-pool (ms) lapack-openblas (ms) vs lapack-openblas notes
cholesky/n16/b1024 0.159257 ± 0.00103866 (W 0.00145458) 1.544 ± 0.00549866 (W 0.0100414) 🔵 tlinalg-pool 9.70× faster than lapack-openblas
cholesky/n32/b1024 0.788136 ± 0.00355739 (W 0.00802258) 6.31548 ± 0.110247 (W 0.0103851) 🔵 tlinalg-pool 8.01× faster than lapack-openblas
cholesky/n4/b1024 0.0275915 ± 0.000276575 (W 9.05184e-05) 0.231838 ± 0.00176635 (W 0.00175238) 🔵 tlinalg-pool 8.40× faster than lapack-openblas
cholesky/n8/b1024 0.054752 ± 0.000614653 (W 0.000386386) 0.486092 ± 0.00168639 (W 0.00253009) 🔵 tlinalg-pool 8.88× faster than lapack-openblas
eig/n16/b1024 16.4531 ± 0.0238473 (W 0.0139802) 199.277 ± 0.413577 (W 0.326167) 🔵 tlinalg-pool 12.11× faster than lapack-openblas
eig/n32/b1024 67.8968 ± 0.266678 (W 0.0442335) 887.128 ± 2.66053 (W 0.744715) 🔵 tlinalg-pool 13.07× faster than lapack-openblas
eig/n4/b1024 1.37833 ± 0.00625651 (W 0.00309467) 13.254 ± 0.0560174 (W 0.0270338) 🔵 tlinalg-pool 9.62× faster than lapack-openblas
eig/n8/b1024 4.55381 ± 0.0210702 (W 0.0045855) 46.7242 ± 0.513448 (W 0.042586) 🔵 tlinalg-pool 10.26× faster than lapack-openblas
eigh/n16/b1024 4.02836 ± 0.0178157 (W 0.00924572) 45.3489 ± 0.123892 (W 0.069982) 🔵 tlinalg-pool 11.26× faster than lapack-openblas
eigh/n32/b1024 17.4876 ± 0.0460321 (W 0.0194357) 188.369 ± 0.625392 (W 0.189625) 🔵 tlinalg-pool 10.77× faster than lapack-openblas
eigh/n4/b1024 0.385589 ± 0.00189634 (W 0.0016078) 3.60396 ± 0.0067923 (W 0.00612087) 🔵 tlinalg-pool 9.35× faster than lapack-openblas
eigh/n8/b1024 1.17973 ± 0.00667414 (W 0.00229871) 12.4114 ± 0.0463819 (W 0.0122804) 🔵 tlinalg-pool 10.52× faster than lapack-openblas
eigvals/n16/b1024 9.92603 ± 0.0591576 (W 0.00795602) 149.391 ± 0.391381 (W 0.241759) 🔵 tlinalg-pool 15.05× faster than lapack-openblas
eigvals/n32/b1024 43.029 ± 0.184073 (W 0.0166894) 601.064 ± 1.83733 (W 0.667594) 🔵 tlinalg-pool 13.97× faster than lapack-openblas
eigvals/n4/b1024 0.696284 ± 0.013261 (W 0.00324871) 10.0491 ± 0.0311495 (W 0.012939) 🔵 tlinalg-pool 14.43× faster than lapack-openblas
eigvals/n8/b1024 2.61834 ± 0.010365 (W 0.00151636) 37.0604 ± 0.151491 (W 0.047249) 🔵 tlinalg-pool 14.15× faster than lapack-openblas
eigvalsh/n16/b1024 3.04073 ± 0.0150375 (W 0.00329569) 19.4154 ± 0.0330966 (W 0.0348549) 🔵 tlinalg-pool 6.39× faster than lapack-openblas
eigvalsh/n32/b1024 11.5661 ± 0.038041 (W 0.00742776) 73.2501 ± 0.140848 (W 0.18621) 🔵 tlinalg-pool 6.33× faster than lapack-openblas
eigvalsh/n4/b1024 0.314271 ± 0.00389868 (W 0.00196613) 2.12523 ± 0.0139032 (W 0.00506523) 🔵 tlinalg-pool 6.76× faster than lapack-openblas
eigvalsh/n8/b1024 0.949742 ± 0.00468983 (W 0.00193887) 6.17475 ± 0.0229133 (W 0.0170308) 🔵 tlinalg-pool 6.50× faster than lapack-openblas
full_piv_lu/n16/b1024 0.700959 ± 0.0114697 (W 0.00693635) 33.3627 ± 0.253327 (W 0.0571457) 🔵 tlinalg-pool 47.60× faster than lapack-openblas
full_piv_lu/n32/b1024 3.04988 ± 0.00601346 (W 0.0153116) 199.8 ± 0.932354 (W 0.506835) 🔵 tlinalg-pool 65.51× faster than lapack-openblas
full_piv_lu/n4/b1024 0.089202 ± 0.0195535 (W 0.00223192) 0.910949 ± 0.00738149 (W 0.00227128) 🔵 tlinalg-pool 10.21× faster than lapack-openblas ! between-start range over 5%; p position effect over 5%
full_piv_lu/n8/b1024 0.216544 ± 0.00166869 (W 0.00280027) 3.89587 ± 0.0247942 (W 0.00739432) 🔵 tlinalg-pool 17.99× faster than lapack-openblas
full_piv_lu_solve/n16/b1024 0.813813 ± 0.0065582 (W 0.00296521) 37.8926 ± 0.123346 (W 0.0667898) 🔵 tlinalg-pool 46.56× faster than lapack-openblas
full_piv_lu_solve/n32/b1024 2.76164 ± 0.0177748 (W 0.0141706) 211.169 ± 0.601992 (W 0.349811) 🔵 tlinalg-pool 76.47× faster than lapack-openblas
full_piv_lu_solve/n4/b1024 0.121357 ± 0.00209886 (W 0.00251767) 2.18829 ± 0.0135075 (W 0.00406473) 🔵 tlinalg-pool 18.03× faster than lapack-openblas
full_piv_lu_solve/n8/b1024 0.289766 ± 0.00251552 (W 0.0025748) 6.20082 ± 0.0618638 (W 0.011878) 🔵 tlinalg-pool 21.40× faster than lapack-openblas
householder_apply/n16/b1024 0.625387 ± 0.0119581 (W 0.00444785) 7.22676 ± 0.156721 (W 0.0178936) 🔵 tlinalg-pool 11.56× faster than lapack-openblas
householder_apply/n32/b1024 3.57525 ± 0.0156446 (W 0.0130742) 35.9034 ± 0.295803 (W 0.0375794) 🔵 tlinalg-pool 10.04× faster than lapack-openblas
householder_apply/n4/b1024 0.0526185 ± 0.00215527 (W 0.000123449) 0.429612 ± 0.00164544 (W 0.00260464) 🔵 tlinalg-pool 8.16× faster than lapack-openblas ! between-start range over 5%
householder_apply/n8/b1024 0.154348 ± 0.00456334 (W 0.00129419) 1.48348 ± 0.0071478 (W 0.00471445) 🔵 tlinalg-pool 9.61× faster than lapack-openblas ! between-start range over 5%
householder_factor/n16/b1024 0.610665 ± 0.00810943 (W 0.00375274) 7.35798 ± 0.0387324 (W 0.02374) 🔵 tlinalg-pool 12.05× faster than lapack-openblas
householder_factor/n32/b1024 2.60039 ± 0.03213 (W 0.00891376) 27.3285 ± 0.0814685 (W 0.0337291) 🔵 tlinalg-pool 10.51× faster than lapack-openblas
householder_factor/n4/b1024 0.0787575 ± 0.000444746 (W 0.00174085) 0.92657 ± 0.00270134 (W 0.00221623) 🔵 tlinalg-pool 11.76× faster than lapack-openblas
householder_factor/n8/b1024 0.20084 ± 0.00161951 (W 0.00177071) 2.36745 ± 0.0113182 (W 0.00426332) 🔵 tlinalg-pool 11.79× faster than lapack-openblas
lu/n16/b1024 0.451849 ± 0.00545225 (W 0.0067817) 3.9615 ± 0.13928 (W 0.0128001) 🔵 tlinalg-pool 8.77× faster than lapack-openblas ! between-start range over 5%
lu/n32/b1024 2.18331 ± 0.012896 (W 0.0115653) 15.4739 ± 0.10468 (W 0.0206575) 🔵 tlinalg-pool 7.09× faster than lapack-openblas
lu/n4/b1024 0.0634035 ± 0.00110509 (W 0.000958212) 0.379243 ± 0.0047094 (W 0.00285604) 🔵 tlinalg-pool 5.98× faster than lapack-openblas
lu/n8/b1024 0.136054 ± 0.00171974 (W 0.0020534) 1.03326 ± 0.0274728 (W 0.00244628) 🔵 tlinalg-pool 7.59× faster than lapack-openblas ! between-start range over 5%
lu_factor/n16/b1024 0.286153 ± 0.00428108 (W 0.00148544) 2.67642 ± 0.00781021 (W 0.00442222) 🔵 tlinalg-pool 9.35× faster than lapack-openblas
lu_factor/n32/b1024 1.24855 ± 0.00921724 (W 0.00691534) 11.1836 ± 0.0506941 (W 0.0451721) 🔵 tlinalg-pool 8.96× faster than lapack-openblas
lu_factor/n4/b1024 0.0608435 ± 0.000330306 (W 0.000571555) 0.271467 ± 0.0021826 (W 0.00174316) 🔵 tlinalg-pool 4.46× faster than lapack-openblas
lu_factor/n8/b1024 0.104851 ± 0.00109129 (W 0.00175533) 0.75632 ± 0.00144255 (W 0.002189) 🔵 tlinalg-pool 7.21× faster than lapack-openblas
lu_factor_solve/n16/b1024 0.455175 ± 0.000648186 (W 0.00337846) 5.02967 ± 0.0587444 (W 0.00832606) 🔵 tlinalg-pool 11.05× faster than lapack-openblas
lu_factor_solve/n32/b1024 1.70367 ± 0.0118262 (W 0.00524951) 15.3664 ± 0.062087 (W 0.0177884) 🔵 tlinalg-pool 9.02× faster than lapack-openblas
lu_factor_solve/n4/b1024 0.071248 ± 0.00079098 (W 0.00100132) 1.11088 ± 0.009298 (W 0.00364057) 🔵 tlinalg-pool 15.59× faster than lapack-openblas
lu_factor_solve/n8/b1024 0.160705 ± 0.00287389 (W 0.00132202) 2.29475 ± 0.0295728 (W 0.00616518) 🔵 tlinalg-pool 14.28× faster than lapack-openblas
lu_solve_prepared/n16/b1024 0.184329 ± 0.00173415 (W 0.00186127) 1.93875 ± 0.00841969 (W 0.00534189) 🔵 tlinalg-pool 10.52× faster than lapack-openblas
lu_solve_prepared/n32/b1024 0.873241 ± 0.00549215 (W 0.00440819) 4.02883 ± 0.020794 (W 0.0142927) 🔵 tlinalg-pool 4.61× faster than lapack-openblas
lu_solve_prepared/n4/b1024 0.0337535 ± 0.000487231 (W 0.000434394) 0.736144 ± 0.0257148 (W 0.00687378) 🔵 tlinalg-pool 21.81× faster than lapack-openblas ! between-start range over 5%
lu_solve_prepared/n8/b1024 0.076092 ± 0.000454849 (W 0.000329986) 1.33527 ± 0.0234547 (W 0.00522284) 🔵 tlinalg-pool 17.55× faster than lapack-openblas
qr/n16/b1024 1.39993 ± 0.0116288 (W 0.00456594) 12.6687 ± 0.0559292 (W 0.0100561) 🔵 tlinalg-pool 9.05× faster than lapack-openblas
qr/n32/b1024 9.00105 ± 0.0460372 (W 0.0187552) 49.6932 ± 0.311267 (W 0.0395781) 🔵 tlinalg-pool 5.52× faster than lapack-openblas
qr/n4/b1024 0.120816 ± 0.000822733 (W 0.000981448) 1.41452 ± 0.00583912 (W 0.00630851) 🔵 tlinalg-pool 11.71× faster than lapack-openblas
qr/n8/b1024 0.343129 ± 0.00227337 (W 0.00213184) 3.82147 ± 0.0181062 (W 0.00704435) 🔵 tlinalg-pool 11.14× faster than lapack-openblas
rank_revealing_qr/n16/b1024 1.58682 ± 0.00824498 (W 0.00322515) 16.8038 ± 0.0718976 (W 0.0340178) 🔵 tlinalg-pool 10.59× faster than lapack-openblas
rank_revealing_qr/n32/b1024 8.48042 ± 0.0357949 (W 0.0145924) 65.9249 ± 0.366738 (W 0.128677) 🔵 tlinalg-pool 7.77× faster than lapack-openblas
rank_revealing_qr/n4/b1024 0.195781 ± 0.00133197 (W 0.00148975) 1.83216 ± 0.0109604 (W 0.00528043) 🔵 tlinalg-pool 9.36× faster than lapack-openblas
rank_revealing_qr/n8/b1024 0.493375 ± 0.00311423 (W 0.0027148) 5.0154 ± 0.0240928 (W 0.00892492) 🔵 tlinalg-pool 10.17× faster than lapack-openblas
solve/n16/b1024 0.502684 ± 0.0106806 (W 0.00370911) 5.23085 ± 0.0494202 (W 0.0148813) 🔵 tlinalg-pool 10.41× faster than lapack-openblas
solve/n32/b1024 1.857 ± 0.0148855 (W 0.00383936) 16.2361 ± 0.0679426 (W 0.0261413) 🔵 tlinalg-pool 8.74× faster than lapack-openblas
solve/n4/b1024 0.0836465 ± 0.00106773 (W 0.00103217) 1.46229 ± 0.0166907 (W 0.0076383) 🔵 tlinalg-pool 17.48× faster than lapack-openblas
solve/n8/b1024 0.188136 ± 0.0060726 (W 0.00186065) 2.31756 ± 0.0154626 (W 0.0213636) 🔵 tlinalg-pool 12.32× faster than lapack-openblas ! between-start range over 5%
svd_full/n16/b1024 7.45161 ± 0.163487 (W 0.0104577) 93.4606 ± 0.124076 (W 0.275314) 🔵 tlinalg-pool 12.54× faster than lapack-openblas ! between-start range over 5%
svd_full/n32/b1024 31.0982 ± 0.120904 (W 0.0385905) 297.442 ± 0.602756 (W 0.445782) 🔵 tlinalg-pool 9.56× faster than lapack-openblas
svd_full/n4/b1024 0.674449 ± 0.0222475 (W 0.00242499) 7.86595 ± 0.0409101 (W 0.0158447) 🔵 tlinalg-pool 11.66× faster than lapack-openblas ! between-start range over 5%
svd_full/n8/b1024 2.10848 ± 0.0710424 (W 0.00174323) 24.8597 ± 0.0544166 (W 0.026577) 🔵 tlinalg-pool 11.79× faster than lapack-openblas ! between-start range over 5%
svd_thin/n16/b1024 7.43395 ± 0.131604 (W 0.0138429) 93.7738 ± 0.277578 (W 0.187002) 🔵 tlinalg-pool 12.61× faster than lapack-openblas
svd_thin/n32/b1024 31.0929 ± 0.130428 (W 0.0436691) 296.409 ± 1.19662 (W 0.367159) 🔵 tlinalg-pool 9.53× faster than lapack-openblas
svd_thin/n4/b1024 0.667902 ± 0.0282992 (W 0.00422807) 7.91632 ± 0.048634 (W 0.0121122) 🔵 tlinalg-pool 11.85× faster than lapack-openblas ! between-start range over 5%
svd_thin/n8/b1024 2.17544 ± 0.0678544 (W 0.00379189) 24.9239 ± 0.036912 (W 0.0208885) 🔵 tlinalg-pool 11.46× faster than lapack-openblas ! between-start range over 5%
svd_values/n16/b1024 3.18532 ± 0.0136457 (W 0.00208356) 33.4652 ± 0.0882183 (W 0.0277493) 🔵 tlinalg-pool 10.51× faster than lapack-openblas
svd_values/n32/b1024 11.3738 ± 0.0480991 (W 0.0084497) 110.254 ± 0.646616 (W 0.2538) 🔵 tlinalg-pool 9.69× faster than lapack-openblas
svd_values/n4/b1024 0.444388 ± 0.00328041 (W 0.00211202) 4.65059 ± 0.0296487 (W 0.0061206) 🔵 tlinalg-pool 10.47× faster than lapack-openblas
svd_values/n8/b1024 1.15058 ± 0.00658974 (W 0.00184838) 11.2713 ± 0.0717445 (W 0.00983786) 🔵 tlinalg-pool 9.80× faster than lapack-openblas
triangular_solve/n16/b1024 0.12829 ± 0.00145889 (W 0.00147589) 1.35913 ± 0.0191829 (W 0.00562788) 🔵 tlinalg-pool 10.59× faster than lapack-openblas
triangular_solve/n32/b1024 0.546847 ± 0.00572213 (W 0.00595351) 3.03749 ± 0.0129041 (W 0.00485001) 🔵 tlinalg-pool 5.55× faster than lapack-openblas
triangular_solve/n4/b1024 0.0240245 ± 0.000309937 (W 0.000137831) 0.790566 ± 0.0140814 (W 0.00239568) 🔵 tlinalg-pool 32.91× faster than lapack-openblas
triangular_solve/n8/b1024 0.0546765 ± 0.000359348 (W 0.000370225) 0.908322 ± 0.0153052 (W 0.00641218) 🔵 tlinalg-pool 16.61× faster than lapack-openblas

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

case tlinalg-pool (ms) lapack-openblas (ms) vs lapack-openblas notes
cholesky/n128/b1 0.0705165 ± 0.000828188 (W 0.00180657) 0.077074 ± 0.00108625 (W 0.00087669) 🔵 tlinalg-pool 1.09× faster than lapack-openblas
lu/n128/b1 0.173969 ± 0.00253684 (W 0.00172381) 0.14509 ± 0.00143983 (W 0.00195492) 🔴 tlinalg-pool 1.20× slower than lapack-openblas
lu_factor/n128/b1 0.093354 ± 0.00100087 (W 0.000211139) 0.130783 ± 0.000820345 (W 0.00185299) 🔵 tlinalg-pool 1.40× faster than lapack-openblas
lu_factor_solve/n128/b1 0.104976 ± 0.000849844 (W 0.000491159) 0.147435 ± 0.000689228 (W 0.00170845) 🔵 tlinalg-pool 1.40× faster than lapack-openblas
lu_solve_prepared/n128/b1 0.011271 ± 9.25509e-05 (W 4.10396e-05) 0.012338 ± 0.000132288 (W 0.000115601) 🔵 tlinalg-pool 1.09× faster than lapack-openblas dispatch-dominated (row under 20 us)
solve/n128/b1 0.119714 ± 0.000754551 (W 0.00173954) 0.153907 ± 0.000438551 (W 0.00219973) 🔵 tlinalg-pool 1.29× faster than lapack-openblas
svd_values/n128/b1 1.14043 ± 0.00341274 (W 0.00511815) 1.38743 ± 0.00429392 (W 0.00280299) 🔵 tlinalg-pool 1.22× faster than lapack-openblas
triangular_solve/n128/b1 0.00795 ± 1.70783e-05 (W 5.12379e-05) 0.006888 ± 0.000154381 (W 0.000107213) 🔴 tlinalg-pool 1.15× slower than lapack-openblas dispatch-dominated (row under 20 us)

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

case tlinalg-pool (ms) lapack-openblas (ms) vs lapack-openblas notes
cholesky/n128/b1 0.157778 ± 0.00117117 (W 0.00124498) 0.157559 ± 0.00120141 (W 0.00196674) ·
lu/n128/b1 0.319571 ± 0.00241293 (W 0.00279699) 0.234071 ± 0.00264405 (W 0.00208313) 🔴 tlinalg-pool 1.37× slower than lapack-openblas
lu_factor/n128/b1 0.225035 ± 0.000959822 (W 0.00174813) 0.1898 ± 0.00217693 (W 0.00188755) 🔴 tlinalg-pool 1.19× slower than lapack-openblas
lu_factor_solve/n128/b1 0.25971 ± 0.00116272 (W 0.0018995) 0.222325 ± 0.00197714 (W 0.00216142) 🔴 tlinalg-pool 1.17× slower than lapack-openblas
lu_solve_prepared/n128/b1 0.0336775 ± 0.000110877 (W 0.000168908) 0.0296805 ± 5.55158e-05 (W 0.000192801) 🔴 tlinalg-pool 1.13× slower than lapack-openblas
solve/n128/b1 0.283965 ± 0.00203427 (W 0.00151882) 0.237223 ± 0.000527402 (W 0.00349621) 🔴 tlinalg-pool 1.20× slower than lapack-openblas
svd_values/n128/b1 1.43802 ± 0.00518201 (W 0.00321026) 2.58498 ± 0.0105203 (W 0.00425194) 🔵 tlinalg-pool 1.80× faster than lapack-openblas
triangular_solve/n128/b1 0.020117 ± 6.62338e-05 (W 0.000168854) 0.013505 ± 0.000135277 (W 0.000132378) 🔴 tlinalg-pool 1.49× slower than lapack-openblas

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

case tlinalg-pool (ms) lapack-openblas (ms) vs lapack-openblas notes
eigh/n512/b1 27.3505 ± 0.0578615 (W 0.0230387) 23.4501 ± 0.0644224 (W 0.0242599) 🔴 tlinalg-pool 1.17× slower than lapack-openblas
eigvalsh/n512/b1 16.214 ± 0.0334409 (W 0.0146255) 11.2364 ± 0.0463387 (W 0.0178833) 🔴 tlinalg-pool 1.44× slower than lapack-openblas
householder_apply/n512/b1 1.55536 ± 0.00381001 (W 0.00347993) 9.81095 ± 0.0915604 (W 0.0213836) 🔵 tlinalg-pool 6.31× faster than lapack-openblas
householder_factor/n512/b1 1.84506 ± 0.00718713 (W 0.00492137) 6.65989 ± 0.0366841 (W 0.0176109) 🔵 tlinalg-pool 3.61× faster than lapack-openblas
qr/n512/b1 5.47599 ± 0.0131954 (W 0.0109176) 13.4853 ± 0.0715864 (W 0.0185606) 🔵 tlinalg-pool 2.46× faster than lapack-openblas
rank_revealing_qr/n512/b1 7.25994 ± 0.0136999 (W 0.00994734) 22.8718 ± 0.0101188 (W 0.0942788) 🔵 tlinalg-pool 3.15× faster than lapack-openblas
svd_thin/n512/b1 38.5701 ± 0.132366 (W 0.0307242) 84.9901 ± 0.248708 (W 0.104427) 🔵 tlinalg-pool 2.20× faster than lapack-openblas

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

case tlinalg-pool (ms) lapack-openblas (ms) vs lapack-openblas notes
eigh/n512/b1 37.0138 ± 0.0767851 (W 0.0254273) 203.375 ± 0.575511 (W 0.349192) 🔵 tlinalg-pool 5.49× faster than lapack-openblas
eigvalsh/n512/b1 20.4466 ± 0.0699934 (W 0.0195009) 29.7391 ± 0.0927531 (W 0.0309846) 🔵 tlinalg-pool 1.45× faster than lapack-openblas
householder_apply/n512/b1 5.67833 ± 0.00893137 (W 0.010198) 14.6198 ± 0.277443 (W 0.0282451) 🔵 tlinalg-pool 2.57× faster than lapack-openblas
householder_factor/n512/b1 5.26653 ± 0.0232988 (W 0.00863078) 10.7222 ± 0.024677 (W 0.0194345) 🔵 tlinalg-pool 2.04× faster than lapack-openblas
qr/n512/b1 13.3313 ± 0.0558323 (W 0.0246606) 21.7257 ± 0.0818613 (W 0.0292186) 🔵 tlinalg-pool 1.63× faster than lapack-openblas
rank_revealing_qr/n512/b1 14.3005 ± 0.0307523 (W 0.0290572) 24.3503 ± 0.0630077 (W 0.0417404) 🔵 tlinalg-pool 1.70× faster than lapack-openblas
svd_thin/n512/b1 74.9962 ± 0.197543 (W 0.116033) 97.8644 ± 0.970394 (W 0.295174) 🔵 tlinalg-pool 1.30× faster than lapack-openblas

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

case tlinalg-pool (ms) lapack-openblas (ms) vs lapack-openblas notes
householder_factor/4096x256/b1 6.02684 ± 0.0345719 (W 0.0244677) 17.9783 ± 0.115682 (W 0.0430487) 🔵 tlinalg-pool 2.98× faster than lapack-openblas
qr/4096x256/b1 21.8066 ± 0.022948 (W 0.0470504) 35.8284 ± 0.197215 (W 0.0415782) 🔵 tlinalg-pool 1.64× faster than lapack-openblas
rank_revealing_qr/4096x256/b1 21.4119 ± 0.0468023 (W 0.0632051) 39.0542 ± 0.184177 (W 0.0347143) 🔵 tlinalg-pool 1.82× faster than lapack-openblas
svd_thin/4096x256/b1 35.2304 ± 0.0406626 (W 0.0619395) 60.8544 ± 0.29715 (W 0.0903935) 🔵 tlinalg-pool 1.73× faster than lapack-openblas

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

case tlinalg-pool (ms) lapack-openblas (ms) vs lapack-openblas notes
householder_factor/4096x256/b1 17.653 ± 0.0490225 (W 0.0514272) 20.12 ± 0.0928749 (W 0.0274036) 🔵 tlinalg-pool 1.14× faster than lapack-openblas
qr/4096x256/b1 58.4649 ± 0.151674 (W 0.0781798) 40.1287 ± 0.111999 (W 0.0526735) 🔴 tlinalg-pool 1.46× slower than lapack-openblas
rank_revealing_qr/4096x256/b1 59.6647 ± 0.21807 (W 0.182363) 51.4944 ± 0.277735 (W 0.121319) 🔴 tlinalg-pool 1.16× slower than lapack-openblas
svd_thin/4096x256/b1 108.818 ± 0.224917 (W 0.276155) 85.2422 ± 0.472405 (W 0.0977869) 🔴 tlinalg-pool 1.28× slower than lapack-openblas

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.