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 1.13%, worst 97.44% among cells of 1 us or more (a sub-microsecond row is timer and scheduler jitter, not kernel time); 30 of 396 engine cells exceed 5% and carry !; median within-process relative SD 0.34%. 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 0.11816 ± 0.00144156 (W 0.00195627) 1.05378 ± 0.0047196 (W 0.00189261) 🔵 tlinalg-pool 8.92× faster than lapack-mkl
cholesky/n32/b1024 0.47261 ± 0.00202699 (W 0.0044669) 3.72177 ± 0.0276019 (W 0.014555) 🔵 tlinalg-pool 7.87× faster than lapack-mkl
cholesky/n4/b1024 0.180246 ± 0.000474068 (W 0.00124504) 0.120043 ± 0.00270533 (W 0.000985078) 🔴 tlinalg-pool 1.50× slower than lapack-mkl
cholesky/n8/b1024 0.359496 ± 0.00249846 (W 0.00145839) 0.323663 ± 0.00203801 (W 0.0017502) 🔴 tlinalg-pool 1.11× slower than lapack-mkl
eig/n16/b1024 11.8592 ± 0.0248078 (W 0.024848) 52.7487 ± 0.225565 (W 0.0587057) 🔵 tlinalg-pool 4.45× faster than lapack-mkl
eig/n32/b1024 53.2362 ± 0.0640746 (W 0.0834665) 283.232 ± 0.368463 (W 0.317788) 🔵 tlinalg-pool 5.32× faster than lapack-mkl
eig/n4/b1024 6.19311 ± 0.00872016 (W 0.00736154) 2.83519 ± 0.0100826 (W 0.0030865) 🔴 tlinalg-pool 2.18× slower than lapack-mkl
eig/n8/b1024 2.95749 ± 0.0094049 (W 0.00520947) 11.659 ± 0.0389843 (W 0.0243671) 🔵 tlinalg-pool 3.94× faster than lapack-mkl
eigh/n16/b1024 3.80283 ± 0.261475 (W 0.010128) 23.4481 ± 0.0791156 (W 0.0228858) 🔵 tlinalg-pool 6.17× faster than lapack-mkl ! between-start range over 5%
eigh/n32/b1024 13.8743 ± 0.223488 (W 0.0197439) 100.538 ± 0.420364 (W 0.23531) 🔵 tlinalg-pool 7.25× faster than lapack-mkl
eigh/n4/b1024 2.54698 ± 0.0071688 (W 0.00807891) 2.86083 ± 0.0215404 (W 0.00391321) 🔵 tlinalg-pool 1.12× faster than lapack-mkl
eigh/n8/b1024 1.11581 ± 0.0594232 (W 0.00400823) 7.76804 ± 0.0291926 (W 0.00826225) 🔵 tlinalg-pool 6.96× faster than lapack-mkl ! between-start range over 5%
eigvals/n16/b1024 8.16628 ± 0.0546839 (W 0.0188581) 36.4317 ± 0.0641599 (W 0.0198027) 🔵 tlinalg-pool 4.46× faster than lapack-mkl
eigvals/n32/b1024 36.3319 ± 0.0631985 (W 0.0883505) 169.038 ± 0.523315 (W 0.270615) 🔵 tlinalg-pool 4.65× faster than lapack-mkl
eigvals/n4/b1024 3.17788 ± 0.00890266 (W 0.00389388) 2.10349 ± 0.00433425 (W 0.00173553) 🔴 tlinalg-pool 1.51× slower than lapack-mkl
eigvals/n8/b1024 1.93278 ± 0.00821512 (W 0.00533842) 8.91508 ± 0.00535516 (W 0.0153767) 🔵 tlinalg-pool 4.61× faster than lapack-mkl
eigvalsh/n16/b1024 2.80348 ± 0.00312577 (W 0.00635393) 12.9887 ± 0.0110118 (W 0.0130209) 🔵 tlinalg-pool 4.63× faster than lapack-mkl
eigvalsh/n32/b1024 10.9147 ± 0.0307473 (W 0.0312863) 42.9908 ± 0.0844967 (W 0.0293525) 🔵 tlinalg-pool 3.94× faster than lapack-mkl
eigvalsh/n4/b1024 2.03018 ± 0.00429639 (W 0.00385414) 1.50636 ± 0.0114289 (W 0.0072123) 🔴 tlinalg-pool 1.35× slower than lapack-mkl
eigvalsh/n8/b1024 0.878141 ± 0.00275443 (W 0.00188356) 4.43659 ± 0.0190481 (W 0.0046929) 🔵 tlinalg-pool 5.05× faster than lapack-mkl
full_piv_lu/n16/b1024 0.4598 ± 0.00576516 (W 0.00595304) 3.97896 ± 0.0416646 (W 0.0132564) 🔵 tlinalg-pool 8.65× faster than lapack-mkl
full_piv_lu/n32/b1024 1.92922 ± 0.0487688 (W 0.0154352) 26.1093 ± 0.0676489 (W 0.0194422) 🔵 tlinalg-pool 13.53× faster than lapack-mkl ! between-start range over 5%
full_piv_lu/n4/b1024 0.404844 ± 0.00616303 (W 0.00306533) 0.427105 ± 0.0101113 (W 0.00187067) 🔵 tlinalg-pool 1.05× faster than lapack-mkl
full_piv_lu/n8/b1024 1.06747 ± 0.00584804 (W 0.00140735) 0.988058 ± 0.0197048 (W 0.00291438) 🔴 tlinalg-pool 1.08× slower than lapack-mkl
full_piv_lu_solve/n16/b1024 0.467441 ± 0.00635259 (W 0.00571056) 5.80022 ± 0.0177901 (W 0.023089) 🔵 tlinalg-pool 12.41× faster than lapack-mkl
full_piv_lu_solve/n32/b1024 1.58272 ± 0.0154707 (W 0.0269313) 29.5597 ± 0.0217285 (W 0.0362495) 🔵 tlinalg-pool 18.68× faster than lapack-mkl
full_piv_lu_solve/n4/b1024 0.516157 ± 0.00918342 (W 0.00369929) 1.13677 ± 0.00693131 (W 0.00747968) 🔵 tlinalg-pool 2.20× faster than lapack-mkl
full_piv_lu_solve/n8/b1024 0.16953 ± 0.00124213 (W 0.00144469) 1.97278 ± 0.0199891 (W 0.00592752) 🔵 tlinalg-pool 11.64× faster than lapack-mkl
householder_apply/n16/b1024 0.638849 ± 0.00474801 (W 0.00539493) 2.69055 ± 0.0341321 (W 0.0205845) 🔵 tlinalg-pool 4.21× faster than lapack-mkl
householder_apply/n32/b1024 2.5237 ± 0.0329392 (W 0.0241492) 13.2509 ± 0.172242 (W 0.0580387) 🔵 tlinalg-pool 5.25× faster than lapack-mkl
householder_apply/n4/b1024 0.387874 ± 0.0145873 (W 0.00176908) 0.324398 ± 0.00212146 (W 0.00188943) 🔴 tlinalg-pool 1.20× slower than lapack-mkl ! between-start range over 5%
householder_apply/n8/b1024 0.179378 ± 0.0032174 (W 0.00127903) 0.796562 ± 0.0102122 (W 0.00489705) 🔵 tlinalg-pool 4.44× faster than lapack-mkl
householder_factor/n16/b1024 0.49791 ± 0.00696704 (W 0.00238599) 2.94034 ± 0.00781064 (W 0.00342453) 🔵 tlinalg-pool 5.91× faster than lapack-mkl
householder_factor/n32/b1024 1.699 ± 0.0193508 (W 0.0109188) 12.2133 ± 0.0244821 (W 0.0180051) 🔵 tlinalg-pool 7.19× faster than lapack-mkl
householder_factor/n4/b1024 0.472704 ± 0.0121099 (W 0.00196333) 0.305203 ± 0.000654684 (W 0.00150186) 🔴 tlinalg-pool 1.55× slower than lapack-mkl ! between-start range over 5%
householder_factor/n8/b1024 0.180314 ± 0.0030229 (W 0.000989288) 0.837235 ± 0.000257387 (W 0.00186398) 🔵 tlinalg-pool 4.64× faster than lapack-mkl
lu/n16/b1024 0.381081 ± 0.00546883 (W 0.00507207) 1.6597 ± 0.0709782 (W 0.00328129) 🔵 tlinalg-pool 4.36× faster than lapack-mkl ! between-start range over 5%
lu/n32/b1024 1.32936 ± 0.00971138 (W 0.0111012) 11.8367 ± 0.0715406 (W 0.0273531) 🔵 tlinalg-pool 8.90× faster than lapack-mkl
lu/n4/b1024 0.309763 ± 0.00235548 (W 0.00208239) 0.168041 ± 0.000514188 (W 0.00120425) 🔴 tlinalg-pool 1.84× slower than lapack-mkl
lu/n8/b1024 0.865545 ± 0.00975462 (W 0.00841614) 0.51714 ± 0.00930959 (W 0.00162204) 🔴 tlinalg-pool 1.67× slower than lapack-mkl
lu_factor/n16/b1024 0.269961 ± 0.0031475 (W 0.00933431) 1.19543 ± 0.0219353 (W 0.00277083) 🔵 tlinalg-pool 4.43× faster than lapack-mkl
lu_factor/n32/b1024 0.746059 ± 0.00441663 (W 0.0103484) 8.91864 ± 0.0428401 (W 0.00970524) 🔵 tlinalg-pool 11.95× faster than lapack-mkl
lu_factor/n4/b1024 0.235572 ± 0.00100239 (W 0.00124075) 0.065116 ± 0.00163536 (W 0.000233289) 🔴 tlinalg-pool 3.62× slower than lapack-mkl ! between-start range over 5%
lu_factor/n8/b1024 0.625053 ± 0.00764757 (W 0.0043897) 0.258957 ± 0.00260471 (W 0.00196132) 🔴 tlinalg-pool 2.41× slower than lapack-mkl
lu_factor_solve/n16/b1024 0.339516 ± 0.00306921 (W 0.00383229) 2.39592 ± 0.0328189 (W 0.00663494) 🔵 tlinalg-pool 7.06× faster than lapack-mkl
lu_factor_solve/n32/b1024 0.899804 ± 0.0120415 (W 0.00897025) 13.058 ± 0.0666914 (W 0.0089373) 🔵 tlinalg-pool 14.51× faster than lapack-mkl
lu_factor_solve/n4/b1024 0.299334 ± 0.00485133 (W 0.0016378) 0.179723 ± 0.00173571 (W 0.00135435) 🔴 tlinalg-pool 1.67× slower than lapack-mkl
lu_factor_solve/n8/b1024 0.129802 ± 0.000811224 (W 0.00173819) 0.821025 ± 0.00195482 (W 0.00246347) 🔵 tlinalg-pool 6.33× faster than lapack-mkl
lu_solve_prepared/n16/b1024 0.106978 ± 0.000927392 (W 0.00131231) 1.0448 ± 0.0113723 (W 0.00278094) 🔵 tlinalg-pool 9.77× faster than lapack-mkl
lu_solve_prepared/n32/b1024 0.298284 ± 0.0125048 (W 0.00825076) 4.01292 ± 0.0333861 (W 0.0122729) 🔵 tlinalg-pool 13.45× faster than lapack-mkl ! between-start range over 5%
lu_solve_prepared/n4/b1024 0.070716 ± 0.000201561 (W 0.000140473) 0.130617 ± 0.00109725 (W 0.00139523) 🔵 tlinalg-pool 1.85× faster than lapack-mkl
lu_solve_prepared/n8/b1024 0.230517 ± 0.00120624 (W 0.00169671) 0.584804 ± 0.00185926 (W 0.00236185) 🔵 tlinalg-pool 2.54× faster than lapack-mkl
qr/n16/b1024 1.15447 ± 0.0184059 (W 0.00725262) 5.27501 ± 0.0444724 (W 0.00778422) 🔵 tlinalg-pool 4.57× faster than lapack-mkl
qr/n32/b1024 4.58438 ± 0.0133174 (W 0.0222059) 21.1541 ± 0.0707778 (W 0.0366872) 🔵 tlinalg-pool 4.61× faster than lapack-mkl
qr/n4/b1024 0.767939 ± 0.0085408 (W 0.0021626) 0.721904 ± 0.00407835 (W 0.00166761) 🔴 tlinalg-pool 1.06× slower than lapack-mkl
qr/n8/b1024 2.43052 ± 0.0479365 (W 0.00266579) 1.68659 ± 0.0230013 (W 0.00330049) 🔴 tlinalg-pool 1.44× slower than lapack-mkl
rank_revealing_qr/n16/b1024 1.35195 ± 0.0235322 (W 0.0110695) 8.3491 ± 0.0413144 (W 0.0328257) 🔵 tlinalg-pool 6.18× faster than lapack-mkl
rank_revealing_qr/n32/b1024 4.5939 ± 0.0533953 (W 0.0733823) 31.1327 ± 0.0507464 (W 0.0587301) 🔵 tlinalg-pool 6.78× faster than lapack-mkl
rank_revealing_qr/n4/b1024 1.2732 ± 0.0276169 (W 0.00222673) 1.21547 ± 0.0184932 (W 0.00288108) ·
rank_revealing_qr/n8/b1024 3.39964 ± 0.0617467 (W 0.00537711) 2.94237 ± 0.0167314 (W 0.00519293) 🔴 tlinalg-pool 1.16× slower than lapack-mkl
solve/n16/b1024 0.367439 ± 0.0234606 (W 0.00264413) 2.49184 ± 0.0643437 (W 0.00325987) 🔵 tlinalg-pool 6.78× faster than lapack-mkl ! between-start range over 5%
solve/n32/b1024 0.957036 ± 0.0103328 (W 0.0172835) 13.6251 ± 0.0732898 (W 0.0212336) 🔵 tlinalg-pool 14.24× faster than lapack-mkl
solve/n4/b1024 0.372424 ± 0.00451688 (W 0.00304182) 0.235736 ± 0.00132863 (W 0.00167166) 🔴 tlinalg-pool 1.58× slower than lapack-mkl
solve/n8/b1024 0.134049 ± 0.0650218 (W 0.00298282) 0.896969 ± 0.00988452 (W 0.00289819) 🔵 tlinalg-pool 6.69× faster than lapack-mkl ! between-start range over 5%; p position effect over 5%
svd_full/n16/b1024 7.13045 ± 0.0461361 (W 0.0722109) 51.5142 ± 0.0381934 (W 0.070309) 🔵 tlinalg-pool 7.22× faster than lapack-mkl
svd_full/n32/b1024 25.8327 ± 0.305905 (W 0.243245) 195.083 ± 0.516915 (W 0.148465) 🔵 tlinalg-pool 7.55× faster than lapack-mkl
svd_full/n4/b1024 4.41298 ± 0.0533854 (W 0.0110598) 6.19924 ± 0.00413809 (W 0.0111332) 🔵 tlinalg-pool 1.40× faster than lapack-mkl
svd_full/n8/b1024 2.08238 ± 0.144288 (W 0.0244668) 18.9269 ± 0.0191579 (W 0.0113063) 🔵 tlinalg-pool 9.09× faster than lapack-mkl ! between-start range over 5%
svd_thin/n16/b1024 7.13321 ± 0.0568622 (W 0.0404927) 51.4693 ± 0.0425392 (W 0.0566181) 🔵 tlinalg-pool 7.22× faster than lapack-mkl
svd_thin/n32/b1024 25.8222 ± 1.51538 (W 0.949076) 194.823 ± 0.813349 (W 0.13055) 🔵 tlinalg-pool 7.54× faster than lapack-mkl ! between-start range over 5%
svd_thin/n4/b1024 4.36568 ± 0.0547672 (W 0.00571141) 6.24945 ± 0.0215901 (W 0.00856845) 🔵 tlinalg-pool 1.43× faster than lapack-mkl
svd_thin/n8/b1024 2.1003 ± 0.0798912 (W 0.0204702) 18.9087 ± 0.0257276 (W 0.0152622) 🔵 tlinalg-pool 9.00× faster than lapack-mkl ! between-start range over 5%
svd_values/n16/b1024 5.13785 ± 0.0102969 (W 0.006938) 19.2346 ± 0.0618029 (W 0.00862122) 🔵 tlinalg-pool 3.74× faster than lapack-mkl
svd_values/n32/b1024 19.9715 ± 0.0331836 (W 0.0291881) 64.6301 ± 0.239906 (W 0.0669885) 🔵 tlinalg-pool 3.24× faster than lapack-mkl
svd_values/n4/b1024 3.17163 ± 0.0160105 (W 0.00618161) 2.67119 ± 0.00256029 (W 0.00554454) 🔴 tlinalg-pool 1.19× slower than lapack-mkl
svd_values/n8/b1024 1.43233 ± 0.00498556 (W 0.0039559) 6.72611 ± 0.0257945 (W 0.00735482) 🔵 tlinalg-pool 4.70× faster than lapack-mkl
triangular_solve/n16/b1024 0.079583 ± 0.00251456 (W 0.000324752) 0.602336 ± 0.0039071 (W 0.00424021) 🔵 tlinalg-pool 7.57× faster than lapack-mkl ! between-start range over 5%
triangular_solve/n32/b1024 0.211118 ± 0.00230304 (W 0.00493161) 2.65688 ± 0.0266847 (W 0.00486194) 🔵 tlinalg-pool 12.58× faster than lapack-mkl
triangular_solve/n4/b1024 0.093213 ± 0.000185221 (W 0.000694343) 0.312501 ± 0.0131218 (W 0.00337833) 🔵 tlinalg-pool 3.35× faster than lapack-mkl ! between-start range over 5%
triangular_solve/n8/b1024 0.219507 ± 0.000418277 (W 0.00164106) 0.410702 ± 0.00208294 (W 0.00294179) 🔵 tlinalg-pool 1.87× 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 0.165167 ± 0.00285707 (W 0.00257786) 1.79895 ± 0.00564825 (W 0.00795481) 🔵 tlinalg-pool 10.89× faster than lapack-mkl
cholesky/n32/b1024 0.792058 ± 0.00559249 (W 0.00831354) 8.82948 ± 0.0747306 (W 0.0192738) 🔵 tlinalg-pool 11.15× faster than lapack-mkl
cholesky/n4/b1024 0.201809 ± 0.000722179 (W 0.00142189) 0.177214 ± 0.000544343 (W 0.00109625) 🔴 tlinalg-pool 1.14× slower than lapack-mkl
cholesky/n8/b1024 0.416632 ± 0.0030258 (W 0.00261668) 0.472433 ± 0.00227097 (W 0.00179459) 🔵 tlinalg-pool 1.13× faster than lapack-mkl
eig/n16/b1024 16.3237 ± 0.0406195 (W 0.0341502) 174.913 ± 0.345452 (W 0.22987) 🔵 tlinalg-pool 10.72× faster than lapack-mkl
eig/n32/b1024 67.1865 ± 0.0218717 (W 0.185408) 790.948 ± 1.26925 (W 2.09175) 🔵 tlinalg-pool 11.77× faster than lapack-mkl
eig/n4/b1024 10.3213 ± 0.0545377 (W 0.00973861) 12.424 ± 0.0121179 (W 0.00518603) 🔵 tlinalg-pool 1.20× faster than lapack-mkl
eig/n8/b1024 4.41723 ± 0.0307641 (W 0.00730629) 43.1874 ± 0.0757051 (W 0.0397118) 🔵 tlinalg-pool 9.78× faster than lapack-mkl
eigh/n16/b1024 4.13608 ± 0.175486 (W 0.0657377) 32.0976 ± 0.0715638 (W 0.016766) 🔵 tlinalg-pool 7.76× faster than lapack-mkl ! between-start range over 5%
eigh/n32/b1024 17.4418 ± 0.288005 (W 0.064869) 151.171 ± 0.307535 (W 0.0721474) 🔵 tlinalg-pool 8.67× faster than lapack-mkl
eigh/n4/b1024 2.89691 ± 0.0120012 (W 0.00600325) 2.84663 ± 0.00842104 (W 0.00349538) ·
eigh/n8/b1024 1.17594 ± 0.0225432 (W 0.0175849) 8.94099 ± 0.037576 (W 0.00627186) 🔵 tlinalg-pool 7.60× faster than lapack-mkl
eigvals/n16/b1024 9.73352 ± 0.0148217 (W 0.00706188) 124.671 ± 0.202277 (W 0.0924196) 🔵 tlinalg-pool 12.81× faster than lapack-mkl
eigvals/n32/b1024 42.5667 ± 0.0209696 (W 0.0249829) 521.974 ± 0.841399 (W 0.580095) 🔵 tlinalg-pool 12.26× faster than lapack-mkl
eigvals/n4/b1024 4.99917 ± 0.0563388 (W 0.00781479) 9.54806 ± 0.00644153 (W 0.00605725) 🔵 tlinalg-pool 1.91× faster than lapack-mkl
eigvals/n8/b1024 2.51759 ± 0.00813904 (W 0.00699654) 34.2088 ± 0.0279121 (W 0.0233198) 🔵 tlinalg-pool 13.59× faster than lapack-mkl
eigvalsh/n16/b1024 3.01117 ± 0.0128937 (W 0.0115094) 18.299 ± 0.016168 (W 0.0194512) 🔵 tlinalg-pool 6.08× faster than lapack-mkl
eigvalsh/n32/b1024 11.7829 ± 0.0235692 (W 0.0231587) 77.0694 ± 0.0434207 (W 0.0371674) 🔵 tlinalg-pool 6.54× faster than lapack-mkl
eigvalsh/n4/b1024 2.36142 ± 0.0200882 (W 0.00767829) 1.79692 ± 0.0152792 (W 0.00159269) 🔴 tlinalg-pool 1.31× slower than lapack-mkl
eigvalsh/n8/b1024 0.931736 ± 0.00136474 (W 0.00303744) 5.53719 ± 0.00987975 (W 0.00642559) 🔵 tlinalg-pool 5.94× faster than lapack-mkl
full_piv_lu/n16/b1024 0.612035 ± 0.00216031 (W 0.0152629) 5.41381 ± 0.047409 (W 0.0176671) 🔵 tlinalg-pool 8.85× faster than lapack-mkl
full_piv_lu/n32/b1024 2.97553 ± 0.0224426 (W 0.0214036) 27.0369 ± 0.0307568 (W 0.0345904) 🔵 tlinalg-pool 9.09× faster than lapack-mkl
full_piv_lu/n4/b1024 0.462672 ± 0.00499384 (W 0.00298534) 0.527087 ± 0.00753213 (W 0.00276861) 🔵 tlinalg-pool 1.14× faster than lapack-mkl
full_piv_lu/n8/b1024 0.189966 ± 0.00371953 (W 0.00514785) 1.29344 ± 0.0104595 (W 0.00271855) 🔵 tlinalg-pool 6.81× faster than lapack-mkl
full_piv_lu_solve/n16/b1024 0.707432 ± 0.0110544 (W 0.0113343) 8.44327 ± 0.00757532 (W 0.0145129) 🔵 tlinalg-pool 11.94× faster than lapack-mkl
full_piv_lu_solve/n32/b1024 2.63965 ± 0.0172518 (W 0.0249751) 34.4224 ± 0.0733062 (W 0.0462211) 🔵 tlinalg-pool 13.04× faster than lapack-mkl
full_piv_lu_solve/n4/b1024 0.77799 ± 0.00371637 (W 0.0047459) 1.60625 ± 0.00644341 (W 0.00653963) 🔵 tlinalg-pool 2.06× faster than lapack-mkl
full_piv_lu_solve/n8/b1024 0.267145 ± 0.00187 (W 0.00307652) 3.20344 ± 0.0125906 (W 0.00382873) 🔵 tlinalg-pool 11.99× faster than lapack-mkl
householder_apply/n16/b1024 0.635574 ± 0.00595288 (W 0.00577262) 6.13049 ± 0.066583 (W 0.0184729) 🔵 tlinalg-pool 9.65× faster than lapack-mkl
householder_apply/n32/b1024 3.57131 ± 0.0442559 (W 0.034025) 37.7888 ± 0.277852 (W 0.0435326) 🔵 tlinalg-pool 10.58× faster than lapack-mkl
householder_apply/n4/b1024 0.357461 ± 0.00630281 (W 0.00237194) 0.41537 ± 0.00422641 (W 0.00284802) 🔵 tlinalg-pool 1.16× faster than lapack-mkl
householder_apply/n8/b1024 0.154496 ± 0.00207461 (W 0.00195847) 1.24156 ± 0.0521208 (W 0.00571554) 🔵 tlinalg-pool 8.04× faster than lapack-mkl ! between-start range over 5%
householder_factor/n16/b1024 0.624251 ± 0.0146723 (W 0.0111572) 6.30446 ± 0.0156077 (W 0.00668291) 🔵 tlinalg-pool 10.10× faster than lapack-mkl ! between-start range over 5%
householder_factor/n32/b1024 2.62336 ± 0.0444875 (W 0.0338397) 27.6297 ± 0.0376572 (W 0.0193492) 🔵 tlinalg-pool 10.53× faster than lapack-mkl
householder_factor/n4/b1024 0.567893 ± 0.00329165 (W 0.00202445) 0.513624 ± 0.000882723 (W 0.00176218) 🔴 tlinalg-pool 1.11× slower than lapack-mkl
householder_factor/n8/b1024 0.203637 ± 0.00113619 (W 0.00177262) 1.47712 ± 0.000858516 (W 0.00183881) 🔵 tlinalg-pool 7.25× faster than lapack-mkl
lu/n16/b1024 0.45428 ± 0.00503117 (W 0.00881368) 3.57882 ± 0.0232788 (W 0.010662) 🔵 tlinalg-pool 7.88× faster than lapack-mkl
lu/n32/b1024 2.12015 ± 0.0201004 (W 0.0104729) 19.654 ± 0.0639959 (W 0.0302205) 🔵 tlinalg-pool 9.27× faster than lapack-mkl
lu/n4/b1024 0.39222 ± 0.00551881 (W 0.00277203) 0.281026 ± 0.00195863 (W 0.00215989) 🔴 tlinalg-pool 1.40× slower than lapack-mkl
lu/n8/b1024 0.152397 ± 0.00152253 (W 0.00307445) 0.89525 ± 0.0192834 (W 0.00204753) 🔵 tlinalg-pool 5.87× faster than lapack-mkl
lu_factor/n16/b1024 0.310295 ± 0.0035677 (W 0.0120439) 2.64562 ± 0.0190857 (W 0.00434134) 🔵 tlinalg-pool 8.53× faster than lapack-mkl
lu_factor/n32/b1024 1.3051 ± 0.0200737 (W 0.0127557) 15.4033 ± 0.0790734 (W 0.0177741) 🔵 tlinalg-pool 11.80× faster than lapack-mkl
lu_factor/n4/b1024 0.300873 ± 0.00270131 (W 0.00235366) 0.182868 ± 0.00038405 (W 0.00121939) 🔴 tlinalg-pool 1.65× slower than lapack-mkl
lu_factor/n8/b1024 0.114873 ± 0.000564835 (W 0.00153982) 0.619419 ± 0.00216494 (W 0.00221938) 🔵 tlinalg-pool 5.39× faster than lapack-mkl
lu_factor_solve/n16/b1024 0.476079 ± 0.00382671 (W 0.00251843) 5.28099 ± 0.0182198 (W 0.00355848) 🔵 tlinalg-pool 11.09× faster than lapack-mkl
lu_factor_solve/n32/b1024 1.80251 ± 0.0182351 (W 0.0187586) 20.667 ± 0.0222673 (W 0.0203899) 🔵 tlinalg-pool 11.47× faster than lapack-mkl
lu_factor_solve/n4/b1024 0.456411 ± 0.00179545 (W 0.00210495) 0.415648 ± 0.000620113 (W 0.00180195) 🔴 tlinalg-pool 1.10× slower than lapack-mkl
lu_factor_solve/n8/b1024 0.172934 ± 0.0015887 (W 0.00208106) 1.66176 ± 0.00418885 (W 0.00141049) 🔵 tlinalg-pool 9.61× faster than lapack-mkl
lu_solve_prepared/n16/b1024 0.198777 ± 0.00356605 (W 0.00365503) 2.59259 ± 0.0160642 (W 0.00540144) 🔵 tlinalg-pool 13.04× faster than lapack-mkl
lu_solve_prepared/n32/b1024 0.922353 ± 0.0117793 (W 0.0133873) 4.99463 ± 0.0370721 (W 0.0132105) 🔵 tlinalg-pool 5.42× faster than lapack-mkl
lu_solve_prepared/n4/b1024 0.166223 ± 0.00176096 (W 0.00164584) 0.240449 ± 0.000495363 (W 0.00153186) 🔵 tlinalg-pool 1.45× faster than lapack-mkl
lu_solve_prepared/n8/b1024 0.087873 ± 0.000620549 (W 0.000766418) 1.06903 ± 0.00322176 (W 0.00286678) 🔵 tlinalg-pool 12.17× faster than lapack-mkl
qr/n16/b1024 1.40608 ± 0.00130453 (W 0.00633763) 10.511 ± 0.0420574 (W 0.0142856) 🔵 tlinalg-pool 7.48× faster than lapack-mkl
qr/n32/b1024 9.40335 ± 0.0604833 (W 0.0380481) 49.4689 ± 0.18023 (W 0.019917) 🔵 tlinalg-pool 5.26× faster than lapack-mkl
qr/n4/b1024 0.871581 ± 0.00519215 (W 0.00234312) 0.932743 ± 0.0156053 (W 0.00248417) 🔵 tlinalg-pool 1.07× faster than lapack-mkl
qr/n8/b1024 0.342309 ± 0.00394485 (W 0.00200921) 2.65537 ± 0.00393418 (W 0.00751371) 🔵 tlinalg-pool 7.76× faster than lapack-mkl
rank_revealing_qr/n16/b1024 1.59635 ± 0.0733241 (W 0.0114495) 15.8155 ± 0.0430919 (W 0.0160928) 🔵 tlinalg-pool 9.91× faster than lapack-mkl ! between-start range over 5%
rank_revealing_qr/n32/b1024 8.62972 ± 0.045571 (W 0.0325646) 69.1725 ± 0.292239 (W 0.0593122) 🔵 tlinalg-pool 8.02× faster than lapack-mkl
rank_revealing_qr/n4/b1024 1.48392 ± 0.030017 (W 0.00226129) 1.78756 ± 0.0060614 (W 0.00475047) 🔵 tlinalg-pool 1.20× faster than lapack-mkl
rank_revealing_qr/n8/b1024 0.49468 ± 0.00856051 (W 0.00368858) 4.79864 ± 0.0119172 (W 0.00706367) 🔵 tlinalg-pool 9.70× faster than lapack-mkl
solve/n16/b1024 0.529887 ± 0.0151663 (W 0.0115682) 5.54956 ± 0.0107933 (W 0.00570418) 🔵 tlinalg-pool 10.47× faster than lapack-mkl ! between-start range over 5%
solve/n32/b1024 1.88574 ± 0.0140666 (W 0.00952354) 21.7089 ± 0.0713661 (W 0.0331966) 🔵 tlinalg-pool 11.51× faster than lapack-mkl
solve/n4/b1024 0.570728 ± 0.00671252 (W 0.00265192) 0.475144 ± 0.00190779 (W 0.00237915) 🔴 tlinalg-pool 1.20× slower than lapack-mkl
solve/n8/b1024 0.194761 ± 0.00655113 (W 0.00269998) 1.74153 ± 0.00899743 (W 0.00177228) 🔵 tlinalg-pool 8.94× faster than lapack-mkl ! between-start range over 5%
svd_full/n16/b1024 7.71446 ± 0.0646465 (W 0.025055) 67.3185 ± 0.0908306 (W 0.217786) 🔵 tlinalg-pool 8.73× faster than lapack-mkl
svd_full/n32/b1024 34.6566 ± 0.806786 (W 0.170511) 285.181 ± 0.622173 (W 0.295084) 🔵 tlinalg-pool 8.23× faster than lapack-mkl ! between-start range over 5%
svd_full/n4/b1024 5.25212 ± 0.0284442 (W 0.00326105) 7.57654 ± 0.0207038 (W 0.00973944) 🔵 tlinalg-pool 1.44× faster than lapack-mkl
svd_full/n8/b1024 2.31814 ± 0.0875153 (W 0.0352854) 22.5428 ± 0.0371875 (W 0.010712) 🔵 tlinalg-pool 9.72× faster than lapack-mkl ! between-start range over 5%
svd_thin/n16/b1024 7.80754 ± 0.0856146 (W 0.0236423) 67.2165 ± 0.141455 (W 0.033909) 🔵 tlinalg-pool 8.61× faster than lapack-mkl
svd_thin/n32/b1024 33.3068 ± 0.414238 (W 0.113304) 285.872 ± 0.787426 (W 0.233632) 🔵 tlinalg-pool 8.58× faster than lapack-mkl
svd_thin/n4/b1024 5.2533 ± 0.0124963 (W 0.00446239) 7.55285 ± 0.00891062 (W 0.00868025) 🔵 tlinalg-pool 1.44× faster than lapack-mkl
svd_thin/n8/b1024 2.18021 ± 0.0819182 (W 0.0238073) 22.5419 ± 0.0794562 (W 0.0146204) 🔵 tlinalg-pool 10.34× faster than lapack-mkl ! between-start range over 5%
svd_values/n16/b1024 5.80273 ± 0.0739596 (W 0.0082379) 28.1506 ± 0.0398454 (W 0.0234935) 🔵 tlinalg-pool 4.85× faster than lapack-mkl
svd_values/n32/b1024 22.4645 ± 0.0531168 (W 0.0320753) 105.38 ± 0.0694728 (W 0.0543694) 🔵 tlinalg-pool 4.69× faster than lapack-mkl
svd_values/n4/b1024 4.05353 ± 0.00793004 (W 0.00341962) 3.54829 ± 0.0102753 (W 0.00464115) 🔴 tlinalg-pool 1.14× slower than lapack-mkl
svd_values/n8/b1024 1.67877 ± 0.00930833 (W 0.00215997) 9.08353 ± 0.0122686 (W 0.0101364) 🔵 tlinalg-pool 5.41× faster than lapack-mkl
triangular_solve/n16/b1024 0.138837 ± 0.00110373 (W 0.00134077) 1.39141 ± 0.0174466 (W 0.00498318) 🔵 tlinalg-pool 10.02× faster than lapack-mkl
triangular_solve/n32/b1024 0.615366 ± 0.00212721 (W 0.0159747) 3.55413 ± 0.00715131 (W 0.0168041) 🔵 tlinalg-pool 5.78× faster than lapack-mkl
triangular_solve/n4/b1024 0.165195 ± 0.000878523 (W 0.00131313) 0.302554 ± 0.00159831 (W 0.00170742) 🔵 tlinalg-pool 1.83× faster than lapack-mkl
triangular_solve/n8/b1024 0.065862 ± 0.000385386 (W 0.000682452) 0.608469 ± 0.00254983 (W 0.00356578) 🔵 tlinalg-pool 9.24× 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 0.065301 ± 0.00188863 (W 0.000150734) 0.092186 ± 0.000448718 (W 0.000141155) 🔵 tlinalg-pool 1.41× faster than lapack-mkl ! between-start range over 5%
lu/n128/b1 0.195353 ± 0.00159333 (W 0.00406133) 0.135059 ± 0.000443082 (W 0.00179415) 🔴 tlinalg-pool 1.45× slower than lapack-mkl
lu_factor/n128/b1 0.118199 ± 0.00310637 (W 0.00438064) 0.120438 ± 0.000439928 (W 0.00146366) · ! between-start range over 5%
lu_factor_solve/n128/b1 0.12412 ± 0.00201912 (W 0.00226076) 0.143707 ± 0.00044106 (W 0.000999988) 🔵 tlinalg-pool 1.16× faster than lapack-mkl
lu_solve_prepared/n128/b1 0.010289 ± 3.41565e-05 (W 8.55951e-05) 0.015764 ± 0.000130449 (W 0.000157886) 🔵 tlinalg-pool 1.53× faster than lapack-mkl dispatch-dominated (row under 20 us)
solve/n128/b1 0.138081 ± 0.00102486 (W 0.00307723) 0.149757 ± 0.000615386 (W 0.00191584) 🔵 tlinalg-pool 1.08× faster than lapack-mkl
svd_values/n128/b1 2.37647 ± 0.00188902 (W 0.00261572) 1.14152 ± 0.00169936 (W 0.00256725) 🔴 tlinalg-pool 2.08× slower than lapack-mkl
triangular_solve/n128/b1 0.007138 ± 9.10586e-05 (W 9.86153e-05) 0.010119 ± 0.000365594 (W 0.000216709) 🔵 tlinalg-pool 1.42× faster than lapack-mkl ! between-start range over 5%; 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 0.187959 ± 0.00152285 (W 0.00280749) 0.409348 ± 0.00199171 (W 0.00307837) 🔵 tlinalg-pool 2.18× faster than lapack-mkl
lu/n128/b1 0.350141 ± 0.00335503 (W 0.00444658) 0.450431 ± 0.0018315 (W 0.00264609) 🔵 tlinalg-pool 1.29× faster than lapack-mkl
lu_factor/n128/b1 0.271012 ± 0.00312387 (W 0.00249348) 0.417436 ± 0.00120914 (W 0.00278952) 🔵 tlinalg-pool 1.54× faster than lapack-mkl
lu_factor_solve/n128/b1 0.305878 ± 0.00243842 (W 0.00515817) 0.448104 ± 0.000684027 (W 0.00214974) 🔵 tlinalg-pool 1.46× faster than lapack-mkl
lu_solve_prepared/n128/b1 0.031754 ± 0.000109479 (W 0.000110076) 0.0252815 ± 6.70342e-05 (W 0.000140717) 🔴 tlinalg-pool 1.26× slower than lapack-mkl
solve/n128/b1 0.327182 ± 0.00116895 (W 0.00451576) 0.463969 ± 0.00164177 (W 0.00195804) 🔵 tlinalg-pool 1.42× faster than lapack-mkl
svd_values/n128/b1 3.26103 ± 0.00638476 (W 0.0127987) 1.92164 ± 0.00986727 (W 0.00543672) 🔴 tlinalg-pool 1.70× slower than lapack-mkl
triangular_solve/n128/b1 0.0187645 ± 8.97144e-05 (W 0.000104229) 0.012954 ± 5.19615e-05 (W 9.07845e-05) 🔴 tlinalg-pool 1.45× slower than lapack-mkl dispatch-dominated (row under 20 us)

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

case tlinalg-pool (ms) lapack-mkl (ms) vs lapack-mkl notes
eigh/n512/b1 29.1749 ± 0.072167 (W 0.0794185) 14.9922 ± 0.0328072 (W 0.0422729) 🔴 tlinalg-pool 1.95× slower than lapack-mkl
eigvalsh/n512/b1 17.6522 ± 0.0473387 (W 0.0258605) 12.6446 ± 0.0229647 (W 0.0116131) 🔴 tlinalg-pool 1.40× slower than lapack-mkl
householder_apply/n512/b1 1.76002 ± 0.00316849 (W 0.00497625) 3.14314 ± 0.0472926 (W 0.0198665) 🔵 tlinalg-pool 1.79× faster than lapack-mkl
householder_factor/n512/b1 2.08716 ± 0.13318 (W 0.0181278) 2.17973 ± 0.00396754 (W 0.00507886) · ! between-start range over 5%
qr/n512/b1 6.06002 ± 0.00523791 (W 0.009003) 8.55763 ± 0.0276456 (W 0.0167526) 🔵 tlinalg-pool 1.41× faster than lapack-mkl
rank_revealing_qr/n512/b1 9.02999 ± 0.0274011 (W 0.0255858) 24.7301 ± 0.108759 (W 0.0946693) 🔵 tlinalg-pool 2.74× faster than lapack-mkl
svd_thin/n512/b1 59.7688 ± 0.244838 (W 0.0572341) 48.9218 ± 0.0930879 (W 0.064136) 🔴 tlinalg-pool 1.22× slower than lapack-mkl

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

case tlinalg-pool (ms) lapack-mkl (ms) vs lapack-mkl notes
eigh/n512/b1 40.4349 ± 0.43125 (W 0.365728) 65.5663 ± 0.169379 (W 0.0720538) 🔵 tlinalg-pool 1.62× faster than lapack-mkl
eigvalsh/n512/b1 23.8219 ± 0.065593 (W 0.0549455) 28.8931 ± 0.0583789 (W 0.0224125) 🔵 tlinalg-pool 1.21× faster than lapack-mkl
householder_apply/n512/b1 5.6775 ± 0.0160003 (W 0.00859838) 8.4166 ± 0.137729 (W 0.0111147) 🔵 tlinalg-pool 1.48× faster than lapack-mkl
householder_factor/n512/b1 5.63332 ± 0.00650171 (W 0.00728457) 5.63259 ± 0.00342713 (W 0.00580989) ·
qr/n512/b1 14.8809 ± 0.0075053 (W 0.0262273) 15.3025 ± 0.0174279 (W 0.0263164) ·
rank_revealing_qr/n512/b1 16.8189 ± 0.0545835 (W 0.0258467) 60.5837 ± 0.0918355 (W 0.154234) 🔵 tlinalg-pool 3.60× faster than lapack-mkl
svd_thin/n512/b1 95.3558 ± 0.102113 (W 0.287231) 85.3061 ± 0.760821 (W 0.0874368) 🔴 tlinalg-pool 1.12× slower 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 6.97688 ± 0.0143854 (W 0.0257235) 6.18701 ± 0.0197274 (W 0.035676) 🔴 tlinalg-pool 1.13× slower than lapack-mkl
qr/4096x256/b1 23.4463 ± 0.0654522 (W 0.0534801) 18.2969 ± 0.0507366 (W 0.0441115) 🔴 tlinalg-pool 1.28× slower than lapack-mkl
rank_revealing_qr/4096x256/b1 24.5803 ± 0.0972784 (W 0.0829737) 61.0475 ± 2.0044 (W 0.944583) 🔵 tlinalg-pool 2.48× faster than lapack-mkl ! between-start range over 5%
svd_thin/4096x256/b1 40.7083 ± 0.16389 (W 0.0682436) 34.287 ± 0.239118 (W 0.0739752) 🔴 tlinalg-pool 1.19× slower than lapack-mkl

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

case tlinalg-pool (ms) lapack-mkl (ms) vs lapack-mkl notes
householder_factor/4096x256/b1 18.7655 ± 0.037154 (W 0.0373203) 17.0212 ± 0.141927 (W 0.159903) 🔴 tlinalg-pool 1.10× slower than lapack-mkl
qr/4096x256/b1 55.4432 ± 0.181878 (W 0.903627) 40.1698 ± 0.173905 (W 0.177043) 🔴 tlinalg-pool 1.38× slower than lapack-mkl
rank_revealing_qr/4096x256/b1 61.5753 ± 0.700023 (W 0.876444) 161.578 ± 1.06051 (W 0.520752) 🔵 tlinalg-pool 2.62× faster than lapack-mkl
svd_thin/4096x256/b1 110.764 ± 0.498402 (W 0.981802) 73.6806 ± 0.415224 (W 0.309084) 🔴 tlinalg-pool 1.50× 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.