tlinalg-scaling 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 1.57%, worst 12.05% among cells of 1 us or more (a sub-microsecond row is timer and scheduler jitter, not kernel time); 27 of 396 engine cells exceed 5% and carry !; median within-process relative SD 0.32%. A cell whose ±, ! or p is as large as the difference you are reading is not evidence of a regression. The vs tlinalg-1lane column is the rotation-averaged ratio exp(mean over starts of log(engine median / tlinalg-1lane 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-1lane (ms) tlinalg-pool (ms) vs tlinalg-pool notes
cholesky/n16/b1024 0.855849 ± 0.0291032 (W 0.00195178) 0.115596 ± 0.00297433 (W 0.00131191) 🔴 tlinalg-1lane 7.40× slower than tlinalg-pool ! between-start range over 5%
cholesky/n32/b1024 3.20277 ± 0.0344566 (W 0.0222837) 0.477874 ± 0.0236719 (W 0.0055627) 🔴 tlinalg-1lane 6.70× slower than tlinalg-pool ! between-start range over 5%
cholesky/n4/b1024 0.17919 ± 0.000665212 (W 0.00162742) 0.179998 ± 0.000707896 (W 0.00154636) ·
cholesky/n8/b1024 0.359067 ± 0.00779763 (W 0.00249448) 0.046963 ± 0.00113275 (W 0.000118693) 🔴 tlinalg-1lane 7.65× slower than tlinalg-pool
eig/n16/b1024 96.3224 ± 0.232625 (W 0.148996) 12.0664 ± 0.0226326 (W 0.0155419) 🔴 tlinalg-1lane 7.98× slower than tlinalg-pool
eig/n32/b1024 431.294 ± 1.31978 (W 2.26262) 54.1353 ± 0.161072 (W 0.0599572) 🔴 tlinalg-1lane 7.97× slower than tlinalg-pool
eig/n4/b1024 6.40936 ± 0.0384349 (W 0.0144335) 0.814918 ± 0.00734586 (W 0.0039772) 🔴 tlinalg-1lane 7.87× slower than tlinalg-pool
eig/n8/b1024 24.1055 ± 0.0553928 (W 0.0354991) 3.02484 ± 0.0150027 (W 0.00528488) 🔴 tlinalg-1lane 7.97× slower than tlinalg-pool
eigh/n16/b1024 30.0237 ± 0.142752 (W 0.0199264) 3.78373 ± 0.0102178 (W 0.00674009) 🔴 tlinalg-1lane 7.93× slower than tlinalg-pool
eigh/n32/b1024 111.788 ± 0.262223 (W 0.244854) 13.9812 ± 0.0605025 (W 0.0152887) 🔴 tlinalg-1lane 8.00× slower than tlinalg-pool
eigh/n4/b1024 2.53722 ± 0.0185145 (W 0.00557133) 0.322693 ± 0.00247533 (W 0.00180856) 🔴 tlinalg-1lane 7.86× slower than tlinalg-pool
eigh/n8/b1024 8.69073 ± 0.084301 (W 0.0150905) 1.0969 ± 0.0086676 (W 0.00140577) 🔴 tlinalg-1lane 7.92× slower than tlinalg-pool
eigvals/n16/b1024 66.6984 ± 0.184383 (W 0.0455893) 8.40082 ± 0.0140545 (W 0.00504718) 🔴 tlinalg-1lane 7.94× slower than tlinalg-pool
eigvals/n32/b1024 292.463 ± 0.431381 (W 0.576801) 36.7273 ± 0.105627 (W 0.191188) 🔴 tlinalg-1lane 7.96× slower than tlinalg-pool
eigvals/n4/b1024 3.42968 ± 0.0369116 (W 0.00700129) 0.440434 ± 0.00255486 (W 0.00317691) 🔴 tlinalg-1lane 7.79× slower than tlinalg-pool
eigvals/n8/b1024 15.7784 ± 0.0619819 (W 0.0294281) 1.99883 ± 0.00683196 (W 0.0042415) 🔴 tlinalg-1lane 7.89× slower than tlinalg-pool
eigvalsh/n16/b1024 22.7451 ± 0.153439 (W 0.0198078) 2.85937 ± 0.0131626 (W 0.00274532) 🔴 tlinalg-1lane 7.95× slower than tlinalg-pool
eigvalsh/n32/b1024 85.1785 ± 0.241138 (W 0.0809479) 10.6777 ± 0.0541594 (W 0.00634604) 🔴 tlinalg-1lane 7.98× slower than tlinalg-pool
eigvalsh/n4/b1024 2.08059 ± 0.00260483 (W 0.00224398) 0.263557 ± 0.000780352 (W 0.00321034) 🔴 tlinalg-1lane 7.89× slower than tlinalg-pool
eigvalsh/n8/b1024 6.9658 ± 0.0236987 (W 0.00678945) 0.883954 ± 0.00519868 (W 0.00158982) 🔴 tlinalg-1lane 7.88× slower than tlinalg-pool
full_piv_lu/n16/b1024 3.89463 ± 0.0309431 (W 0.0158966) 0.512506 ± 0.0053001 (W 0.0046377) 🔴 tlinalg-1lane 7.60× slower than tlinalg-pool
full_piv_lu/n32/b1024 14.9886 ± 0.151486 (W 0.0253562) 2.0321 ± 0.0125992 (W 0.0168606) 🔴 tlinalg-1lane 7.38× slower than tlinalg-pool
full_piv_lu/n4/b1024 0.492873 ± 0.00424397 (W 0.00365398) 0.493069 ± 0.00219258 (W 0.00347171) ·
full_piv_lu/n8/b1024 1.29494 ± 0.0127305 (W 0.0103216) 0.181891 ± 0.00180557 (W 0.00207077) 🔴 tlinalg-1lane 7.12× slower than tlinalg-pool
full_piv_lu_solve/n16/b1024 3.97782 ± 0.0419627 (W 0.0106712) 0.518838 ± 0.0104859 (W 0.00271659) 🔴 tlinalg-1lane 7.67× slower than tlinalg-pool
full_piv_lu_solve/n32/b1024 12.7556 ± 0.22577 (W 0.071261) 1.63829 ± 0.0144249 (W 0.0170708) 🔴 tlinalg-1lane 7.79× slower than tlinalg-pool
full_piv_lu_solve/n4/b1024 0.643456 ± 0.00153958 (W 0.00375278) 0.085125 ± 0.00068985 (W 0.00205518) 🔴 tlinalg-1lane 7.56× slower than tlinalg-pool
full_piv_lu_solve/n8/b1024 1.58164 ± 0.0186299 (W 0.0103959) 0.205722 ± 0.00048924 (W 0.00152377) 🔴 tlinalg-1lane 7.69× slower than tlinalg-pool
householder_apply/n16/b1024 4.90249 ± 0.146096 (W 0.00771353) 0.610801 ± 0.0190718 (W 0.00305887) 🔴 tlinalg-1lane 8.03× slower than tlinalg-pool ! between-start range over 5%
householder_apply/n32/b1024 20.3604 ± 0.786062 (W 0.04494) 2.46243 ± 0.0650348 (W 0.0103442) 🔴 tlinalg-1lane 8.27× slower than tlinalg-pool ! between-start range over 5%
householder_apply/n4/b1024 0.402465 ± 0.00935227 (W 0.00233096) 0.0512605 ± 0.00121339 (W 0.000251588) 🔴 tlinalg-1lane 7.85× slower than tlinalg-pool ! between-start range over 5%
householder_apply/n8/b1024 1.3187 ± 0.0650999 (W 0.00368994) 0.166221 ± 0.00555407 (W 0.0015534) 🔴 tlinalg-1lane 7.93× slower than tlinalg-pool ! between-start range over 5%
householder_factor/n16/b1024 4.02627 ± 0.061204 (W 0.00458937) 0.511219 ± 0.00918973 (W 0.00217504) 🔴 tlinalg-1lane 7.88× slower than tlinalg-pool
householder_factor/n32/b1024 13.4184 ± 0.308597 (W 0.0424563) 1.67912 ± 0.0257903 (W 0.00485654) 🔴 tlinalg-1lane 7.99× slower than tlinalg-pool ! between-start range over 5%
householder_factor/n4/b1024 0.489 ± 0.00325598 (W 0.00205088) 0.0639655 ± 0.000667716 (W 0.000332418) 🔴 tlinalg-1lane 7.64× slower than tlinalg-pool
householder_factor/n8/b1024 1.30977 ± 0.0226862 (W 0.00249652) 0.168345 ± 0.00267909 (W 0.000867601) 🔴 tlinalg-1lane 7.78× slower than tlinalg-pool
lu/n16/b1024 2.91453 ± 0.0350465 (W 0.0142099) 0.382903 ± 0.00483953 (W 0.00258095) 🔴 tlinalg-1lane 7.61× slower than tlinalg-pool
lu/n32/b1024 10.3944 ± 0.1278 (W 0.0373975) 1.35435 ± 0.00344806 (W 0.00811035) 🔴 tlinalg-1lane 7.67× slower than tlinalg-pool
lu/n4/b1024 0.305559 ± 0.0035299 (W 0.00294347) 0.304856 ± 0.00308516 (W 0.0033308) ·
lu/n8/b1024 0.878233 ± 0.0130956 (W 0.00322447) 0.135704 ± 0.00140518 (W 0.0014278) 🔴 tlinalg-1lane 6.47× slower than tlinalg-pool
lu_factor/n16/b1024 1.91506 ± 0.0249218 (W 0.013055) 0.252263 ± 0.00212198 (W 0.00150737) 🔴 tlinalg-1lane 7.59× slower than tlinalg-pool
lu_factor/n32/b1024 5.66441 ± 0.0239483 (W 0.0145566) 0.723919 ± 0.00431761 (W 0.00184592) 🔴 tlinalg-1lane 7.82× slower than tlinalg-pool
lu_factor/n4/b1024 0.236664 ± 0.000912643 (W 0.0026237) 0.236148 ± 0.00167616 (W 0.00155674) ·
lu_factor/n8/b1024 0.609159 ± 0.00623041 (W 0.00475437) 0.0967765 ± 0.000975413 (W 0.00147184) 🔴 tlinalg-1lane 6.29× slower than tlinalg-pool
lu_factor_solve/n16/b1024 2.4989 ± 0.0264915 (W 0.00846591) 0.322345 ± 0.00508703 (W 0.00216781) 🔴 tlinalg-1lane 7.75× slower than tlinalg-pool
lu_factor_solve/n32/b1024 7.01706 ± 0.0244384 (W 0.011271) 0.887383 ± 0.00628826 (W 0.00462447) 🔴 tlinalg-1lane 7.91× slower than tlinalg-pool
lu_factor_solve/n4/b1024 0.298426 ± 0.00287018 (W 0.0024136) 0.057302 ± 0.00179883 (W 0.000687229) 🔴 tlinalg-1lane 5.21× slower than tlinalg-pool ! between-start range over 5%
lu_factor_solve/n8/b1024 0.837041 ± 0.00774732 (W 0.00861025) 0.119423 ± 0.00152607 (W 0.00149415) 🔴 tlinalg-1lane 7.01× slower than tlinalg-pool
lu_solve_prepared/n16/b1024 0.620486 ± 0.00330643 (W 0.00357745) 0.097026 ± 0.000704381 (W 0.00172004) 🔴 tlinalg-1lane 6.40× slower than tlinalg-pool
lu_solve_prepared/n32/b1024 2.36994 ± 0.0245499 (W 0.0204599) 0.230733 ± 0.00244539 (W 0.00195802) 🔴 tlinalg-1lane 10.27× slower than tlinalg-pool
lu_solve_prepared/n4/b1024 0.0732025 ± 0.00016526 (W 0.00100761) 0.020814 ± 0.000122848 (W 0.000135739) 🔴 tlinalg-1lane 3.52× slower than tlinalg-pool
lu_solve_prepared/n8/b1024 0.237337 ± 0.00111239 (W 0.00259421) 0.0436525 ± 0.000365632 (W 0.00049437) 🔴 tlinalg-1lane 5.44× slower than tlinalg-pool
qr/n16/b1024 9.03765 ± 0.159149 (W 0.0154025) 1.13528 ± 0.0190053 (W 0.00244876) 🔴 tlinalg-1lane 7.96× slower than tlinalg-pool
qr/n32/b1024 33.5354 ± 0.106137 (W 0.0291889) 4.23196 ± 0.00843614 (W 0.00830256) 🔴 tlinalg-1lane 7.92× slower than tlinalg-pool
qr/n4/b1024 0.770061 ± 0.00521924 (W 0.00410492) 0.101009 ± 0.00186983 (W 0.00066209) 🔴 tlinalg-1lane 7.62× slower than tlinalg-pool
qr/n8/b1024 2.41926 ± 0.0386027 (W 0.006063) 0.309712 ± 0.00426611 (W 0.00199856) 🔴 tlinalg-1lane 7.81× slower than tlinalg-pool
rank_revealing_qr/n16/b1024 10.8266 ± 0.0966263 (W 0.0192324) 1.35361 ± 0.0140052 (W 0.00250411) 🔴 tlinalg-1lane 8.00× slower than tlinalg-pool
rank_revealing_qr/n32/b1024 34.4205 ± 0.50422 (W 0.0415197) 4.37063 ± 0.0250375 (W 0.0105433) 🔴 tlinalg-1lane 7.88× slower than tlinalg-pool
rank_revealing_qr/n4/b1024 1.34628 ± 0.0234005 (W 0.00392067) 0.174407 ± 0.003185 (W 0.00204238) 🔴 tlinalg-1lane 7.72× slower than tlinalg-pool
rank_revealing_qr/n8/b1024 3.45962 ± 0.0299402 (W 0.00500486) 0.444638 ± 0.0196609 (W 0.00270843) 🔴 tlinalg-1lane 7.78× slower than tlinalg-pool ! between-start range over 5%
solve/n16/b1024 2.79951 ± 0.00902763 (W 0.015382) 0.358156 ± 0.00489281 (W 0.00324112) 🔴 tlinalg-1lane 7.82× slower than tlinalg-pool
solve/n32/b1024 7.63433 ± 0.0424703 (W 0.0249661) 0.961314 ± 0.0102158 (W 0.00518737) 🔴 tlinalg-1lane 7.94× slower than tlinalg-pool
solve/n4/b1024 0.398651 ± 0.00610768 (W 0.00378354) 0.0530095 ± 0.00107019 (W 0.00044153) 🔴 tlinalg-1lane 7.52× slower than tlinalg-pool
solve/n8/b1024 0.9533 ± 0.00926522 (W 0.00415344) 0.125198 ± 0.00167272 (W 0.00146855) 🔴 tlinalg-1lane 7.61× slower than tlinalg-pool
svd_full/n16/b1024 53.3921 ± 0.176067 (W 0.0375215) 6.68884 ± 0.0377078 (W 0.0105584) 🔴 tlinalg-1lane 7.98× slower than tlinalg-pool
svd_full/n32/b1024 187.035 ± 4.09193 (W 0.437846) 25.4885 ± 0.72332 (W 0.0210242) 🔴 tlinalg-1lane 7.34× slower than tlinalg-pool ! between-start range over 5%
svd_full/n4/b1024 4.38734 ± 0.0253421 (W 0.00708009) 0.553285 ± 0.00594591 (W 0.0033564) 🔴 tlinalg-1lane 7.93× slower than tlinalg-pool
svd_full/n8/b1024 14.9096 ± 0.369134 (W 0.0241369) 1.88544 ± 0.0481036 (W 0.00230222) 🔴 tlinalg-1lane 7.91× slower than tlinalg-pool ! between-start range over 5%
svd_thin/n16/b1024 53.4391 ± 0.25332 (W 0.0457313) 6.69341 ± 0.0260079 (W 0.00778176) 🔴 tlinalg-1lane 7.98× slower than tlinalg-pool
svd_thin/n32/b1024 187.212 ± 4.17959 (W 0.324838) 25.5112 ± 0.667937 (W 0.0188084) 🔴 tlinalg-1lane 7.34× slower than tlinalg-pool ! between-start range over 5%
svd_thin/n4/b1024 4.36506 ± 0.0295653 (W 0.00495021) 0.552784 ± 0.00278392 (W 0.00285223) 🔴 tlinalg-1lane 7.90× slower than tlinalg-pool
svd_thin/n8/b1024 14.8115 ± 0.420935 (W 0.0266155) 1.89821 ± 0.0480317 (W 0.00566023) 🔴 tlinalg-1lane 7.80× slower than tlinalg-pool ! between-start range over 5%
svd_values/n16/b1024 22.1934 ± 0.0847129 (W 0.0371547) 2.80804 ± 0.0156281 (W 0.00184502) 🔴 tlinalg-1lane 7.90× slower than tlinalg-pool
svd_values/n32/b1024 74.5443 ± 0.252253 (W 0.108117) 9.45575 ± 0.0565566 (W 0.00645828) 🔴 tlinalg-1lane 7.88× slower than tlinalg-pool
svd_values/n4/b1024 2.88112 ± 0.0246642 (W 0.00557846) 0.365038 ± 0.00378279 (W 0.00181227) 🔴 tlinalg-1lane 7.89× slower than tlinalg-pool
svd_values/n8/b1024 7.90004 ± 0.0446828 (W 0.00773972) 1.00182 ± 0.00664936 (W 0.00130818) 🔴 tlinalg-1lane 7.89× slower than tlinalg-pool
triangular_solve/n16/b1024 0.523161 ± 0.00252261 (W 0.00318308) 0.067923 ± 0.000406449 (W 0.000355283) 🔴 tlinalg-1lane 7.70× slower than tlinalg-pool
triangular_solve/n32/b1024 1.34304 ± 0.0104851 (W 0.0201505) 0.15789 ± 0.00137315 (W 0.00235982) 🔴 tlinalg-1lane 8.51× slower than tlinalg-pool
triangular_solve/n4/b1024 0.093564 ± 0.000990301 (W 0.00174753) 0.014277 ± 0.000508211 (W 0.000129012) 🔴 tlinalg-1lane 6.55× slower than tlinalg-pool ! between-start range over 5%
triangular_solve/n8/b1024 0.226043 ± 0.000346012 (W 0.00201615) 0.030708 ± 7.56351e-05 (W 0.000139216) 🔴 tlinalg-1lane 7.36× slower than tlinalg-pool

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

case tlinalg-1lane (ms) tlinalg-pool (ms) vs tlinalg-pool notes
cholesky/n16/b1024 1.28538 ± 0.0239876 (W 0.00824835) 0.15977 ± 0.00112832 (W 0.00166947) 🔴 tlinalg-1lane 8.05× slower than tlinalg-pool
cholesky/n32/b1024 5.65626 ± 0.0507389 (W 0.0160628) 0.793188 ± 0.00520949 (W 0.00607663) 🔴 tlinalg-1lane 7.13× slower than tlinalg-pool
cholesky/n4/b1024 0.19737 ± 6.93319e-05 (W 0.00218379) 0.027612 ± 0.000214399 (W 0.000133912) 🔴 tlinalg-1lane 7.15× slower than tlinalg-pool
cholesky/n8/b1024 0.419466 ± 0.0141926 (W 0.0030408) 0.0555145 ± 0.00132994 (W 0.000799489) 🔴 tlinalg-1lane 7.56× slower than tlinalg-pool ! between-start range over 5%
eig/n16/b1024 130.849 ± 0.424093 (W 0.20312) 16.491 ± 0.0393093 (W 0.0193509) 🔴 tlinalg-1lane 7.93× slower than tlinalg-pool
eig/n32/b1024 540.891 ± 0.521459 (W 0.560875) 67.8147 ± 0.0988268 (W 0.168301) 🔴 tlinalg-1lane 7.98× slower than tlinalg-pool
eig/n4/b1024 10.8051 ± 0.0261343 (W 0.020983) 1.38413 ± 0.00557563 (W 0.00280759) 🔴 tlinalg-1lane 7.81× slower than tlinalg-pool
eig/n8/b1024 35.7582 ± 0.142647 (W 0.0371665) 4.57954 ± 0.00716057 (W 0.00606449) 🔴 tlinalg-1lane 7.81× slower than tlinalg-pool
eigh/n16/b1024 32.2372 ± 0.0338834 (W 0.0970892) 4.01829 ± 0.00822771 (W 0.00606951) 🔴 tlinalg-1lane 8.02× slower than tlinalg-pool
eigh/n32/b1024 139.376 ± 0.229955 (W 0.0567097) 17.4101 ± 0.105418 (W 0.016723) 🔴 tlinalg-1lane 8.01× slower than tlinalg-pool
eigh/n4/b1024 2.98734 ± 0.0267806 (W 0.003892) 0.385568 ± 0.000198534 (W 0.00198619) 🔴 tlinalg-1lane 7.75× slower than tlinalg-pool
eigh/n8/b1024 9.32093 ± 0.0344586 (W 0.01151) 1.1855 ± 0.0136439 (W 0.00258504) 🔴 tlinalg-1lane 7.86× slower than tlinalg-pool
eigvals/n16/b1024 78.494 ± 0.214892 (W 0.0389553) 9.90462 ± 0.0157364 (W 0.0057164) 🔴 tlinalg-1lane 7.92× slower than tlinalg-pool
eigvals/n32/b1024 341.618 ± 0.76001 (W 0.512053) 43.3047 ± 0.256161 (W 0.0176132) 🔴 tlinalg-1lane 7.89× slower than tlinalg-pool
eigvals/n4/b1024 5.50508 ± 0.0220492 (W 0.00929968) 0.702493 ± 0.0125656 (W 0.00136886) 🔴 tlinalg-1lane 7.84× slower than tlinalg-pool
eigvals/n8/b1024 20.5258 ± 0.132634 (W 0.0152265) 2.60502 ± 0.0165216 (W 0.00346474) 🔴 tlinalg-1lane 7.88× slower than tlinalg-pool
eigvalsh/n16/b1024 24.1794 ± 0.0725263 (W 0.0128363) 3.04515 ± 0.00269629 (W 0.0025504) 🔴 tlinalg-1lane 7.94× slower than tlinalg-pool
eigvalsh/n32/b1024 92.1772 ± 0.101092 (W 0.225622) 11.5653 ± 0.0297567 (W 0.00837582) 🔴 tlinalg-1lane 7.97× slower than tlinalg-pool
eigvalsh/n4/b1024 2.49101 ± 0.016896 (W 0.00495819) 0.317123 ± 0.00260755 (W 0.0027383) 🔴 tlinalg-1lane 7.86× slower than tlinalg-pool
eigvalsh/n8/b1024 7.54171 ± 0.0326531 (W 0.00874277) 0.956037 ± 0.00216682 (W 0.00191545) 🔴 tlinalg-1lane 7.89× slower than tlinalg-pool
full_piv_lu/n16/b1024 5.21645 ± 0.193348 (W 0.0210197) 0.69198 ± 0.0120868 (W 0.00705793) 🔴 tlinalg-1lane 7.54× slower than tlinalg-pool ! between-start range over 5%
full_piv_lu/n32/b1024 21.4471 ± 0.26812 (W 0.0241189) 3.03571 ± 0.00758872 (W 0.0149521) 🔴 tlinalg-1lane 7.06× slower than tlinalg-pool
full_piv_lu/n4/b1024 0.590545 ± 0.00927172 (W 0.00389761) 0.0880455 ± 0.000572177 (W 0.000605743) 🔴 tlinalg-1lane 6.71× slower than tlinalg-pool
full_piv_lu/n8/b1024 1.52066 ± 0.0185886 (W 0.00774503) 0.211531 ± 0.0053198 (W 0.00272508) 🔴 tlinalg-1lane 7.19× slower than tlinalg-pool ! between-start range over 5%
full_piv_lu_solve/n16/b1024 6.26332 ± 0.0700505 (W 0.083906) 0.80608 ± 0.00675734 (W 0.00332452) 🔴 tlinalg-1lane 7.77× slower than tlinalg-pool
full_piv_lu_solve/n32/b1024 21.5046 ± 0.302278 (W 0.0158413) 2.75114 ± 0.0254169 (W 0.00697919) 🔴 tlinalg-1lane 7.82× slower than tlinalg-pool
full_piv_lu_solve/n4/b1024 0.903879 ± 0.00800231 (W 0.00410868) 0.118688 ± 0.00262013 (W 0.00269116) 🔴 tlinalg-1lane 7.62× slower than tlinalg-pool
full_piv_lu_solve/n8/b1024 2.262 ± 0.00992288 (W 0.0114047) 0.28849 ± 0.00377799 (W 0.00227125) 🔴 tlinalg-1lane 7.84× slower than tlinalg-pool
householder_apply/n16/b1024 5.55637 ± 0.103586 (W 0.0150257) 0.628277 ± 0.00755659 (W 0.00358075) 🔴 tlinalg-1lane 8.84× slower than tlinalg-pool
householder_apply/n32/b1024 27.6293 ± 0.252139 (W 0.0460723) 3.50438 ± 0.0138348 (W 0.0135921) 🔴 tlinalg-1lane 7.88× slower than tlinalg-pool
householder_apply/n4/b1024 0.392699 ± 0.00893044 (W 0.00270337) 0.0517825 ± 0.00125398 (W 0.000178476) 🔴 tlinalg-1lane 7.58× slower than tlinalg-pool
householder_apply/n8/b1024 1.21038 ± 0.0180582 (W 0.00365588) 0.152191 ± 0.00173001 (W 0.00200301) 🔴 tlinalg-1lane 7.95× slower than tlinalg-pool
householder_factor/n16/b1024 4.81899 ± 0.0277782 (W 0.0165255) 0.607811 ± 0.0086288 (W 0.00296049) 🔴 tlinalg-1lane 7.93× slower than tlinalg-pool
householder_factor/n32/b1024 20.3651 ± 0.231054 (W 0.0165742) 2.58815 ± 0.00630614 (W 0.00793076) 🔴 tlinalg-1lane 7.87× slower than tlinalg-pool
householder_factor/n4/b1024 0.598907 ± 0.00572545 (W 0.00328803) 0.0789025 ± 0.000233924 (W 0.000264616) 🔴 tlinalg-1lane 7.59× slower than tlinalg-pool
householder_factor/n8/b1024 1.58249 ± 0.00112833 (W 0.00199674) 0.203464 ± 0.00138985 (W 0.00160078) 🔴 tlinalg-1lane 7.78× slower than tlinalg-pool
lu/n16/b1024 3.32176 ± 0.041246 (W 0.0272603) 0.442744 ± 0.00321664 (W 0.00749766) 🔴 tlinalg-1lane 7.50× slower than tlinalg-pool
lu/n32/b1024 14.9569 ± 0.0427859 (W 0.0414373) 2.16555 ± 0.00775342 (W 0.0115644) 🔴 tlinalg-1lane 6.91× slower than tlinalg-pool
lu/n4/b1024 0.365368 ± 0.00272097 (W 0.00300299) 0.0633085 ± 0.000539309 (W 0.000682587) 🔴 tlinalg-1lane 5.77× slower than tlinalg-pool
lu/n8/b1024 0.950569 ± 0.00583529 (W 0.00847487) 0.135986 ± 0.00106466 (W 0.00251411) 🔴 tlinalg-1lane 6.99× slower than tlinalg-pool
lu_factor/n16/b1024 2.15059 ± 0.0385855 (W 0.00591663) 0.283028 ± 0.00420529 (W 0.00148772) 🔴 tlinalg-1lane 7.60× slower than tlinalg-pool
lu_factor/n32/b1024 9.82913 ± 0.0298877 (W 0.0197885) 1.24296 ± 0.0139209 (W 0.00367804) 🔴 tlinalg-1lane 7.91× slower than tlinalg-pool
lu_factor/n4/b1024 0.289491 ± 0.00194583 (W 0.00236056) 0.0620865 ± 0.00117632 (W 0.0008339) 🔴 tlinalg-1lane 4.66× slower than tlinalg-pool
lu_factor/n8/b1024 0.708804 ± 0.0110046 (W 0.00366876) 0.102968 ± 0.0023637 (W 0.00156562) 🔴 tlinalg-1lane 6.88× slower than tlinalg-pool
lu_factor_solve/n16/b1024 3.51095 ± 0.0490205 (W 0.0090668) 0.453072 ± 0.0048761 (W 0.00334839) 🔴 tlinalg-1lane 7.75× slower than tlinalg-pool
lu_factor_solve/n32/b1024 13.517 ± 0.07813 (W 0.0126448) 1.71451 ± 0.00919679 (W 0.00602896) 🔴 tlinalg-1lane 7.88× slower than tlinalg-pool
lu_factor_solve/n4/b1024 0.456199 ± 0.00332448 (W 0.00331504) 0.0727255 ± 0.00135844 (W 0.000676138) 🔴 tlinalg-1lane 6.27× slower than tlinalg-pool
lu_factor_solve/n8/b1024 1.19613 ± 0.0137258 (W 0.00273542) 0.15773 ± 0.00117848 (W 0.00191939) 🔴 tlinalg-1lane 7.58× slower than tlinalg-pool
lu_solve_prepared/n16/b1024 1.35571 ± 0.0106795 (W 0.00490564) 0.184124 ± 0.00218684 (W 0.00134483) 🔴 tlinalg-1lane 7.36× slower than tlinalg-pool
lu_solve_prepared/n32/b1024 6.86552 ± 0.0714994 (W 0.0301493) 0.939333 ± 0.00963784 (W 0.0124752) 🔴 tlinalg-1lane 7.31× slower than tlinalg-pool
lu_solve_prepared/n4/b1024 0.176455 ± 0.000765631 (W 0.00199462) 0.033978 ± 0.000302326 (W 0.000405034) 🔴 tlinalg-1lane 5.19× slower than tlinalg-pool
lu_solve_prepared/n8/b1024 0.500232 ± 0.00710698 (W 0.0031523) 0.0760175 ± 0.000182671 (W 0.00169845) 🔴 tlinalg-1lane 6.58× slower than tlinalg-pool
qr/n16/b1024 11.1791 ± 0.0670875 (W 0.012904) 1.40611 ± 0.00523828 (W 0.00481913) 🔴 tlinalg-1lane 7.95× slower than tlinalg-pool
qr/n32/b1024 71.7039 ± 0.421098 (W 0.059286) 8.93241 ± 0.0662829 (W 0.0108604) 🔴 tlinalg-1lane 8.03× slower than tlinalg-pool
qr/n4/b1024 0.933366 ± 0.00695303 (W 0.00163928) 0.120058 ± 0.00111786 (W 0.000669157) 🔴 tlinalg-1lane 7.77× slower than tlinalg-pool
qr/n8/b1024 2.72505 ± 0.0258193 (W 0.00530172) 0.342433 ± 0.00199048 (W 0.00251128) 🔴 tlinalg-1lane 7.96× slower than tlinalg-pool
rank_revealing_qr/n16/b1024 12.596 ± 0.0700264 (W 0.0197643) 1.58619 ± 0.0113445 (W 0.00336078) 🔴 tlinalg-1lane 7.94× slower than tlinalg-pool
rank_revealing_qr/n32/b1024 67.0404 ± 0.288876 (W 0.0455975) 8.43201 ± 0.0325196 (W 0.0142625) 🔴 tlinalg-1lane 7.95× slower than tlinalg-pool
rank_revealing_qr/n4/b1024 1.51263 ± 0.0206142 (W 0.00255275) 0.194875 ± 0.00292921 (W 0.00164045) 🔴 tlinalg-1lane 7.76× slower than tlinalg-pool
rank_revealing_qr/n8/b1024 3.88745 ± 0.0123452 (W 0.00521444) 0.492556 ± 0.00410607 (W 0.00245592) 🔴 tlinalg-1lane 7.89× slower than tlinalg-pool
solve/n16/b1024 3.92546 ± 0.0395587 (W 0.00714371) 0.502991 ± 0.0049424 (W 0.00230284) 🔴 tlinalg-1lane 7.80× slower than tlinalg-pool
solve/n32/b1024 14.4318 ± 0.0583992 (W 0.0158539) 1.84476 ± 0.0139515 (W 0.0034312) 🔴 tlinalg-1lane 7.82× slower than tlinalg-pool
solve/n4/b1024 0.629698 ± 0.0103083 (W 0.00337184) 0.082674 ± 0.000958606 (W 0.000886017) 🔴 tlinalg-1lane 7.62× slower than tlinalg-pool
solve/n8/b1024 1.43473 ± 0.00711314 (W 0.00915072) 0.18492 ± 0.000607923 (W 0.00149538) 🔴 tlinalg-1lane 7.76× slower than tlinalg-pool
svd_full/n16/b1024 57.7402 ± 0.467004 (W 0.0388564) 7.22931 ± 0.15502 (W 0.0099396) 🔴 tlinalg-1lane 7.99× slower than tlinalg-pool
svd_full/n32/b1024 246.361 ± 1.06191 (W 0.385223) 31.0429 ± 0.284233 (W 0.0435566) 🔴 tlinalg-1lane 7.94× slower than tlinalg-pool
svd_full/n4/b1024 5.2355 ± 0.0160379 (W 0.00811481) 0.661181 ± 0.00504798 (W 0.00200157) 🔴 tlinalg-1lane 7.92× slower than tlinalg-pool
svd_full/n8/b1024 16.5342 ± 0.0242796 (W 0.0134403) 2.10683 ± 0.0249002 (W 0.00427323) 🔴 tlinalg-1lane 7.85× slower than tlinalg-pool
svd_thin/n16/b1024 57.7949 ± 0.178256 (W 0.330102) 7.25694 ± 0.102446 (W 0.0074772) 🔴 tlinalg-1lane 7.96× slower than tlinalg-pool
svd_thin/n32/b1024 245.974 ± 0.713985 (W 0.444368) 31.0439 ± 0.338167 (W 0.0243581) 🔴 tlinalg-1lane 7.92× slower than tlinalg-pool
svd_thin/n4/b1024 5.23331 ± 0.00779485 (W 0.00519314) 0.677936 ± 0.0350737 (W 0.0017882) 🔴 tlinalg-1lane 7.72× slower than tlinalg-pool ! between-start range over 5%
svd_thin/n8/b1024 16.5213 ± 0.0415848 (W 0.0785085) 2.13234 ± 0.0534661 (W 0.00464835) 🔴 tlinalg-1lane 7.75× slower than tlinalg-pool
svd_values/n16/b1024 25.622 ± 0.0526437 (W 0.0158991) 3.2085 ± 0.0234414 (W 0.00277774) 🔴 tlinalg-1lane 7.99× slower than tlinalg-pool
svd_values/n32/b1024 90.6482 ± 0.419792 (W 0.0833006) 11.4469 ± 0.057231 (W 0.0177248) 🔴 tlinalg-1lane 7.92× slower than tlinalg-pool
svd_values/n4/b1024 3.53501 ± 0.0206521 (W 0.00623745) 0.448059 ± 0.00293904 (W 0.00213503) 🔴 tlinalg-1lane 7.89× slower than tlinalg-pool
svd_values/n8/b1024 9.15925 ± 0.0293712 (W 0.0105678) 1.1539 ± 0.0058777 (W 0.00166669) 🔴 tlinalg-1lane 7.94× slower than tlinalg-pool
triangular_solve/n16/b1024 1.02693 ± 0.0152713 (W 0.00528121) 0.129217 ± 0.00163603 (W 0.00172495) 🔴 tlinalg-1lane 7.95× slower than tlinalg-pool
triangular_solve/n32/b1024 4.27327 ± 0.0420663 (W 0.0147631) 0.539872 ± 0.00224006 (W 0.00442122) 🔴 tlinalg-1lane 7.92× slower than tlinalg-pool
triangular_solve/n4/b1024 0.172242 ± 0.00105772 (W 0.00128183) 0.0239945 ± 0.000162656 (W 0.00101274) 🔴 tlinalg-1lane 7.18× slower than tlinalg-pool
triangular_solve/n8/b1024 0.414199 ± 0.0078952 (W 0.00355543) 0.054852 ± 0.00106515 (W 0.000402374) 🔴 tlinalg-1lane 7.55× slower than tlinalg-pool

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

case tlinalg-1lane (ms) tlinalg-pool (ms) vs tlinalg-pool notes
cholesky/n128/b1 0.0669 ± 0.00212853 (W 9.57283e-05) 0.0690595 ± 0.00196527 (W 0.00161333) · ! between-start range over 5%
lu/n128/b1 0.168801 ± 0.00110254 (W 0.0016513) 0.173409 ± 0.00165156 (W 0.00166176) ·
lu_factor/n128/b1 0.0955385 ± 0.000667572 (W 0.00203813) 0.093059 ± 0.00116777 (W 0.00030513) ·
lu_factor_solve/n128/b1 0.104777 ± 0.000945151 (W 0.00178702) 0.103918 ± 0.000677575 (W 0.000433825) ·
lu_solve_prepared/n128/b1 0.009964 ± 7.14143e-05 (W 5.69318e-05) 0.011351 ± 6.20134e-05 (W 3.03041e-05) 🔵 tlinalg-1lane 1.14× faster than tlinalg-pool dispatch-dominated (row under 20 us)
solve/n128/b1 0.112991 ± 0.000793057 (W 0.000115658) 0.118311 ± 0.000839641 (W 0.00132211) ·
svd_values/n128/b1 1.05009 ± 0.00180357 (W 0.00322754) 1.14114 ± 0.00505631 (W 0.00243071) 🔵 tlinalg-1lane 1.09× faster than tlinalg-pool
triangular_solve/n128/b1 0.006918 ± 6.0208e-05 (W 4.47201e-05) 0.00801 ± 8.05828e-05 (W 5.02259e-05) 🔵 tlinalg-1lane 1.16× faster than tlinalg-pool dispatch-dominated (row under 20 us)

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

case tlinalg-1lane (ms) tlinalg-pool (ms) vs tlinalg-pool notes
cholesky/n128/b1 0.177813 ± 0.000227549 (W 0.00202236) 0.156769 ± 0.00229206 (W 0.00151937) 🔴 tlinalg-1lane 1.13× slower than tlinalg-pool
lu/n128/b1 0.358216 ± 0.00334689 (W 0.00334171) 0.314957 ± 0.00196992 (W 0.00385947) 🔴 tlinalg-1lane 1.14× slower than tlinalg-pool
lu_factor/n128/b1 0.273166 ± 0.00287235 (W 0.00254956) 0.2254 ± 0.00111694 (W 0.00173016) 🔴 tlinalg-1lane 1.21× slower than tlinalg-pool
lu_factor_solve/n128/b1 0.304059 ± 0.00127159 (W 0.00224354) 0.258043 ± 0.0021565 (W 0.0019576) 🔴 tlinalg-1lane 1.18× slower than tlinalg-pool
lu_solve_prepared/n128/b1 0.0317295 ± 0.000113946 (W 0.000106186) 0.033873 ± 0.000224512 (W 0.000816128) 🔵 tlinalg-1lane 1.07× faster than tlinalg-pool
solve/n128/b1 0.314994 ± 0.00120712 (W 0.00282996) 0.282965 ± 0.00211172 (W 0.00220004) 🔴 tlinalg-1lane 1.11× slower than tlinalg-pool
svd_values/n128/b1 1.7639 ± 0.00335475 (W 0.00795172) 1.43433 ± 0.00532239 (W 0.00249503) 🔴 tlinalg-1lane 1.23× slower than tlinalg-pool
triangular_solve/n128/b1 0.0189055 ± 0.000105894 (W 0.000104426) 0.0201325 ± 0.000212715 (W 0.000137296) 🔵 tlinalg-1lane 1.06× faster than tlinalg-pool dispatch-dominated (row under 20 us)

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

case tlinalg-1lane (ms) tlinalg-pool (ms) vs tlinalg-pool notes
eigh/n512/b1 56.8057 ± 0.286211 (W 0.247835) 27.3049 ± 0.0240236 (W 0.0169525) 🔴 tlinalg-1lane 2.08× slower than tlinalg-pool
eigvalsh/n512/b1 23.4726 ± 0.267259 (W 0.0276072) 16.2183 ± 0.0475803 (W 0.0159732) 🔴 tlinalg-1lane 1.45× slower than tlinalg-pool
householder_apply/n512/b1 10.2447 ± 0.0567724 (W 0.0103856) 1.59405 ± 0.0026899 (W 0.00323868) 🔴 tlinalg-1lane 6.43× slower than tlinalg-pool
householder_factor/n512/b1 7.10544 ± 0.0306333 (W 0.0105699) 1.84575 ± 0.00981167 (W 0.00475265) 🔴 tlinalg-1lane 3.85× slower than tlinalg-pool
qr/n512/b1 19.1257 ± 0.135494 (W 0.0196129) 5.49195 ± 0.0122292 (W 0.0103794) 🔴 tlinalg-1lane 3.48× slower than tlinalg-pool
rank_revealing_qr/n512/b1 29.6834 ± 0.128296 (W 0.0304971) 7.28305 ± 0.0143347 (W 0.0174838) 🔴 tlinalg-1lane 4.08× slower than tlinalg-pool
svd_thin/n512/b1 107.282 ± 0.840718 (W 0.284136) 38.5445 ± 0.10678 (W 0.0209864) 🔴 tlinalg-1lane 2.78× slower than tlinalg-pool

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

case tlinalg-1lane (ms) tlinalg-pool (ms) vs tlinalg-pool notes
eigh/n512/b1 113.9 ± 0.0547224 (W 0.0797068) 37.0025 ± 0.0375861 (W 0.20581) 🔴 tlinalg-1lane 3.08× slower than tlinalg-pool
eigvalsh/n512/b1 49.4074 ± 0.19457 (W 0.0335589) 20.4494 ± 0.00997976 (W 0.028072) 🔴 tlinalg-1lane 2.42× slower than tlinalg-pool
householder_apply/n512/b1 41.0623 ± 0.0492058 (W 0.0267711) 5.63233 ± 0.0102797 (W 0.0106328) 🔴 tlinalg-1lane 7.29× slower than tlinalg-pool
householder_factor/n512/b1 27.2316 ± 0.0475866 (W 0.0246958) 5.17474 ± 0.0169352 (W 0.00942564) 🔴 tlinalg-1lane 5.26× slower than tlinalg-pool
qr/n512/b1 70.0125 ± 0.19024 (W 0.0407955) 13.3358 ± 0.0380976 (W 0.0166864) 🔴 tlinalg-1lane 5.25× slower than tlinalg-pool
rank_revealing_qr/n512/b1 79.3833 ± 0.212833 (W 0.0476459) 14.2642 ± 0.0349989 (W 0.0122797) 🔴 tlinalg-1lane 5.57× slower than tlinalg-pool
svd_thin/n512/b1 255.569 ± 0.495028 (W 0.964147) 75.0869 ± 0.050694 (W 0.0986574) 🔴 tlinalg-1lane 3.40× slower than tlinalg-pool

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

case tlinalg-1lane (ms) tlinalg-pool (ms) vs tlinalg-pool notes
householder_factor/4096x256/b1 23.1538 ± 0.352265 (W 0.0351949) 5.92498 ± 0.0144641 (W 0.0270192) 🔴 tlinalg-1lane 3.91× slower than tlinalg-pool
qr/4096x256/b1 65.0536 ± 0.0326541 (W 0.0780722) 22.1582 ± 0.0876764 (W 0.0477128) 🔴 tlinalg-1lane 2.94× slower than tlinalg-pool
rank_revealing_qr/4096x256/b1 101.263 ± 0.727538 (W 0.356657) 21.4538 ± 0.140407 (W 0.0578058) 🔴 tlinalg-1lane 4.72× slower than tlinalg-pool
svd_thin/4096x256/b1 99.7401 ± 0.364976 (W 0.257105) 35.0016 ± 0.0365897 (W 0.0532894) 🔴 tlinalg-1lane 2.85× slower than tlinalg-pool

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

case tlinalg-1lane (ms) tlinalg-pool (ms) vs tlinalg-pool notes
householder_factor/4096x256/b1 85.6167 ± 0.0622036 (W 0.171229) 17.444 ± 0.0747473 (W 0.0344407) 🔴 tlinalg-1lane 4.91× slower than tlinalg-pool
qr/4096x256/b1 234.404 ± 0.292511 (W 0.443775) 58.543 ± 0.0484898 (W 0.0761094) 🔴 tlinalg-1lane 4.00× slower than tlinalg-pool
rank_revealing_qr/4096x256/b1 348.83 ± 2.57016 (W 0.75279) 59.746 ± 0.187146 (W 0.141795) 🔴 tlinalg-1lane 5.84× slower than tlinalg-pool
svd_thin/4096x256/b1 363.38 ± 0.258475 (W 0.333036) 108.807 ± 0.0741426 (W 0.256627) 🔴 tlinalg-1lane 3.34× slower than tlinalg-pool

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.