Replaced inner loops with assembly calls for Dge* functions and Dsyrk

[btracey]

These are the functions that are most likely to be used / are already used. Others can be updated as needed.

[btracey]

Pretty significant speedups. Thanks @fhs !

benchmark                           old ns/op      new ns/op      delta       
BenchmarkDgemmSmSmSm                2347           1870           -20.32%     
BenchmarkDgemmMedMedMed             1420777        601110         -57.69%     
BenchmarkDgemmMedLgMed              14150749       5703823        -59.69%     
BenchmarkDgemmLgLgLg                1413082561     588834401      -58.33%     
BenchmarkDgemmLgSmLg                16212486       12189649       -24.81%     
BenchmarkDgemmLgLgSm                16308323       12406908       -23.92%     
BenchmarkDgemmHgHgSm                1690570016     1217346857     -27.99%     
BenchmarkDgemmMedMedMedTNT          1400496        572611         -59.11%     
BenchmarkDgemmMedMedMedNTT          1106778        654992         -40.82%     
BenchmarkDgemmMedMedMedNTNT         1406017        592313         -57.87%     
BenchmarkDgemvSmSmNoTransInc1       157            130            -17.20%     
BenchmarkDgemvSmSmNoTransIncN       209            204            -2.39%      
BenchmarkDgemvSmSmTransInc1         220            141            -35.91%     
BenchmarkDgemvSmSmTransIncN         271            219            -19.19%     
BenchmarkDgemvMedMedNoTransInc1     10003          5276           -47.26%     
BenchmarkDgemvMedMedNoTransIncN     13430          11311          -15.78%     
BenchmarkDgemvMedMedTransInc1       14554          5294           -63.63%     
BenchmarkDgemvMedMedTransIncN       16948          12260          -27.66%     
BenchmarkDgemvLgLgNoTransInc1       1044625        604480         -42.13%     
BenchmarkDgemvLgLgNoTransIncN       1326452        1159622        -12.58%     
BenchmarkDgemvLgLgTransInc1         1351233        611742         -54.73%     
BenchmarkDgemvLgLgTransIncN         1693624        1197201        -29.31%     
BenchmarkDgemvLgSmNoTransInc1       13017          10620          -18.41%     
BenchmarkDgemvLgSmNoTransIncN       17490          18077          +3.36%      
BenchmarkDgemvLgSmTransInc1         18241          10296          -43.56%     
BenchmarkDgemvLgSmTransIncN         22960          18761          -18.29%     
BenchmarkDgemvSmLgNoTransInc1       9231           4594           -50.23%     
BenchmarkDgemvSmLgNoTransIncN       12685          10905          -14.03%     
BenchmarkDgemvSmLgTransInc1         13172          4631           -64.84%     
BenchmarkDgemvSmLgTransIncN         16498          11630          -29.51%     
BenchmarkDgerSmSmInc1               174            125            -28.16%     
BenchmarkDgerSmSmIncN               232            202            -12.93%     
BenchmarkDgerMedMedInc1             13183          5443           -58.71%     
BenchmarkDgerMedMedIncN             16141          12183          -24.52%     
BenchmarkDgerLgLgInc1               1372862        783534         -42.93%     
BenchmarkDgerLgLgIncN               1565488        1243930        -20.54%     
BenchmarkDgerLgSmInc1               15179          10845          -28.55%     
BenchmarkDgerLgSmIncN               21187          19274          -9.03%      
BenchmarkDgerSmLgInc1               12449          3595           -71.12%     
BenchmarkDgerSmLgIncN               14455          10906          -24.55% 

[coveralls]

Coverage decreased (-0.26%) when pulling b4c254ddc0bbc748a90bca0418212dbe42d3421b on replaceinner into 26b876145b7733caf1212969f4715692e9fc67dd on master.

[btracey]

Still a while to cgo, but we're getting there. These level 2 routines are probably parallel in cgo.

benchmark                           old ns/op     new ns/op     delta        
BenchmarkDgemvSmSmNoTransInc1       130           301           +131.54%     
BenchmarkDgemvSmSmNoTransIncN       204           309           +51.47%      
BenchmarkDgemvSmSmTransInc1         141           298           +111.35%     
BenchmarkDgemvSmSmTransIncN         219           316           +44.29%      
BenchmarkDgemvMedMedNoTransInc1     5276          1844          -65.05%      
BenchmarkDgemvMedMedNoTransIncN     11311         2536          -77.58%      
BenchmarkDgemvMedMedTransInc1       5294          2515          -52.49%      
BenchmarkDgemvMedMedTransIncN       12260         2844          -76.80%      
BenchmarkDgemvLgLgNoTransInc1       604480        258565        -57.23%      
BenchmarkDgemvLgLgNoTransIncN       1159622       264518        -77.19%      
BenchmarkDgemvLgLgTransInc1         611742        280277        -54.18%      
BenchmarkDgemvLgLgTransIncN         1197201       289379        -75.83%      
BenchmarkDgemvLgSmNoTransInc1       10620         3304          -68.89%      
BenchmarkDgemvLgSmNoTransIncN       18077         5370          -70.29%      
BenchmarkDgemvLgSmTransInc1         10296         3358          -67.39%      
BenchmarkDgemvLgSmTransIncN         18761         2728          -85.46%      
BenchmarkDgemvSmLgNoTransInc1       4594          2110          -54.07%      
BenchmarkDgemvSmLgNoTransIncN       10905         2281          -79.08%      
BenchmarkDgemvSmLgTransInc1         4631          3666          -20.84%      
BenchmarkDgemvSmLgTransIncN         11630         6218          -46.53%      
BenchmarkDgerSmSmInc1               125           791           +532.80%     
BenchmarkDgerSmSmIncN               202           798           +295.05%     
BenchmarkDgerMedMedInc1             5443          2382          -56.24%      
BenchmarkDgerMedMedIncN             12183         2570          -78.91%      
BenchmarkDgerLgLgInc1               783534        615112        -21.50%      
BenchmarkDgerLgLgIncN               1243930       623505        -49.88%      
BenchmarkDgerLgSmInc1               10845         4477          -58.72%      
BenchmarkDgerLgSmIncN               19274         4390          -77.22%      
BenchmarkDgerSmLgInc1               3595          2278          -36.63%      
BenchmarkDgerSmLgIncN               10906         3055          -71.99%  

[btracey]

Ddot's getting really close though:

brendan:~/Documents/mygo/src/github.com/gonum/blas/native$ benchcmp new.bench cgo.bench 
benchmark                          old ns/op     new ns/op     delta         
BenchmarkDdotSmallBothUnitary      15.8          207           +1210.13%     
BenchmarkDdotSmallIncUni           18.5          214           +1056.76%     
BenchmarkDdotSmallUniInc           18.5          214           +1056.76%     
BenchmarkDdotSmallBothInc          18.6          216           +1061.29%     
BenchmarkDdotMediumBothUnitary     460           435           -5.43%        
BenchmarkDdotMediumIncUni          1078          1054          -2.23%        
BenchmarkDdotMediumUniInc          1096          866           -20.99%       
BenchmarkDdotMediumBothInc         1084          1166          +7.56%        
BenchmarkDdotLargeBothUnitary      49681         39232         -21.03%       
BenchmarkDdotLargeIncUni           198573        170580        -14.10%       
BenchmarkDdotLargeUniInc           118322        101916        -13.87%       
BenchmarkDdotLargeBothInc          347034        274692        -20.85%       
BenchmarkDdotHugeBothUnitary       11061677      10652330      -3.70%        
BenchmarkDdotHugeIncUni            37096808      32448696      -12.53%       
BenchmarkDdotHugeUniInc            24956236      20180868      -19.13%       
BenchmarkDdotHugeBothInc           45096756      41569145      -7.82%   

[btracey]

PTAL

Updated benchmarks:

brendan:~/Documents/mygo/src/github.com/gonum/blas/native$ benchcmp old.txt new.txt
benchmark                           old ns/op      new ns/op      delta       
BenchmarkDgemmSmSmSm                2729           1901           -30.34%     
BenchmarkDgemmMedMedMed             1454813        613739         -57.81%     
BenchmarkDgemmMedLgMed              14976720       5682133        -62.06%     
BenchmarkDgemmLgLgLg                1437288447     583485583      -59.40%     
BenchmarkDgemmLgSmLg                23259519       12217995       -47.47%     
BenchmarkDgemmLgLgSm                23417498       12636051       -46.04%     
BenchmarkDgemmHgHgSm                2353436677     1205930033     -48.76%     
BenchmarkDgemmMedMedMedTNT          1513161        599095         -60.41%     
BenchmarkDgemmMedMedMedNTT          1284047        669217         -47.88%     
BenchmarkDgemmMedMedMedNTNT         1445991        615434         -57.44%     
BenchmarkDgemvSmSmNoTransInc1       206            141            -31.55%     
BenchmarkDgemvSmSmNoTransIncN       247            206            -16.60%     
BenchmarkDgemvSmSmTransInc1         226            143            -36.73%     
BenchmarkDgemvSmSmTransIncN         279            230            -17.56%     
BenchmarkDgemvMedMedNoTransInc1     10785          5446           -49.50%     
BenchmarkDgemvMedMedNoTransIncN     13994          12008          -14.19%     
BenchmarkDgemvMedMedTransInc1       14127          5444           -61.46%     
BenchmarkDgemvMedMedTransIncN       15957          11829          -25.87%     
BenchmarkDgemvLgLgNoTransInc1       1008614        557777         -44.70%     
BenchmarkDgemvLgLgNoTransIncN       1292387        1097074        -15.11%     
BenchmarkDgemvLgLgTransInc1         1257284        561283         -55.36%     
BenchmarkDgemvLgLgTransIncN         1457361        1193900        -18.08%     
BenchmarkDgemvLgSmNoTransInc1       16966          10974          -35.32%     
BenchmarkDgemvLgSmNoTransIncN       22019          18493          -16.01%     
BenchmarkDgemvLgSmTransInc1         18401          10811          -41.25%     
BenchmarkDgemvLgSmTransIncN         23416          18270          -21.98%     
BenchmarkDgemvSmLgNoTransInc1       9830           4777           -51.40%     
BenchmarkDgemvSmLgNoTransIncN       12141          10670          -12.12%     
BenchmarkDgemvSmLgTransInc1         13488          4620           -65.75%     
BenchmarkDgemvSmLgTransIncN         14623          11574          -20.85%     
BenchmarkDgerSmSmInc1               216            131            -39.35%     
BenchmarkDgerSmSmIncN               257            206            -19.84%     
BenchmarkDgerMedMedInc1             13255          5293           -60.07%     
BenchmarkDgerMedMedIncN             15942          12235          -23.25%     
BenchmarkDgerLgLgInc1               1276512        639479         -49.90%     
BenchmarkDgerLgLgIncN               1464653        1194753        -18.43%     
BenchmarkDgerLgSmInc1               18638          12299          -34.01%     
BenchmarkDgerLgSmIncN               22120          18782          -15.09%     
BenchmarkDgerSmLgInc1               11955          3835           -67.92%     
BenchmarkDgerSmLgIncN               14437          10814          -25.10%   

[btracey]

@jonlawlor this should bring the covmat benchmarks up

[coveralls]

Coverage decreased (-0.25%) when pulling e6a3f0fecd3277e7663c6856481b9914d74c1beb on replaceinner into 45eea585d1a9673f2d2aeb424a1c47fd0b958f15 on master.

[kortschak]

LGTM