%PDF-1.6
%
1 0 obj
<>
endobj
2 0 obj
<>
endobj
3 0 obj
<>stream
IEEE
IEEE Transactions on Parallel and Distributed Systems;2012;23;11;10.1109/TPDS.2011.311
Graphics processing unit
matrix multiplication
code generation
automatic tuning
GEMM
BLAS
CUDA
Autotuning GEMM Kernels for the Fermi GPU
Jakub Kurzak
Stanimire Tomov
Jack Dongarra
endstream
endobj
4 0 obj
<>stream
x+ |
endstream
endobj
5 0 obj
<>stream
x0YjbqMpQ%Ѫ5b)1͙`GOZU)ĕ|bo`4&VRo#/,CȮuv&F&g8a4,c2JLMA>
?G8
%֔HZ+=`/g)S5(Z=ŃXPF;/9
endstream
endobj
6 0 obj
<>stream
x+ |
endstream
endobj
7 0 obj
<>stream
x0YjbqMpQ%Ѫ5b)1͙`GOZS+(#hMfuwcG^YVM])L>stream
x+ |
endstream
endobj
9 0 obj
<>stream
x0YjbqMpQ%Ѫ5b)1͙`GOZU)ĕ|bo`4&VRo#/,CȮuv&F&g8a4,c2JLMA>
?G8
%֔HZ+=`/g)S5(Z=ŃXPF;9
endstream
endobj
10 0 obj
<>stream
x+ |
endstream
endobj
11 0 obj
<>stream
xAo@sX4jL@,,횅e`Mfpm%!9'Bނd3o-:j}Q0aTZ'{̳L~kyE;cz<HDU!J3T xmUߛGK*@9
endstream
endobj
12 0 obj
<>stream
x+ |
endstream
endobj
13 0 obj
<>stream
x0YjbqMpQ%Ѫ5b)1͙`GOZU)ĕ|bo`4&VRo#/,CȮuv&F&g8a4,c2JLMA>
?G8
%֔HZ+=`/g)S5(Z=ŃXPF;q9
endstream
endobj
14 0 obj
<>stream
x+ |
endstream
endobj
15 0 obj
<>stream
x0YjbOv> .0J5$ZP,%F^49lIsAbq%ԙ5n:˲nԲ-rODYApr1{XWM~iĺoQ/CŃQ؋HJ+=`'kg)SȫJ=tŃXPF[9
endstream
endobj
16 0 obj
<>stream
x+ |
endstream
endobj
17 0 obj
<>stream
xM@+H"JzJPK86,Vn#z%sy&3ғ悆Gcq%xJZw7V}ZethٕR$gwy*Nج 88ϙ=,ͫҦt?Sa7hQDb5%RfƱJYuvT
o} Ӗi9
endstream
endobj
18 0 obj
<>stream
x+ |
endstream
endobj
19 0 obj
<>stream
x0YjbqMpQ%Ѫ5b)1͙`GOZS+(#hM[Iyƪ@m#R*ٙyx%Ӕʼ*mK36~#((M(@.ZS"afkmlUgLՠkbA9
endstream
endobj
20 0 obj
<>stream
x+ |
endstream
endobj
21 0 obj
<>stream
xK0&[W8&(h՚BﯿhfMf'-$u(0'F!F`2l;N>ѵ/MPdy*Oخ!98ϙ6?[cә7xHq2$qB.quB86: {z^ۺE4` Ēr?9
endstream
endobj
22 0 obj
<>stream
x+ |
endstream
endobj
23 0 obj
<>stream
x0YjbqMpQ%Ѫ5b)1͙`GOZUb
q%췕ԛn:˪6+rɁY"Nخ!8M9˘=̫Ҧt?Sa7h$b5%fƱJYuvT
o} ю9
endstream
endobj
24 0 obj
<>stream
x+ |
endstream
endobj
25 0 obj
<>stream
xAo@s,X4jL@,횅e`MfPЍ^Z!/b!-H6O&JdoSyv7v\U>39O?a?xQ#U3T jN݀#B^%UT?9
endstream
endobj
26 0 obj
<>stream
x+ |
endstream
endobj
27 0 obj
<>stream
x0YjbqMpQ%Ѫ5b)1͙`GOZS+(#hMfuwcG^YVM])L>stream
x+ |
endstream
endobj
29 0 obj
<>stream
x
0Y*HB]TQtY4j$mj"Vf'
*J>~7
ԛn:˪6+rITYApq3{XWMq~ºo C%8WkJ̌c씩uA,(-}9
endstream
endobj
30 0 obj
<>/ProcSet[/PDF/Text]/Font<>>>/CropBox[0 0 567 774]/Parent 49 0 R/Rotate 0/MediaBox[0 0 567 774]>>
endobj
49 0 obj
<>
endobj
31 0 obj
<>stream
HUMo8WH1"iiEx@˴͍ElGi9MQ4|Ӄ(vjvy^3!L>Z
jL%18Ūb7X5c2{$W
caWΥTO{\ s1C?d[iE{7ݹ)bqK"?o3ΨLvZ
EPLF;z$Ӹ.tAJH9"]ވ\ZgkoHZnr!0}ZUa̭ґrEPa.Vj>kXj(xixVwt4gv
sEk<mȿs6l2k
j7O6;B_\
ʿqCǂņjjꓻOˊQUU0պmw.(N]6Ga,qa_D|u6Anػ|F@rAE{D?ަ]΅u4dےtAasgo7}mw +q 쐧P8opqasїRmtQ4eǃo =3sXPjl4
)"MfYL*0n368Pɯ XEaB_Lu{}{u1MEx@G]!VrIRtqQ[`:ӏ
qB#퇴K6ݘ(jϕ+tuCjSȳ ("k C&Sp1ÌY4g@Wh^Kf52^#&MVI.$ Ajڛ:gvxeWB؝)0'Е2'9b~z.|q2.P䡱C>x0lHKv.m\c!rN0xv')c۞ Z ~M3-uN HEo Q1
endstream
endobj
32 0 obj
<>stream
HlUKr6)Ĥ,++Yb[eS+;2g
)ɑr-w5r o_6__FgT aē!wT?vwCp;d4-8l$iNBh=(դjG{t>$G>C'SRA,`Ќ\ُ>%8nC*$
}4܅ص
-V)JI
w~pݔR$G~ò3N%3&.GaQLGW6:ޥ:_%S`q>4h!0PP"hlP*qvxj)\?Jޠr!H
/}/1ӂK!4ǝ[omVpU&p&A9eoঀ/[\Ot5Mo}XpkJR@Sҗ:nz\{c
zr$@ӊ֒lVS °dJv&<Ġ:t"N\Lo{pC%qhϐd0`t;秔NXMbX(k%Gh'#v1+rݪ4[.{Q_0w2YEVa` 8O9O0B%=6+VcRyW+ǷՆRgJjuFs%ʼ7o*@#Oh JqeYFDYoȿÒVK"a*~dghg4ZaN:
טZF%pwƢۋGt}N䐈"Aׄ"vwnp/1ڌRhKc~%o~۞ŗHAIs
pc`eoNT&f Uh\N߶xg0#mapip^3!霅}` 5
endstream
endobj
33 0 obj
<>stream
HTUr6WV,:fsHEDA"!t%`=/Uh%â*dzӫ٘B\Zbp}I3PGg_
~/WH.Jk}81״E{a7;&0Kf]1CuHd췱ϞJU5GR.~C>45eF#ChɢBiQXwލݥewmeٱ~݁p{ԹB6`ܘ/ToΧƠn7b/nHPTӭOFeT盝;;0uM?c*Y'lzH;Sb]5|_sqb5mR\hR8eg.;@(7B87\[x|g˻vCHV:IFjD@