darft of chapter four

• Messages sorted by: [ date ][ thread ][ subject ][ author ]
• Next message: (no name): "Re: YASM"
• Previous message: Richard Frost: "info, hints, commandline in dynamic/io/1sided (fwd)"
• Next in thread: Steve Huss-Lederman: "Re: darft of chapter four"

$F2psBegin
0 setlinecap 0 setlinejoin
-81.0 315.0 translate 0.900 -0.900 scale
2.000 setlinewidth
n 179 79 m 199 139 l gs col-1 s gr
n 197.103 114.334 m 199.000 139.000 l 185.718 118.129 l gs 2 setlinejoin col-1
s gr
n 339 139 m 359 79 l gs col-1 s gr
n 345.718 99.871 m 359.000 79.000 l 357.103 103.666 l gs 2 setlinejoin col-1 s
gr
n 179 319 m 199 259 l gs col-1 s gr
n 185.718 279.871 m 199.000 259.000 l 197.103 283.666 l gs 2 setlinejoin col-1
s gr
n 339 259 m 359 319 l gs col-1 s gr
n 357.103 294.334 m 359.000 319.000 l 345.718 298.129 l gs 2 setlinejoin col-1
s gr
0.500 setlinewidth
1 setlinecap [1 3.000000] 3.000000 setdash
n 379 179 m 379 159 l 179 159 l 179 179 l
clp gs col-1 s gr
[] 0 setdash 0 setlinecap
1 setlinecap [1 3.000000] 3.000000 setdash
n 379 239 m 379 219 l 179 219 l 179 239 l
clp gs col-1 s gr
[] 0 setdash 0 setlinecap
2.000 setlinewidth
1 setlinecap [1 6.000000] 6.000000 setdash
n 219 184 m 219 209 l gs col-1 s gr
[] 0 setdash 0 setlinecap
n 222.000 197.000 m 219.000 209.000 l 216.000 197.000 l gs 2 setlinejoin col-1
s gr
1 setlinecap [1 6.000000] 6.000000 setdash
n 339 209 m 339 184 l gs col-1 s gr
[] 0 setdash 0 setlinecap
n 336.000 196.000 m 339.000 184.000 l 342.000 196.000 l gs 2 setlinejoin col-1
s gr
/Times-Bold findfont 16.00 scalefont setfont
159 64 m
gs 1 -1 scale (PUT) col-1 show gr
/Times-Bold findfont 16.00 scalefont setfont
344 64 m
gs 1 -1 scale (GET) col-1 show gr
/Times-Bold findfont 16.00 scalefont setfont
159 349 m
gs 1 -1 scale (STORE) col-1 show gr
/Times-Bold findfont 16.00 scalefont setfont
339 349 m
gs 1 -1 scale (LOAD) col-1 show gr
/Times-Bold findfont 16.00 scalefont setfont
89 199 m
gs 1 -1 scale (Window) col-1 show gr
/Times-Bold findfont 16.00 scalefont setfont
229 174 m
gs 1 -1 scale (public copy) col-1 show gr
/Times-Bold findfont 16.00 scalefont setfont
234 234 m
gs 1 -1 scale (private copy) col-1 show gr
$F2psEnd
%%EndDocument
75 4 a
endTexFig
541 1002 a Ft(Figure)h(4.1:)j(Sc)o(hematic)d(desciption)h(of)d(windo)o(w)75
1139 y(pro)q(cess,)g(and)h(a)f Fh(public)g Ft(cop)o(y)g(that)g(is)h(accessed)
f(b)o(y)h(RMA)f(op)q(erations)h(\(including)h(RMA)f(op)q(erations)75
1196 y(issued)h(b)o(y)f(the)h(lo)q(cal)g(pro)q(cess\).)k(This)c(is)f
(illustrated)i(in)f(Figure)f(4.7.3.)166 1252 y(Eac)o(h)h(call)h(to)f
Fk(MPI)p 506 1252 14 2 v 15 w(WINDO)o(W)p 717 1252 V 16 w(IN)g
Ft(or)g Fk(MPI)p 933 1252 V 16 w(WINDO)o(W)p 1145 1252 V 16
w(OUT)h Ft(completes)f(an)g Fh(ep)n(o)n(ch)g Ft(in)h(the)g(life-)75
1308 y(time)c(of)f(the)g(corresp)q(onding)h(windo)o(w.)20 b(A)12
b(call)i(to)d Fk(MPI)p 1036 1308 V 16 w(WINDO)o(W)p 1248 1308
V 16 w(OUT)i Ft(completes)g(a)f Fh(private)h Ft(ep)q(o)q(c)o(h)75
1365 y(for)i(the)g(windo)o(w;)g(a)g(call)h(to)f Fk(MPI)p 663
1365 V 15 w(WINDO)o(W)p 874 1365 V 17 w(IN)g Ft(complete)g(a)g
Fh(public)g Ft(ep)q(o)q(c)o(h)i(to)d(the)h(windo)o(w.)166 1421
y(During)20 b(a)g(priv)m(ate)h(ep)q(o)q(c)o(h,)g(a)f(windo)o(w)g(can)g(b)q(e)
h(accessed)f(or)g(up)q(dated)h(b)o(y)f(the)g(lo)q(cal)h(pro)q(cess)75
1478 y(using)15 b(load)g(and)g(store)f(op)q(erations,)h(without)f
(restrictions.)21 b(RMA)14 b(get)h(accesses)g(to)f(the)g(windo)o(w)h(are)75
1534 y(also)g(allo)o(w)o(ed.)20 b(Ho)o(w)o(ev)o(er,)14 b(RMA)i(up)q(dates,)f
(using)h(put)f(or)g(accum)o(ulate)h(calls,)f(are)g(erroneous.)166
1591 y(During)j(a)f(public)i(ep)q(o)q(c)o(h,)g(the)f(windo)o(w)f(can)h(b)q(e)
g(accessed)g(or)f(up)q(dated)h(b)o(y)g(RMA)f(op)q(erations,)75
1647 y(without)g(restrictions.)24 b(\(This)17 b(includes)i(RMA)e(calls)h
(e\013ected)f(b)o(y)f(the)h(lo)q(cal)h(pro)q(cess.\))24 b(Lo)q(cal)18
b(loads)75 1704 y(are)d(also)g(allo)o(w)o(ed.)20 b(Ho)o(w)o(ev)o(er,)14
b(lo)q(cal)i(stores)f(are)g(erroneous.)166 1760 y(A)e(put)h(or)e(accum)o
(ulate)i(call)h(\(during)e(a)g(public)j(ep)q(o)q(c)o(h\))e(up)q(dates)f(the)h
(public)h(cop)o(y)e(of)g(a)g(windo)o(w.)75 1817 y(The)18 b(up)q(dated)g(v)m
(alues)g(are)f(visible)j(to)c(get)h(calls)i(as)e(so)q(on)g(as)g(the)g(up)q
(date)h(completes)g(at)f(the)g(target)75 1873 y(windo)o(w.)i(On)13
b(the)g(other)g(hand,)g(the)g(corresp)q(onding)g(lo)q(cations)h(in)f(the)g
(priv)m(ate)g(cop)o(y)g(ha)o(v)o(e)f(unde\014ned)75 1929 y(v)m(alues)24
b(after)d(the)i(up)q(date.)41 b(The)23 b(call)h(to)d Fk(MPI)p
948 1929 V 16 w(WINDO)o(W)p 1160 1929 V 16 w(IN)h Ft(restores)g(coherence)h
(b)q(et)o(w)o(een)g(the)75 1986 y(priv)m(ate)17 b(and)g(public)h(copies,)f
(making)g(sure)f(that)g(all)h(up)q(dates)g(to)e(the)i(public)h(cop)o(y)f
(e\013ected)f(during)75 2042 y(the)f(public)j(ep)q(o)q(c)o(h)e(b)q(ecome)g
(visible)h(in)f(the)f(priv)m(ate)h(cop)o(y)l(,)f(and)g(accessible)j(b)o(y)d
(lo)q(cal)h(loads.)166 2099 y(A)i(store)g(call)h(\(during)g(a)f(priv)m(ate)h
(ep)q(o)q(c)o(h\))g(up)q(dates)f(the)h(priv)m(ate)g(cop)o(y)f(of)g(the)g
(windo)o(w.)29 b(Suc)o(h)75 2155 y(up)q(date)24 b(is)g(visible)h(to)e(lo)q
(cal)h(loads)g(as)e(so)q(on)i(as)e(completed.)45 b(On)24 b(the)g(other)e
(hand,)k(the)d(corre-)75 2212 y(sp)q(onding)f(lo)q(cations)f(in)g(the)f
(public)j(cop)o(y)d(ha)o(v)o(e)g(unde\014ned)i(v)m(alues)f(after)f(the)g(up)q
(date.)36 b(The)21 b(call)75 2268 y(to)c Fk(MPI)p 218 2268
V 16 w(WINDO)o(W)p 430 2268 V 16 w(OUT)h Ft(restores)f(coherence)i(b)q(et)o
(w)o(een)f(the)g(priv)m(ate)g(and)g(public)i(copies,)f(making)75
2325 y(sure)e(that)f(all)h(up)q(dates)g(to)f(the)h(priv)m(ate)g(cop)o(y)g
(e\013ected)g(during)g(the)g(priv)m(ate)g(ep)q(o)q(c)o(h)g(b)q(ecome)h
(visible)75 2381 y(in)e(the)f(public)j(cop)o(y)l(,)d(and)g(accessible)i(b)o
(y)e(RMA)h(calls.)166 2438 y(The)e(previous)i(rules)f(sp)q(ecify)g(ho)o(w)f
(load)h(and)f(store)g(accesses)h(in)o(teract)f(with)h(RMA)f(accesses.)20
b(It)75 2494 y(remains)c(to)g(sp)q(ecify)h(ho)o(w)e(RMA)h(accesses)g(in)o
(teract)g(with)g(eac)o(h)g(other.)22 b(The)16 b(only)h(nonob)o(vious)f(case)
75 2550 y(is)21 b(when)g(t)o(w)o(o)e(con\015icting)j(RMA)f(op)q(erations)g
(access)g(concurren)o(tly)g(the)g(same)f(windo)o(w)h(lo)q(cation.)75
2607 y(This)g(will)h(generally)f(lead)g(to)f(an)g(unde\014ned)j(result.)35
b(I.e.,)21 b(if)g(a)f(get)g(accesses)h(a)f(lo)q(cation)h(that)e(is)75
2663 y(concurren)o(tly)e(up)q(dated)g(b)o(y)g(a)f(put)g(or)g(an)h(accum)o
(ulate,)g(then)f(the)h(v)m(alue)g(returned)g(is)g(unde\014ned.)26
b(If)1967 46 y Fn(1)1967 103 y(2)1967 159 y(3)1967 215 y(4)1967
272 y(5)1967 328 y(6)1967 385 y(7)1967 441 y(8)1967 498 y(9)1959
554 y(10)1959 611 y(11)1959 667 y(12)1959 724 y(13)1959 780
y(14)1959 836 y(15)1959 893 y(16)1959 949 y(17)1959 1006 y(18)1959
1062 y(19)1959 1119 y(20)1959 1175 y(21)1959 1232 y(22)1959
1288 y(23)1959 1345 y(24)1959 1401 y(25)1959 1457 y(26)1959
1514 y(27)1959 1570 y(28)1959 1627 y(29)1959 1683 y(30)1959
1740 y(31)1959 1796 y(32)1959 1853 y(33)1959 1909 y(34)1959
1966 y(35)1959 2022 y(36)1959 2078 y(37)1959 2135 y(38)1959
2191 y(39)1959 2248 y(40)1959 2304 y(41)1959 2361 y(42)1959
2417 y(43)1959 2474 y(44)1959 2530 y(45)1959 2587 y(46)1959
2643 y(47)1959 2699 y(48)p eop
%%Page: 23 23
bop 75 -100 a Fl(4.7.)34 b(SEMANTICS)16 b(OF)f(RMA)g(OPERA)l(TIONS)853
b Ft(23)75 4 y
28417720 10230378 0 0 39797964 14603550 startTexFig
75 4 a
%%BeginDocument: windows_ops.eps
/$F2psDict 200 dict def
$F2psDict begin
$F2psDict /mtrx matrix put
/l {lineto} bind def
/m {moveto} bind def
/s {stroke} bind def
/n {newpath} bind def
/gs {gsave} bind def
/gr {grestore} bind def
/clp {closepath} bind def
/graycol {dup dup currentrgbcolor 4 -2 roll mul 4 -2 roll mul
4 -2 roll mul setrgbcolor} bind def
/col-1 {} def
/col0 {0 0 0 setrgbcolor} bind def
/col1 {0 0 1 setrgbcolor} bind def
/col2 {0 1 0 setrgbcolor} bind def
/col3 {0 1 1 setrgbcolor} bind def
/col4 {1 0 0 setrgbcolor} bind def
/col5 {1 0 1 setrgbcolor} bind def
/col6 {1 1 0 setrgbcolor} bind def
/col7 {1 1 1 setrgbcolor} bind def
end
/$F2psBegin {$F2psDict begin /$F2psEnteredState save def} def
/$F2psEnd {$F2psEnteredState restore end} def

$F2psBegin
0 setlinecap 0 setlinejoin
-171.0 270.0 translate 0.900 -0.900 scale
/Times-Bold findfont 16.00 scalefont setfont
204 69 m
gs 1 -1 scale (External epoch) col-1 show gr
/Times-Bold findfont 16.00 scalefont setfont
569 69 m
gs 1 -1 scale (Internal Epoch) col-1 show gr
n 679 179 m 679 159 l 619 159 l 619 179 l
clp gs 0.50 setgray fill gr
0.500 setlinewidth
1 setlinecap [1 3.000000] 3.000000 setdash
n 679 179 m 679 159 l 559 159 l 559 179 l
clp gs col-1 s gr
[] 0 setdash 0 setlinecap
0.000 setlinewidth
n 319 179 m 319 159 l 259 159 l 259 179 l
clp gs 0.50 setgray fill gr
0.500 setlinewidth
1 setlinecap [1 3.000000] 3.000000 setdash
n 319 179 m 319 159 l 199 159 l 199 179 l
clp gs col-1 s gr
[] 0 setdash 0 setlinecap
0.000 setlinewidth
n 499 179 m 499 159 l 439 159 l 439 179 l
clp gs 0.50 setgray fill gr
0.500 setlinewidth
1 setlinecap [1 3.000000] 3.000000 setdash
n 499 179 m 499 159 l 379 159 l 379 179 l
clp gs col-1 s gr
[] 0 setdash 0 setlinecap
0.000 setlinewidth
n 859 179 m 859 159 l 799 159 l 799 179 l
clp gs 0.50 setgray fill gr
0.500 setlinewidth
1 setlinecap [1 3.000000] 3.000000 setdash
n 859 179 m 859 159 l 739 159 l 739 179 l
clp gs col-1 s gr
[] 0 setdash 0 setlinecap
0.000 setlinewidth
n 859 239 m 859 219 l 799 219 l 799 239 l
clp gs 0.50 setgray fill gr
0.500 setlinewidth
1 setlinecap [1 3.000000] 3.000000 setdash
n 859 239 m 859 219 l 739 219 l 739 239 l
clp gs col-1 s gr
[] 0 setdash 0 setlinecap
0.000 setlinewidth
n 499 239 m 499 219 l 439 219 l 439 239 l
clp gs 0.50 setgray fill gr
0.500 setlinewidth
1 setlinecap [1 3.000000] 3.000000 setdash
n 499 239 m 499 219 l 379 219 l 379 239 l
clp gs col-1 s gr
[] 0 setdash 0 setlinecap
1.000 setlinewidth
n 322 60 m 352 60 l 352 54 l 371 66 l
352 77 l 352 71 l 322 71 l 322 60 l gs 0.00 setgray fill gr
gs col-1 s gr
n 320.000 68.000 m 322.000 60.000 l 324.000 68.000 l gs 2 setlinejoin col-1 s gr
n 504 59 m 534 59 l 534 53 l 553 65 l
534 76 l 534 70 l 504 70 l 504 59 l gs 0.00 setgray fill gr
gs col-1 s gr
n 502.000 67.000 m 504.000 59.000 l 506.000 67.000 l gs 2 setlinejoin col-1 s gr
n 682 60 m 712 60 l 712 54 l 731 66 l
712 77 l 712 71 l 682 71 l 682 60 l gs 0.00 setgray fill gr
gs col-1 s gr
n 680.000 68.000 m 682.000 60.000 l 684.000 68.000 l gs 2 setlinejoin col-1 s gr
0.500 setlinewidth
1 setlinecap [1 3.000000] 3.000000 setdash
n 559 239 m 559 219 l 679 219 l 679 239 l
clp gs col-1 s gr
[] 0 setdash 0 setlinecap
0.000 setlinewidth
n 619 219 m 619 219 l 619 219 l 619 219 l
clp gs 0.50 setgray fill gr
n 619 239 m 619 219 l 679 219 l 679 239 l
clp gs 0.50 setgray fill gr
n 599 239 m 599 219 l 559 219 l 559 239 l
clp gs 0.75 setgray fill gr
0.500 setlinewidth
1 setlinecap [1 3.000000] 3.000000 setdash
n 319 239 m 319 219 l 199 219 l 199 239 l
clp gs col-1 s gr
[] 0 setdash 0 setlinecap
0.000 setlinewidth
n 259 219 m 259 219 l 259 219 l 259 219 l
clp gs 0.50 setgray fill gr
n 779 239 m 779 219 l 739 219 l 739 239 l
clp gs 0.75 setgray fill gr
n 779 179 m 779 159 l 739 159 l 739 179 l
clp gs 0.75 setgray fill gr
2.000 setlinewidth
n 756.000 196.000 m 759.000 184.000 l 762.000 196.000 l gs 2 setlinejoin col-1
s gr
n 759 184 m 759 214 l gs 0.75 setgray fill gr
gs col-1 s gr
n 218 244 m 208 279 l gs col-1 s gr
n 214.181 268.286 m 208.000 279.000 l 208.412 266.638 l gs 2 setlinejoin col-1
s gr
n 239 149 m 229 114 l gs col-1 s gr
n 229.412 126.362 m 229.000 114.000 l 235.181 124.714 l gs 2 setlinejoin col-1
s gr
n 289 114 m 279 149 l gs col-1 s gr
n 285.181 138.286 m 279.000 149.000 l 279.412 136.638 l gs 2 setlinejoin col-1
s gr
n 644 149 m 649 114 l gs col-1 s gr
n 644.333 125.455 m 649.000 114.000 l 650.273 126.304 l gs 2 setlinejoin col-1
s gr
n 574 279 m 579 244 l gs col-1 s gr
n 574.333 255.455 m 579.000 244.000 l 580.273 256.304 l gs 2 setlinejoin col-1
s gr
n 644 244 m 654 279 l gs col-1 s gr
n 653.588 266.638 m 654.000 279.000 l 647.819 268.286 l gs 2 setlinejoin col-1
s gr
n 469 184 m 469 214 l gs 0.75 setgray fill gr
gs col-1 s gr
n 472.000 202.000 m 469.000 214.000 l 466.000 202.000 l gs 2 setlinejoin col-1
s gr
0.500 setlinewidth
n 319 239 m 319 219 l 259 219 l 259 239 l
clp gs col-1 s gr
n 599 179 m 599 159 l 559 159 l 559 179 l
clp gs col-1 s gr
/Times-Bold findfont 16.00 scalefont setfont
274 109 m
gs 1 -1 scale (PUT) col-1 show gr
/Times-Bold findfont 16.00 scalefont setfont
539 299 m
gs 1 -1 scale (STORE) col-1 show gr
/Times-Bold findfont 16.00 scalefont setfont
634 299 m
gs 1 -1 scale (LOAD) col-1 show gr
/Times-Bold findfont 16.00 scalefont setfont
634 109 m
gs 1 -1 scale (GET) col-1 show gr
/Times-Bold findfont 16.00 scalefont setfont
214 109 m
gs 1 -1 scale (GET) col-1 show gr
/Times-Bold findfont 16.00 scalefont setfont
189 299 m
gs 1 -1 scale (LOAD) col-1 show gr
/Times-Bold findfont 16.00 scalefont setfont
389 69 m
gs 1 -1 scale (WINDOW_IN) col-1 show gr
/Times-Bold findfont 16.00 scalefont setfont
744 69 m
gs 1 -1 scale (WINDOW_OUT) col-1 show gr
/Times-Roman findfont 12.00 scalefont setfont
269 234 m
gs 1 -1 scale (? ? ? ?) col-1 show gr
/Times-Roman findfont 12.00 scalefont setfont
569 174 m
gs 1 -1 scale (? ? ?) col-1 show gr
$F2psEnd
%%EndDocument
75 4 a
endTexFig
573 750 a Ft(Figure)15 b(4.2:)k(Ep)q(o)q(c)o(hs)d(and)f(windo)o(w)h(accesses)
75 882 y(a)e(put)h(or)f(accum)o(ulate)h(up)q(dates)g(a)f(lo)q(cation)h
(concurren)o(tly)g(with)g(another)f(put)h(or)f(accum)o(ulate,)h(then)75
939 y(the)i(lo)q(cation)h(is)g(left)g(with)f(an)h(unde\014ned)h(v)m(alue.)27
b(The)18 b(only)f(exception)i(is)e(the)h(case)f(of)g(concurren)o(t)75
995 y(accum)o(ulates)c(to)e(the)h(same)g(v)m(ariable,)i(with)e(the)h(same)e
(op)q(eration.)19 b(MPI)13 b(will)g(serialize)h(suc)o(h)f(up)q(dates,)75
1051 y(and)k(the)g(\014nal)h(v)m(alue)h(of)d(the)h(v)m(ariable)i(will)g(b)q
(e)e(as)g(if)h(the)f(up)q(dates)g(o)q(ccurred)h(in)g(some)f(serial)h(order.)
75 1108 y(Note,)13 b(ho)o(w)o(ev)o(er,)f(that)h(if)g(t)o(w)o(o)f(accum)o
(ulate)i(op)q(eration)f(up)q(dates)h(sev)o(eral)f(v)m(ariables)i(then)e(the)g
(up)q(dates)75 1164 y(of)k(di\013eren)o(t)g(v)m(ariables)h(ma)o(y)f(b)q(e)g
(serialized)i(in)f(a)f(di\013eren)o(t)g(order:)24 b(the)17
b(up)q(date)g(of)g(eac)o(h)g(v)m(ariable)h(is)75 1221 y(atomic,)h(but)g(the)g
(en)o(tire)h(set)e(of)h(up)q(dates)g(e\013ected)g(b)o(y)g(an)g(accum)o(ulate)
g(call)h(is)f(not)g(atomic.)31 b(P)o(ar-)75 1277 y(tially)19
b(o)o(v)o(erlaping)e(accum)o(ulates)h(\(e.g.,)f(an)g(accum)o(ulate)h(that)e
(up)q(dates)i(a)f(complex)i(n)o(um)o(b)q(er,)e(while)75 1334
y(another)h(concurren)o(t)g(accumlate)h(up)q(dates)g(the)g(real)f(part)g(of)g
(the)g(n)o(um)o(b)q(er\))h(are)f(erroneous.)29 b(Also,)75 1390
y(are)15 b(erroneous)g(concurren)o(t)g(accum)o(ulates)h(to)e(the)i(same)e(v)m
(ariable)j(with)f(distinct)g(op)q(erations.)166 1447 y(A)h(\014nal)g
(restriction:)23 b(if)17 b(t)o(w)o(o)f(distinct)h(windo)o(ws)g(o)o(v)o(erlap)
g(in)g(memory)f(then)h(it)g(is)g(erroneous)g(to)75 1503 y(ha)o(v)o(e)e
(concurren)o(t)g(public)i(ep)q(o)q(c)o(hs)f(for)f(b)q(oth)g(windo)o(ws.)166
1560 y(The)h(sc)o(hematic)g(co)q(des)g(in)h(Figure)f(4.3)e(illustrate)j
(correct)e(use)h(of)g(the)g(calls)g Fk(MPI)p 1605 1560 14 2
v 16 w(WINDO)o(W)p 1817 1560 V 16 w(IN)75 1616 y Ft(and)c Fk(MPI)p
245 1616 V 16 w(WINDO)o(W)p 457 1616 V 16 w(OUT)h Ft(for)f(get)g(after)f
(store,)h(put)g(after)g(store)f(and)i(load)f(after)g(put)g(dep)q(endencies.)p
74 1741 2 69 v 100 1712 a(Origin)p 100 1727 127 2 v 157 w(T)l(arget)p
382 1727 131 2 v 74 1810 2 69 v 382 1781 a(lo)q(cal)k(stores)p
74 1879 V 382 1849 a Fk(OUT)p 74 1935 2 57 v 382 1918 a Ft(some)e(sync.)8
b(to)p 74 1988 V 382 1971 a(guaran)o(tee...)p 74 2044 V 100
2027 a(...)191 b Fo(.)p 74 2100 V 100 2084 a Ft(b)q(eing)16
b(after)p 74 2165 2 69 v 100 2136 a(the)f(out)p 74 2222 2 57
v 100 2205 a Fk(GET)194 b Ft(lo)q(cal)16 b(loads)p 74 2274
V 382 2257 a(allo)o(w)o(ed)p 680 1741 2 69 v 706 1712 a(Origin)p
706 1727 127 2 v 156 w(T)l(arget)p 987 1727 131 2 v 680 1810
2 69 v 987 1781 a(lo)q(cal)h(stores)p 680 1879 V 987 1849 a
Fk(OUT)p 680 1935 2 57 v 987 1918 a Ft(some)e(sync.)8 b(to)p
680 1988 V 987 1971 a(guaran)o(tee...)p 680 2044 V 706 2027
a(...)191 b Fo(.)p 680 2100 V 706 2084 a Ft(b)q(eing)16 b(after)p
680 2165 2 69 v 706 2136 a(the)f(out)p 680 2222 2 57 v 706
2205 a Fk(PUT)190 b Ft(lo)q(cal)17 b(stores)p 680 2274 V 987
2257 a(forbidden)p 1286 1741 2 69 v 1311 1712 a(Origin)p 1311
1727 127 2 v 192 w(T)l(arget)p 1628 1727 131 2 v 1286 1810
2 69 v 1311 1781 a Fk(PUT)p 1286 1866 2 57 v 1311 1849 a Ft(RMA-sync.)p
1286 1919 V 1311 1902 a(to)e(guaran)o(tee)p 1286 1975 V 1577
1958 a Fo(&)6 b Ft(that)15 b(the)p 1286 2027 V 1628 2011 a(data)g(is)p
1286 2080 V 1628 2063 a(stored)g(in)o(to)p 1286 2132 V 1628
2115 a(the)h(target)p 1286 2197 2 69 v 1628 2167 a(windo)o(w)p
1286 2266 V 1628 2236 a Fk(IN)p 1286 2322 2 57 v 1628 2305
a Ft(lo)q(cal)h(load)279 2399 y(Figure)e(4.3:)k(Correct)14
b(use)i(of)f Fk(MPI)p 893 2399 14 2 v 15 w(WINDO)o(W)p 1104
2399 V 17 w(IN)f Ft(and)i Fk(MPI)p 1351 2399 V 16 w(WINDO)o(W)p
1563 2399 V 16 w(OUT)166 2503 y Ft(The)h(example)g(on)g(the)f(left)h(sho)o
(ws)f(a)g(get)g(that)g(uses)h(a)f(v)m(alue)i(written)f(b)o(y)f(a)g(store.)23
b(The)17 b(origin)75 2560 y(get)e(will)i(surely)f(return)f(the)g(v)m(alue)h
(writen)g(b)o(y)f(the)g(target)f(store)h(if)143 2647 y Fo(\017)23
b Ft(The)17 b(target)e(calls)j Fk(MPI)p 607 2647 V 16 w(WINDO)o(W)p
819 2647 V 16 w(OUT)g Ft(after)e(it)h(completes)h(the)f(store)f(and)h(b)q
(efore)g(the)g(get)189 2704 y(starts.)-32 46 y Fn(1)-32 103
y(2)-32 159 y(3)-32 215 y(4)-32 272 y(5)-32 328 y(6)-32 385
y(7)-32 441 y(8)-32 498 y(9)-40 554 y(10)-40 611 y(11)-40 667
y(12)-40 724 y(13)-40 780 y(14)-40 836 y(15)-40 893 y(16)-40
949 y(17)-40 1006 y(18)-40 1062 y(19)-40 1119 y(20)-40 1175
y(21)-40 1232 y(22)-40 1288 y(23)-40 1345 y(24)-40 1401 y(25)-40
1457 y(26)-40 1514 y(27)-40 1570 y(28)-40 1627 y(29)-40 1683
y(30)-40 1740 y(31)-40 1796 y(32)-40 1853 y(33)-40 1909 y(34)-40
1966 y(35)-40 2022 y(36)-40 2078 y(37)-40 2135 y(38)-40 2191
y(39)-40 2248 y(40)-40 2304 y(41)-40 2361 y(42)-40 2417 y(43)-40
2474 y(44)-40 2530 y(45)-40 2587 y(46)-40 2643 y(47)-40 2699
y(48)p eop
%%Page: 24 24
bop 75 -100 a Ft(24)723 b Fl(CHAPTER)15 b(4.)35 b(ONE)16 b(SIDED)f(COMMUNICA)
l(TIONS)189 49 y Ft(Some)e(sync)o(hronization)i(is)g(needed)g(to)e(ensure)h
(that)f(the)h(call)h(to)e(windo)o(w)p 1485 49 14 2 v 17 w(out)g(\(and)h(the)g
(previ-)189 106 y(ous)h(store\))g(precede)i(the)f(get.)22 b(The)16
b(can)g(b)q(e)g(ac)o(hiev)o(ed)h(b)o(y)f(ha)o(ving)g(the)g(target)f(pro)q
(cess)h(send)g(a)189 162 y(message)d(to)f(the)i(origin)g(pro)q(cess)g(\(as)e
(indicated)k(b)o(y)d(the)g(arro)o(w)f(in)j(the)e(\014gure\),)g(or)g(b)o(y)h
(a)f(barrier)189 219 y(sync)o(hronization,)i(or)g(b)o(y)g(another)g(mec)o
(hanism.)143 312 y Fo(\017)23 b Ft(An)o(y)11 b(other)h(put)f(or)h(accum)o
(ulate)g(that)f(up)q(dates)h(the)g(same)f(lo)q(cation)h(either)h(precedes)f
(the)g(ep)q(o)q(c)o(h)189 369 y(where)j(the)g(store)g(o)q(ccurs,)g(or)g
(follo)o(ws)g(the)g(get)g(call.)143 463 y Fo(\017)23 b Ft(An)o(y)13
b(other)g(store)f(that)h(up)q(dates)h(the)f(same)g(lo)q(cation)h(either)g
(precede)h(the)e(store)g(or)g(follo)o(ws)g(the)189 519 y(ep)q(o)q(c)o(h)j
(where)f(the)g(get)g(o)q(ccurs.)166 613 y(The)h(middle)i(example)f(sho)o(ws)e
(a)g(similar)j(case)e(of)f(a)h(put)g(that)f(o)o(v)o(erwrites)g(a)h(store.)21
b(The)16 b(origin)75 669 y(put)f(will)i(surely)f(o)o(v)o(erwrite)f(the)g
(target)f(store)g(if)143 763 y Fo(\017)23 b Ft(The)14 b(target)e(pro)q(cess)i
(calls)h Fk(MPI)p 755 763 V 16 w(WINDO)o(W)p 967 763 V 16 w(OUT)f
Ft(after)f(the)h(store)f(completes)h(and)g(b)q(efore)g(the)189
820 y(put)h(call)h(starts.)189 876 y(As)d(in)i(the)f(previous)h(example,)f
(some)g(sync)o(hronization)g(is)h(needed)g(to)e(ensure)i(that)e(the)h(call)h
(to)189 933 y(windo)o(w)p 342 933 V 16 w(out)g(and)g(the)h(previous)g(store)e
(precede)i(the)g(put.)166 1026 y(The)f(example)h(on)f(the)g(righ)o(t)f(sho)o
(ws)h(a)f(load)i(that)e(uses)h(a)g(v)m(alue)h(written)f(b)o(y)f(a)h(put.)20
b(The)15 b(target)75 1083 y(load)g(will)i(surely)f(return)f(the)h(v)m(alue)g
(written)f(b)o(y)g(the)h(store)e(if)143 1177 y Fo(\017)23 b
Ft(The)14 b(target)f(calls)i Fk(MPI)p 599 1177 V 16 w(WINDO)o(W)p
811 1177 V 16 w(IN)f Ft(after)f(the)h(put)h(completes)f(at)g(the)g(target)f
(and)h(b)q(efore)h(it)189 1233 y(issues)h(the)f(load.)189 1290
y(The)d(target)e(pro)q(cess)i(can)g(v)o(erify)f(that)g(the)h(put)g(has)f
(completed)i(in)f(its)g(memory)f(b)q(efore)h(it)g(issues)189
1346 y(the)h(call)h(to)f(windo)o(w)p 553 1346 V 17 w(in)h(b)o(y)f(c)o(hec)o
(king)h(the)f(v)m(alue)i(of)d(the)i(windo)o(w)f(coun)o(ter)g(\(assuming)g
(that)g(the)189 1403 y(put)18 b(call)i(incremen)o(ts)f(this)g(coun)o(ter\).)
29 b(In)19 b(this)g(case,)g(the)f(comm)o(unication)h(indicated)i(b)o(y)d(an)
189 1459 y(arro)o(w)11 b(in)j(the)f(\014gure)h(is)f(implicit)j({)d(it)g(is)g
(the)h(coun)o(ter)f(incremen)o(t)g(command)g(that)g(piggy-bac)o(ks)189
1515 y(on)19 b(the)g(data)g(transfer)g(e\013ected)g(b)o(y)h(the)f(put.)33
b(Alternativ)o(ely)l(,)21 b(the)f(origin)g(pro)q(cess)f(and)h(the)189
1572 y(target)15 b(pro)q(cess)h(can)h(sync)o(hronize)g(using)g(a)f(barrier)h
(sync)o(hronization.)24 b(A)16 b(third)h(\(inne\014cien)o(t\))189
1628 y(c)o(hoice)f(is)g(that)f(the)h(origin)g(pro)q(cess)g(executes)h(a)e
(fence)h(call)h(after)e(the)h(put,)f(and)h(next)g(sends)g(a)189
1685 y(message)e(to)h(the)g(target.)143 1779 y Fo(\017)23 b
Ft(An)o(y)11 b(other)g(put)h(or)e(accum)o(ulate)i(to)f(the)g(same)h(lo)q
(cation)g(either)g(precedes)g(the)g(put)f(of)g(the)h(origin,)189
1835 y(or)i(follo)o(ws)i(the)f(ep)q(o)q(c)o(h)h(where)f(the)h(load)f(o)q
(ccurs.)143 1929 y Fo(\017)23 b Ft(An)o(y)17 b(other)g(store)f(to)h(that)g
(lo)q(cation)h(either)g(precedes)g(the)f(ep)q(o)q(c)o(h)h(of)f(the)g(put)h
(or)f(follo)o(ws)g(the)189 1985 y(target)d(load.)166 2162 y
Fg(Alternativ)o(es)o(:)37 b Fj(A)15 b(p)q(ossible)g(alternativ)o(e)f(mo)q
(del)f(is)h(to)h(go)f(all)f(the)i(w)o(a)o(y)f(to)o(w)o(ard)g(dela)o(y)o(ed)g
(consistency)m(.)75 2218 y(Namely)m(,)e(up)q(dates)j(to)g(a)f(windo)o(w,)f
(either)i(lo)q(cal)f(or)g(global,)e(are)j(not)f(\(necessarily\))i(visible)e
(to)g(accesses)j(to)d(the)75 2275 y(windo)o(w,)e(un)o(til)g(a)h(call)f(to)h
(windo)o(w)p 633 2275 13 2 v 15 w(in/out)f(o)q(ccurred.)20
b(The)13 b(di\013erence)i(is)e(that)g(in)g(the)h(curren)o(t)g(design)g(a)f(v)
n(alue)75 2331 y(stored)18 b(b)o(y)g(a)f(put)h(m)o(ust)e(b)q(e)i(view)o(ed)g
(b)o(y)f(a)g(subsequen)o(t)j(get)d(ev)o(en)h(if)f(no)g(MPI)h(call)f(o)q
(ccurred)i(on)e(the)h(target)75 2388 y(pro)q(cess.)i(In)14
b(the)g(alternativ)o(e)f(design,)h(some)f(windo)o(w)p 949 2388
V 14 w(in/out)g(call)g(has)h(to)f(o)q(ccur)i(at)f(target)g(pro)q(cess)h(A)f
(b)q(efore)75 2444 y(a)g(put)g(b)o(y)f(pro)q(cess)j(B)e(can)g(b)q(e)h(seen)g
(b)o(y)f(a)f(get)i(from)d(pro)q(cess)j(C.)189 2633 y Fh(R)n(ationale.)73
b Ft(The)21 b(underlying)h(assumption)f(in)h(this)f(design)g(is)h(that)d
(con\015icting)k(accesses)189 2689 y(to)f(data)g(are)h(alw)o(a)o(ys)f(sync)o
(hronized;)28 b(sync)o(hronization)c(b)q(et)o(w)o(een)f(pro)q(cesses)h(is)f
(done)h(with)1967 46 y Fn(1)1967 103 y(2)1967 159 y(3)1967
215 y(4)1967 272 y(5)1967 328 y(6)1967 385 y(7)1967 441 y(8)1967
498 y(9)1959 554 y(10)1959 611 y(11)1959 667 y(12)1959 724
y(13)1959 780 y(14)1959 836 y(15)1959 893 y(16)1959 949 y(17)1959
1006 y(18)1959 1062 y(19)1959 1119 y(20)1959 1175 y(21)1959
1232 y(22)1959 1288 y(23)1959 1345 y(24)1959 1401 y(25)1959
1457 y(26)1959 1514 y(27)1959 1570 y(28)1959 1627 y(29)1959
1683 y(30)1959 1740 y(31)1959 1796 y(32)1959 1853 y(33)1959
1909 y(34)1959 1966 y(35)1959 2022 y(36)1959 2078 y(37)1959
2135 y(38)1959 2191 y(39)1959 2248 y(40)1959 2304 y(41)1959
2361 y(42)1959 2417 y(43)1959 2474 y(44)1959 2530 y(45)1959
2587 y(46)1959 2643 y(47)1959 2699 y(48)p eop
%%Page: 25 25
bop 75 -100 a Fl(4.7.)34 b(SEMANTICS)16 b(OF)f(RMA)g(OPERA)l(TIONS)853
b Ft(25)189 49 y(message-passing)19 b(calls,)i(suc)o(h)f(as)f(send,)h(receiv)
o(e)g(or)f(barrier,)h(or)f(with)h(sync)o(hronization)g(ob-)189
106 y(jects,)c(but)g(not)g(with)g(put/get/accum)o(ulate)g(calls.)24
b(This)17 b(allo)o(ws)f(us)h(to)e(use)i(a)f(dela)o(y)o(ed)h(coher-)189
162 y(ence)h(mo)q(del,)g(where)f(coherence)h(need)g(b)q(e)g(restored)f(only)h
(at)e(sync)o(hronization)i(p)q(oin)o(ts.)26 b(Suc)o(h)189 219
y(mo)q(del)13 b(can)f(b)q(e)h(implemen)o(ted)g(more)f(e\016cien)o(tly)h(on)f
(systems)g(that)f(ha)o(v)o(e)h(hardw)o(are)f(supp)q(ort)i(for)189
275 y(remote)g(memory)g(access,)h(but)g(v)m(arious)g(limitations)h(on)f(the)g
(coherence)h(or)e(atomicit)o(y)h(of)f(those)189 332 y(accesses)18
b(\(see)g(the)h(implemen)o(tation)g(note)f(b)q(elo)o(w\).)30
b(One)19 b(could)g(sp)q(eci\014y)h(that)d(coherence)i(is)189
388 y(implicitly)f(restored)c(at)g(an)o(y)g(p)q(oten)o(tial)i(sync)o
(hronization)g(p)q(oin)o(t,)f(e.g.,)e(at)i(eac)o(h)f(MPI)h(call,)h(but)189
444 y(that)h(w)o(ould)h(b)q(e)h(exp)q(ensiv)o(e.)29 b(Instead,)19
b(the)f(user)g(is)g(ask)o(ed)g(to)f(insert)i(the)f(coherence)h(calls)g(at)189
501 y(appropriate)c(p)q(oin)o(ts.)20 b(\()p Fh(End)c(of)g(r)n(ationale.)p
Ft(\))189 607 y Fh(A)n(dvic)n(e)g(to)j(users.)50 b Ft(An)17
b(execution)h(ma)o(y)e(b)q(e)i(sign\014can)o(tly)f(slo)o(w)o(ed)g(b)o(y)g
(tests)f(that)h(v)o(erify)g(that)189 664 y(a)f(program)g(ob)q(eys)g(the)h
(restrictions)g(on)g(windo)o(w)g(use)g(stated)f(in)h(this)g(section.)25
b(Therefore,)17 b(it)189 720 y(is)f(exp)q(ected)h(tha)f(man)o(y)f(implemen)o
(tations)j(will)f(not)f(test)f(for)h(erroneous)f(use)i(of)e(ep)q(o)q(c)o(hs,)
i(and)189 777 y(the)e(b)q(eha)o(vior)g(of)g(suc)o(h)g(erroneous)g(programs)f
(will)j(b)q(e)f(unde\014ned.)22 b(\()p Fh(End)15 b(of)h(advic)n(e)g(to)h
(users.)p Ft(\))189 883 y Fh(A)n(dvic)n(e)e(to)i(implementors.)189
958 y Ft(Case)d(1:)189 1014 y(A)21 b(distributed)h(memory)e(implemen)o
(tation,)k(where)d(RMA)g(accesses)g(are)g(p)q(erformed)g(b)o(y)g(an)189
1071 y(agen)o(t)14 b(running)i(on)g(the)f(target)f(pro)q(cess.)189
1146 y(Then)h(the)g(calls)h(to)e Fk(MPI)p 626 1146 14 2 v 16
w(WINDO)o(W)p 838 1146 V 16 w(IN)h Ft(and)g Fk(MPI)p 1084 1146
V 16 w(WINDO)o(W)p 1296 1146 V 16 w(OUT)h Ft(are)e(no)q(ops,)h(and)g(the)g(t)
o(w)o(o)189 1202 y(windo)o(w)g(copies)h(are)f(iden)o(tical.)22
b(No)15 b(lo)q(c)o(king)h(is)g(required)189 1278 y(Case)e(2.1:)189
1334 y(A)e(cac)o(he)g(coheren)o(t)g(shared)g(memory)f(implemen)o(tation,)j
(where)e(RMA)g(accesses)g(are)g(p)q(erformed)189 1390 y(b)o(y)j(the)g(origin)
h(pro)q(cess,)f(and)g(b)o(yte)g(and)h(half-w)o(ord)f(writes)g(are)g(supp)q
(orted.)189 1466 y(If)h(the)g(target)f(windo)o(w)h(is)h(in)f(shared)g(memory)
l(,)g(then)g(the)g(pro)q(cess)h(can)f(directly)h(access)f(it)h(for)189
1522 y(put)e(and)g(get)g(op)q(erations;)g(no)g(lo)q(c)o(ks)g(are)g(required)h
(for)f(puts)g(and)g(gets.)20 b(Accum)o(ulate)15 b(up)q(dates)189
1579 y(need)f(b)q(e)h(protected)e(b)o(y)h(lo)q(c)o(ks.)20 b(One)14
b(can)g(ha)o(v)o(e)g(one)g(lo)q(c)o(k)g(p)q(er)g(windo)o(w,)g(or)f(one)h(can)
g(ha)o(v)o(e)g(lo)q(c)o(ks)189 1635 y(that)j(protect)g(parts)g(of)g(the)h
(windo)o(w,)h(with)f(the)g(usual)g(tradeo\013:)24 b(\014ner)18
b(lo)q(c)o(ks)g(supp)q(ort)g(more)189 1691 y(concurrency)e(but)f(increase)h
(lo)q(c)o(king)g(o)o(v)o(erheads.)189 1767 y(If)e(the)g(target)f(windo)o(w)h
(is)g(not)g(in)h(shared)f(memory)l(,)f(then)i(a)e(public)j(windo)o(w)f(cop)o
(y)e(is)i(allo)q(cated)189 1823 y(in)20 b(shared)h(memory)l(.)33
b(Put,)21 b(get)e(and)i(accum)o(ulate)f(accesses)g(\(including)j(RMA)d
(accesses)g(b)o(y)189 1879 y(the)g(windo)o(w)h(o)o(wner\))f(use)h(the)f
(public)j(cop)o(y)l(,)e(whereas)g(lo)q(cal)g(load)g(and)g(store)f(accesses)h
(use)189 1936 y(the)15 b(priv)m(ate)h(cop)o(y)l(.)21 b(A)16
b(call)g(to)f Fk(MPI)p 815 1936 V 16 w(WINDO)o(W)p 1027 1936
V 16 w(IN)g Ft(copies)i(the)e(public)j(cop)o(y)d(on)o(to)g(the)g(priv)m(ate)
189 1992 y(cop)o(y)f(\(only)g(lo)q(cations)h(that)f(w)o(ere)f(up)q(dated)i
(in)h(the)e(last)g(ep)q(o)q(c)o(h)h(need)g(b)q(e)g(copied,)g(but)f(cop)o
(ying)189 2049 y(the)g(en)o(tire)h(windo)o(w)g(will)h(not)e(cause)g
(problems\).)20 b(A)15 b(call)g(to)f Fk(MPI)p 1344 2049 V 16
w(WINDO)o(W)p 1556 2049 V 16 w(OUT)h Ft(copies)g(the)189 2105
y(priv)m(ate)f(cop)o(y)g(on)o(to)f(the)h(public)i(cop)o(y)e(\(only)g(lo)q
(cations)h(that)e(w)o(ere)h(up)q(dated)g(in)h(the)f(last)g(ep)q(o)q(c)o(h)189
2162 y(need)i(b)q(e)g(copied,)g(but)f(cop)o(ying)h(the)f(en)o(tire)h(windo)o
(w)f(causes)g(no)g(harm.\))189 2237 y(V)l(arious)20 b(optimizations)h
(\(similar)g(to)e(those)h(used)h(in)g(DSVM)e(systems\))h(can)g(b)q(e)h
(applied)h(to)189 2293 y(a)o(v)o(oid)15 b(full)h(windo)o(w)g(cop)o(ying.)189
2368 y(Case)e(2.2:)189 2425 y(A)e(cac)o(he)g(coheren)o(t)g(shared)g(memory)f
(implemen)o(tation,)j(where)e(RMA)g(accesses)g(are)g(p)q(erformed)189
2481 y(b)o(y)j(the)g(origin)h(pro)q(cess,)f(but)g(only)h(w)o(ord)f(writes)g
(are)g(supp)q(orted.)189 2557 y(One)d(can)f(use)g(the)g(same)g(implemen)o
(tation)i(as)e(b)q(efore.)18 b(F)l(ull)13 b(w)o(ord)d(write)h(need)h(not)f(b)
q(e)h(protected)189 2613 y(b)o(y)18 b(lo)q(c)o(ks,)h(since)h(there)e(can)h(b)
q(e)g(no)f(concurren)o(t,)h(o)o(v)o(erlaping)g(up)q(date.)30
b(On)19 b(the)g(other)f(hand,)189 2669 y(partial)c(w)o(ord)f(up)q(dates,)i
(that)e(require)i(a)f(read-mo)q(dify-write)h(cycle,)g(ha)o(v)o(e)f(to)f(b)q
(e)i(protected)f(b)o(y)-32 46 y Fn(1)-32 103 y(2)-32 159 y(3)-32
215 y(4)-32 272 y(5)-32 328 y(6)-32 385 y(7)-32 441 y(8)-32
498 y(9)-40 554 y(10)-40 611 y(11)-40 667 y(12)-40 724 y(13)-40
780 y(14)-40 836 y(15)-40 893 y(16)-40 949 y(17)-40 1006 y(18)-40
1062 y(19)-40 1119 y(20)-40 1175 y(21)-40 1232 y(22)-40 1288
y(23)-40 1345 y(24)-40 1401 y(25)-40 1457 y(26)-40 1514 y(27)-40
1570 y(28)-40 1627 y(29)-40 1683 y(30)-40 1740 y(31)-40 1796
y(32)-40 1853 y(33)-40 1909 y(34)-40 1966 y(35)-40 2022 y(36)-40
2078 y(37)-40 2135 y(38)-40 2191 y(39)-40 2248 y(40)-40 2304
y(41)-40 2361 y(42)-40 2417 y(43)-40 2474 y(44)-40 2530 y(45)-40
2587 y(46)-40 2643 y(47)-40 2699 y(48)p eop
%%Page: 26 26
bop 75 -100 a Ft(26)723 b Fl(CHAPTER)15 b(4.)35 b(ONE)16 b(SIDED)f(COMMUNICA)
l(TIONS)189 49 y Ft(lo)q(c)o(ks,)d(since)h(they)f(can)g(b)q(e)h(concurren)o
(t)f(with)g(another)g(nono)o(v)o(erlaping)g(partial)h(w)o(ord)e(up)q(date)i
(to)189 106 y(the)i(same)g(w)o(ord.)k(\(Only)d(up)q(dates)g(that)e(can)i
(su\013er)e(from)h(false)g(sharing)h(need)g(b)q(e)g(protected.\))189
180 y(An)f(additional)i(complexit)o(y)g(o)q(ccurs)f(with)g(windo)o(w)f(b)q
(oundaries)i(that)e(are)g(not)g(w)o(ord)g(aligned.)189 237
y(If)c(p)q(ossible,)i(suc)o(h)f(windo)o(ws)f(should)h(b)q(e)g(padded)g(to)e
(the)h(next)g(w)o(ord)g(b)q(oundary)g(so)g(that)f(up)q(dates)189
293 y(inside)19 b(the)e(windo)o(w)g(do)h(not)e(in)o(terfere)i(with)f(lo)q
(cations)h(outside)g(the)f(windo)o(w.)27 b(If)17 b(this)h(is)f(not)189
349 y(p)q(ossible,)e(then)e(the)g(system)g(has)g(to)g(main)o(tain)g(a)g
(public)j(cop)o(y)d(of)f(the)i(w)o(ord\(s\))d(that)i(o)o(v)o(erlap\(s\))189
406 y(the)h(windo)o(w)g(b)q(oundary)l(.)20 b(RMA)14 b(accesses)g(to)f(this)i
(\(these\))e(w)o(ord\(s\))g(use)h(the)g(public)i(cop)o(y)l(,)e(and)189
462 y(coherence)i(is)g(restored)e(when)i(calls)g(to)f Fk(MPI)p
977 462 14 2 v 16 w(WINDO)o(W)p 1189 462 V 16 w(IN)g Ft(or)f
Fk(MPI)p 1402 462 V 16 w(WINDO)o(W)p 1614 462 V 16 w(OUT)i
Ft(o)q(ccur.)189 537 y(Case)e(3:)189 593 y(A)k(noncoheren)o(t)h(shared)f
(memory)g(implemen)o(tation,)i(or)e(an)h(implemen)o(tation)g(using)h(nonco-)
189 650 y(heren)o(t)15 b(DMA)g(facilities.)22 b(There)15 b(are)g(sev)o(eral)g
(cases)g(to)g(consider)231 742 y(1.)22 b(RMA)f(is)h(implemen)o(ted)h(using)g
(non-coheren)o(t)e(shared)h(memory)l(,)g(with)g(write-through)289
798 y(noncoheren)o(t)c(cac)o(hes.)30 b(Both)18 b(lo)q(cal)i(and)f(remote)e
(writes)i(go)f(directly)i(to)d(memory)l(,)i(but)289 855 y(b)q(oth)c(lo)q(cal)
h(and)g(remote)e(reads)h(ma)o(y)g(go)f(to)h(cac)o(he)g(\(Cra)o(y)f(T90,)g
(Cra)o(y)h(T3D?\).)231 926 y(2.)22 b(RMA)16 b(is)i(implemen)o(ted)g(using)g
(non-coheren)o(t)f(shared)f(memory)l(,)h(with)g(write-bac)o(k)g(non-)289
983 y(coheren)o(t)g(cac)o(hes.)28 b(Both)17 b(lo)q(cal)i(and)f(remote)f
(accesses)h(ma)o(y)f(go)g(to)g(cac)o(he,)h(rather)f(than)289
1039 y(memory)d(\(NEC)h(Cenju-3,)g(Mercury)g(RA)o(CE)g(i860?\).)231
1111 y(3.)22 b(RMA)d(is)h(implemen)o(ted)h(using)f(a)f(non-coheren)o(t)h(DMA)
e(engine.)34 b(Remote)19 b(accesses)h(go)289 1167 y(directly)g(to)e(memory)l
(,)h(whereas)g(lo)q(cal)h(accesses)f(\(b)q(oth)g(reads)g(and)g(writes\))f(ma)
o(y)g(go)h(to)289 1224 y(cac)o(he)c(\(Cra)o(y)f(T3D?\).)189
1316 y(In)h(these)g(cases,)f(the)h(public)i(cop)o(y)d(of)h(the)g(windo)o(w)f
(is)i(in)f(memory)l(,)f(whereas)h(the)g(priv)m(ate)g(cop)o(y)189
1373 y(of)f(the)i(windo)o(w)f(ma)o(y)g(b)q(e)g(partially)i(con)o(tained)e(in)
h(the)g(lo)q(cal)g(pro)q(cessor)f(cac)o(he.)189 1447 y(Assume)d(that)f(the)h
(windo)o(w)h(is)f(cac)o(he)g(line)i(aligned.)20 b(A)12 b(call)i(to)d
Fk(MPI)p 1368 1447 V 16 w(WINDO)o(W)p 1580 1447 V 16 w(OUT)i
Ft(is)f(a)g(no)q(op)189 1503 y(in)j(the)f(\014rst)h(case)f(\(except)h(that)e
(a)i(memory)f(sync)g(ma)o(y)g(b)q(e)h(needed)h(to)e(force)g(the)h(stores)e
(out)h(of)189 1560 y(the)j(store)g(queue,)h(on)g(some)f(systems\).)25
b(In)19 b(the)e(second)h(and)g(third)g(case,)f(the)h(call)g(will)h(\015ush)
189 1616 y(the)14 b(windo)o(w)h(cac)o(he)g(lines)i(from)d(the)h(calling)h
(pro)q(cessor)f(cac)o(he)g(to)f(memory)g(\(it)h(is)g(su\016cien)o(t)h(to)189
1673 y(\015ush)f(dirt)o(y)h(lines\).)189 1747 y(A)j(call)i(to)e
Fk(MPI)p 475 1747 V 15 w(WINDO)o(W)p 686 1747 V 16 w(IN)h Ft(will)h(in)o(v)m
(alidate)g(the)f(windo)o(w)f(cac)o(he)h(lines)h(from)e(the)g(calling)189
1804 y(pro)q(cessor)d(cac)o(he)g(\(it)g(is)h(su\016cien)o(t)g(to)e(in)o(v)m
(alidate)j(lines)g(that)e(w)o(ere)g(remotely)g(up)q(dated)h(in)g(the)189
1860 y(last)e(ep)q(o)q(c)o(h\).)20 b(Since)d(no)e(lo)q(cal)i(up)q(dates)e(w)o
(ere)g(p)q(erformed)h(in)g(the)f(last)g(ep)q(o)q(c)o(h,)h(all)g(the)g(windo)o
(w)189 1916 y(cac)o(he)22 b(lines)h(are)e(clean,)j(and)e(a)f(\015ush)h(call)h
(ma)o(y)e(b)q(e)h(used,)h(instead)g(of)e(an)g(in)o(v)m(alidate.)41
b(\(A)189 1973 y(\\\015ush")21 b(copies)h(bac)o(k)e(the)h(dirt)o(y)g(lines)i
(to)d(memory)h(and)g(in)o(v)m(alidate)i(the)e(en)o(tries)g(in)h(cac)o(he;)189
2029 y(an)17 b(\\in)o(v)m(alidate")i(do)q(es)f(not)g(cop)o(y)f(bac)o(k)h(the)
f(lines)j(in)e(memory)l(.)27 b(On)18 b(some)g(systems,)f(\015ush)h(is)189
2086 y(allo)o(w)o(ed)c(in)g(user)g(mo)q(de,)g(whereas)g(in)o(v)m(alidate)h
(requires)g(sup)q(ervisory)f(mo)q(de,)g(so)g(that)f(a)g(\015ush)h(is)189
2142 y(c)o(heap)q(er.)20 b(W)l(e)14 b(assume)h(that)f(the)g(\015ush)h
(implemen)o(tation)h(is)f(smart)f(enough)h(not)f(to)g(cop)o(y)g(bac)o(k)189
2199 y(to)g(memory)h(clean)h(lines.\))189 2273 y(In)h(addition,)g(in)g(the)f
(second)h(case,)g(the)f(co)q(de)h(for)f(put,)g(accum)o(ulate)h(or)e(RMW)i
(requires)g(that)189 2330 y(the)f(written)h(data)f(b)q(e)h(\015ushed)h(from)e
(the)h(origin)g(cac)o(he)g(b)q(efore)g(the)g(RMA)g(call)g(is)h(allo)o(w)o(ed)
f(to)189 2386 y(complete)c(at)f(the)h(target.)18 b(In)c(the)f(\014rst)f
(case,)h(a)g(memory)f(sync)h(ma)o(y)f(b)q(e)i(needed)g(to)e(\015ush)i(stores)
189 2442 y(out)g(of)h(the)g(store)g(queue.)189 2517 y(The)k(problems)g(of)g
(partial)g(cac)o(he)g(line)i(up)q(dates)e(and)g(of)f(cac)o(he)i(lines)g(that)
e(o)o(v)o(erlap)h(windo)o(w)189 2573 y(b)q(oundaries)d(are)f(solv)o(ed)h(as)e
(in)i(previous)g(Case)f(2.2.)189 2648 y(\()p Fh(End)g(of)i(advic)n(e)f(to)g
(implementors.)p Ft(\))-117 2704 y Fo(?)f Fn(\(July\))1967
46 y(1)1967 103 y(2)1967 159 y(3)1967 215 y(4)1967 272 y(5)1967
328 y(6)1967 385 y(7)1967 441 y(8)1967 498 y(9)1959 554 y(10)1959
611 y(11)1959 667 y(12)1959 724 y(13)1959 780 y(14)1959 836
y(15)1959 893 y(16)1959 949 y(17)1959 1006 y(18)1959 1062 y(19)1959
1119 y(20)1959 1175 y(21)1959 1232 y(22)1959 1288 y(23)1959
1345 y(24)1959 1401 y(25)1959 1457 y(26)1959 1514 y(27)1959
1570 y(28)1959 1627 y(29)1959 1683 y(30)1959 1740 y(31)1959
1796 y(32)1959 1853 y(33)1959 1909 y(34)1959 1966 y(35)1959
2022 y(36)1959 2078 y(37)1959 2135 y(38)1959 2191 y(39)1959
2248 y(40)1959 2304 y(41)1959 2361 y(42)1959 2417 y(43)1959
2474 y(44)1959 2530 y(45)1959 2587 y(46)1959 2643 y(47)1959
2699 y(48)p eop
%%Page: 27 27
bop 75 -100 a Fl(4.8.)34 b(SYNCHR)o(ONIZA)l(TION)17 b(OPERA)l(TIONS)876
b Ft(27)75 49 y Fk(Registers)16 b(and)h(Com)o(piler)c(Optimizations)44
b Ft(Another)15 b(coherence)h(problem)g(exists)g(b)q(et)o(w)o(een)g(v)m
(ariables)75 106 y(k)o(ept)21 b(in)g(registers)g(and)g(the)f(memory)h(v)m
(alue)h(of)e(these)h(v)m(ariables.)38 b(An)21 b(RMA)g(call)g(ma)o(y)f(access)
h(a)75 162 y(v)m(ariable)c(in)g(memory)f(\(or)f(cac)o(he\),)h(while)h(the)f
(up-to-date)g(v)m(alue)i(of)d(this)i(v)m(ariable)g(is)g(in)g(register.)22
b(A)75 219 y(get)12 b(will)j(not)d(return)g(the)h(latest)g(v)m(ariable)h(v)m
(alue,)g(and)f(a)f(put)h(ma)o(y)f(b)q(e)h(o)o(v)o(erwritten)f(when)h(the)g
(register)75 275 y(is)21 b(stored)f(bac)o(k)h(in)h(memory)l(.)36
b(This)21 b(problem,)h(whic)o(h)g(also)e(a\017icts)h(in)h(some)e(cases)h
(send/receiv)o(e)75 332 y(comm)o(unication,)16 b(is)f(discussed)i(more)e(at)f
(length)i(in)g(Section)g Fm(??)p Ft(.)166 388 y(MPI)k(implemen)o(tations)i
(will)f(a)o(v)o(oid)f(this)h(problem)g(for)e(standard)h(conforming)h(C)f
(programs.)75 444 y(High)13 b(qualit)o(y)h(MPI)f(F)l(ortran)e(implemen)o
(tations)j(will)h(a)o(v)o(oid)d(this)i(problem)f(for)g(standard)f(conforming)
75 501 y(F)l(ortran)i(programs)h(as)g(w)o(ell,)h(without)g(disabling)h
(compiler)g(optimizations.)22 b(Ho)o(w)o(ev)o(er,)14 b(MPI)h(imple-)75
557 y(men)o(tations)h(are)h(not)f(required)i(to)e(a)o(v)o(oid)h(altogether)f
(the)h(register)f(coherence)i(problem,)g(since)g(this)75 614
y(requires)d(compiler)h(supp)q(ort)e(that)g(is)h(not)f(curren)o(tly)h(a)o(v)m
(ailable)h(on)f(most)e(F)l(ortran)g(en)o(vironmen)o(ts.)20
b(In)75 670 y(order)14 b(to)f(a)o(v)o(oid)g(register)h(coherence)h(problems,)
f(users)g(should)h(restrict)e(their)h(use)h(of)e(RMA)h(windo)o(ws)75
727 y(to)h(v)m(ariables)h(stored)f(in)h Fi(COMMON)f Ft(blo)q(c)o(ks,)h(or)e
(to)h(v)m(ariables)i(that)d(w)o(ere)h(declared)i Fi(V)o(OLA)m(TILE)d
Ft(\(while)75 783 y Fi(V)o(OLA)m(TILE)e Ft(is)h(no)g(a)f(standard)h(F)l
(ortran)f(declaration,)h(it)h(is)f(supp)q(orted)g(b)o(y)g(man)o(y)g(F)l
(ortran)e(compilers\).)75 840 y(Thenm)18 b(the)g(use)f(of)h
Fk(MPI)p 535 840 14 2 v 15 w(WINDO)o(W)p 746 840 V 16 w(IN)g
Ft(and)g Fk(MPI)p 998 840 V 15 w(WINDO)o(W)p 1209 840 V 16
w(OUT)h Ft(will)g(prev)o(en)o(t)e(register)h(coher-)75 896
y(ence)i(problems.)33 b(Details,)21 b(and)f(an)f(additional)i(solution)f(are)
f(discussed)i(in)f(Section)h Fm(??)32 b Ft(on)19 b(page)75
953 y Fm(??)p Ft(.)75 1096 y Fq(4.8)59 b(Synchronization)20
b(op)r(erations)1890 1152 y Fo(>)c Fn(\(July\))75 1197 y Ft(MPI)i(pro)o(vide)
g(Read)p 450 1197 V 17 w(Mo)q(dify)p 608 1197 V 16 w(W)l(rite)g(op)q
(erations)g(on)f(opaque)h(sync)o(hronization)h(ob)s(jects.)26
b(that)17 b(can)75 1254 y(b)q(e)f(used)g(to)e(implemen)o(t)j(v)m(arious)e
(forms)g(of)g(sync)o(hronization.)75 1405 y Fk(MPI)p 160 1405
V 16 w(RMW)p 288 1405 V 16 w(CREA)l(TE\(op,)h(comm)m(,)c(new)o(comm)n(\))117
1482 y Fj(IN)155 b Fk(op)541 b Fj(sync)o(hronization)14 b(op)q(eration)117
1557 y(IN)155 b Fk(comm)466 b Fj(comm)o(unicator)11 b(for)j(participating)f
(pro)q(cesses\(handle\))117 1632 y(OUT)108 b Fk(new)o(comm)393
b Fj(new)15 b(comm)o(uni)o(cator)d(used)i(for)g(sync)o(hronization\(handle\))
75 1757 y Fp(int)23 b(MPI)p 245 1757 15 2 v 17 w(Rmw)p 334
1757 V 17 w(create\(MPI)p 591 1757 V 16 w(Op)g(op,)h(MPI)p
846 1757 V 17 w(Comm)f(comm,)g(MPI)p 1197 1757 V 17 w(Comm*)g(newcomm\))75
1843 y(MPI)p 150 1843 V 17 w(RMW)p 239 1843 V 17 w(CREATE\(OP,)f(COMM,)h
(NEWCOMM,)g(IERROR\))170 1899 y(INTEGER)g(OP,)h(COMM,)f(NEWCOMM,)g(IERROR)166
1986 y Ft(Creates)13 b(a)h(sync)o(hronization)g(ob)s(ject)g(and)g(returns)g
(a)f(comm)o(unicator)h(handle)h(that)e(can)h(b)q(e)g(used)75
2042 y(later)d(to)f(access)h(the)g(ob)s(ject.)18 b(The)11 b(ob)s(ject)f(is)h
(a)f(\\hidden")i(in)o(teger)f(v)m(ariable)i(that)d(can)h(only)g(b)q(e)g
(accessed)75 2099 y(b)o(y)k(pro)q(cesses)g(in)h(the)f(group)f(of)h
Fk(comm)m Ft(,)c(using)16 b(the)f(RMW)f(call)i(b)q(elo)o(w.)21
b(The)15 b(ob)s(ject)f(is)h(initialize)q(d)j(to)75 2155 y(zero.)i(The)15
b(comm)o(unicator)g Fk(new)o(comm)10 b Ft(has)15 b(the)g(same)g(group)g(as)g
Fk(comm)m Ft(.)166 2212 y(The)g(follo)o(wing)h(op)q(erations)g(are)e(supp)q
(orted:)21 b Fi(MPI)p 1057 2212 13 2 v 14 w(SW)m(AP)m(,)12
b(MPI)p 1281 2212 V 14 w(SUM)k Ft(and)f Fi(MPI)p 1562 2212
V 14 w(BAND)p Ft(.)166 2268 y(The)h(returned)g(comm)o(unicator)g
Fk(new)o(comm)10 b Ft(can)16 b(b)q(e)h(used)f(as)f(a)h(regular)g(comm)o
(unicator,)f(in)i(ad-)75 2325 y(dition)h(to)d(b)q(eeing)j(used)f(for)f(sync)o
(hronization.)24 b(The)17 b(new)g(comm)o(unicator)f(carries)g(the)h
(information)75 2381 y(that)c(w)o(as)g(cac)o(hed)h(with)f Fk(comm)n
Ft(,)d(as)j(if)h(it)g(w)o(as)f(created)g(b)o(y)h(duplicating)h
Fk(comm)n Ft(.)h(If)e(the)f(comm)o(unicator)75 2438 y(is)j(duplicated)h(than)
e(the)g(new)h(cop)o(y)f(is)g(asso)q(ciated)h(with)f(the)h(same)f(sync)o
(hronization)h(ob)s(ject.)166 2570 y Fg(Alternativ)o(es)o(:)35
b Fj(This)13 b(is)h(ugly)f({)h(the)g(follo)o(wing)e(c)o(hoice)i(seems)g(b)q
(etter:)166 2620 y(If)e Fi(new)o(comm)d Fj(is)j(duplicated,)h(then)g(the)h
(newly)f(crearted)h(comm)o(unicator)c(will)i(not)h(b)q(e)g(asso)q(ciated)h
(with)e(a)75 2670 y(sync)o(hronizer;)j(only)e(comm)o(unicators)e(created)16
b(b)o(y)d(a)h(call)f(to)h Fi(MPI)p 1147 2670 V 14 w(RMW)p 1263
2670 V 15 w(CREA)m(TE)f Fj(carry)h(suc)o(h)h(asso)q(ciation.)-32
46 y Fn(1)-32 103 y(2)-32 159 y(3)-32 215 y(4)-32 272 y(5)-32
328 y(6)-32 385 y(7)-32 441 y(8)-32 498 y(9)-40 554 y(10)-40
611 y(11)-40 667 y(12)-40 724 y(13)-40 780 y(14)-40 836 y(15)-40
893 y(16)-40 949 y(17)-40 1006 y(18)-40 1062 y(19)-40 1119
y(20)-40 1175 y(21)-40 1232 y(22)-40 1288 y(23)-40 1345 y(24)-40
1401 y(25)-40 1457 y(26)-40 1514 y(27)-40 1570 y(28)-40 1627
y(29)-40 1683 y(30)-40 1740 y(31)-40 1796 y(32)-40 1853 y(33)-40
1909 y(34)-40 1966 y(35)-40 2022 y(36)-40 2078 y(37)-40 2135
y(38)-40 2191 y(39)-40 2248 y(40)-40 2304 y(41)-40 2361 y(42)-40
2417 y(43)-40 2474 y(44)-40 2530 y(45)-40 2587 y(46)-40 2643
y(47)-40 2699 y(48)p eop
%%Page: 28 28
bop 75 -100 a Ft(28)723 b Fl(CHAPTER)15 b(4.)35 b(ONE)16 b(SIDED)f(COMMUNICA)
l(TIONS)166 49 y Fj(W)m(e)e(should)h(mak)o(e)e(here)k(the)e(same)f(c)o(hoice)
h(that)g(w)o(e)g(mak)o(e)f(for)g(RMA)h(comm)o(unicators.)75
283 y Fk(MPI)p 160 283 14 2 v 16 w(RMW\(inva)o(r,)g(outva)o(r,)g(comm)m(\))
117 360 y Fj(IN)155 b Fk(inva)o(r)495 b Fj(input)14 b(to)g(RMW)f(op)q
(eration)h(\(in)o(teger\))117 435 y(OUT)108 b Fk(outva)o(r)467
b Fj(output)14 b(of)g(RMW)f(op)q(eration)g(\(in)o(teger\))117
510 y(IN)155 b Fk(comm)466 b Fj(comm)o(unicator)11 b(asso)q(ciated)k(with)f
(sync)o(hronizer)h(\(handle\))75 635 y Fp(int)23 b(MPI)p 245
635 15 2 v 17 w(Rmw\(int)g(invar,)g(int)h(*outvar,)e(MPI)p
1002 635 V 17 w(Comm)h(comm\))75 721 y(MPI)p 150 721 V 17 w(RMW\(INVAR,)f
(OUTVAR,)h(COMM,)g(IERROR\))170 778 y(INTEGER)g(INVAR,)g(OUTVAR,)g(COMM,)g
(IERROR)166 864 y Ft(The)15 b(call)i(has)e(the)g(same)g(e\013ect)g(as)g(an)g
(atomic)g(execution)h(of)f(the)g(co)q(de)75 958 y Fp(outvar)95
b(=)23 b(synchronizer)75 1014 y(targetvar)g(=)g(synchronizer)g(op)g(invar)75
1108 y Ft(where)c Fk(synchronizer)g Ft(and)g Fk(op)g Ft(are,)g(resp.,)g(the)g
(sync)o(hronization)g(ob)s(ject)f(and)h(the)g(sync)o(hronization)75
1165 y(op)q(eration)e(asso)q(ciated)g(with)g Fk(comm)m Ft(.)k(I.e.,)c(the)f
(call)i(returns)f(the)f(old)i(v)m(alue)g(of)e Fp(synchronizer)f
Ft(and)75 1221 y(up)q(date)i(it,)g(according)g(to)g(the)f(op)q(eration)h
Fk(op)p Ft(.)25 b(The)17 b(call)h(completes)f(after)f Fp(synchonizer)f
Ft(has)i(b)q(een)75 1278 y(up)q(dated)f(and)f(the)h(old)f(v)m(alue)i
(returned)e(at)g(the)g(origin.)166 1334 y(This)k(call)i(can)e(b)q(e)g(used)h
(to)e(implemen)o(t)j(a)d(test)h(and)g(set)g(op)q(eration.)31
b(If)19 b Fk(comm)13 b Ft(is)20 b(asso)q(ciated)75 1391 y(with)d(the)f(op)q
(eration)h Fi(MPI)p 541 1391 13 2 v 15 w(SW)m(AP)e Ft(then)i(A)f(call)i(to)e
Fk(MPI)p 1065 1391 14 2 v 16 w(RMW\(inva)o(r,)f(outva)o(r,)i(comm)m(\))c
Ft(with)k Fk(inva)o(r)f(=)75 1447 y(1)f Ft(is)h(a)g(lo)q(c)o(k)g(op)q
(eration)g(that)e(succeeds)j(if)f Fk(outva)o(r)f(=0)h Ft(\(lo)q(c)o(k)g(w)o
(as)f(o\013)t(\),)e(fails)k(if)f Fk(outva)o(r)f(=)h(1)f Ft(\(lo)q(c)o(k)h(w)o
(as)75 1503 y(on\);)e(the)i(call)g(with)g Fk(inva)o(r)e(=)i(0)f
Ft(is)g(an)g(unlo)q(c)o(k)i(op)q(eration.)166 1560 y(The)12
b(call,)h(with)g(the)f Fi(MPI)p 602 1560 13 2 v 14 w(SUM)g
Ft(op)q(eration)g(implemen)o(ts)h(a)f(fetc)o(h&add)g(that)g(is)g(useful)h
(for)e(v)m(arious)75 1616 y(coutner)k(based)h(sync)o(hronization)g(op)q
(erations.)166 1749 y Fg(Alternativ)o(es)o(:)166 1799 y Fj(W)m(e)d(can)h
(declare)h(an)f(arra)o(y)f(of)h(sync)o(hronizers,)h(rather)f(than)g(one.)k
(Both)d(calls)e(with)h(ha)o(v)o(e)f(an)h(additional)75 1848
y(argumen)o(t:)j(n)o(um)o(b)q(er)c(of)g(ob)r(jects)i(in)f(the)g(create)i
(call)d(and)g(index)h(in)g(the)g(use)h(call.)166 1898 y(W)m(e)e(can)g(asso)q
(ciate)h(an)f(op)q(eration)g(with)g(the)h(RMW)e(call,)g(rather)i(than)f(with)
g(the)h(create)h(call.)i(The)c(giv)o(es)75 1948 y(more)g(\015exibilit)o(y)g
(to)i(the)g(user.)21 b(E.g.,)13 b(curren)o(tly)m(,)i(a)f(sync)o(hronization)g
(ob)r(ject)i(asso)q(ciated)f(with)f(the)h(op)q(eration)75 1998
y Fi(MPI)p 152 1998 V 14 w(SW)m(AP)c Fj(do)q(es)i(not)f(supp)q(ort)h(a)f
(nondestructiv)o(e)i(read.)k(Ho)o(w)o(ev)o(er,)12 b(the)h(c)o(hange)g(will)d
(prev)o(en)o(t)j(optimizations)75 2048 y(that)h(tak)o(e)g(adv)n(an)o(tage)f
(of)g(hardw)o(are)h(that)g(supp)q(ort)h(some)e(but)h(no)g(all)f(of)g(the)h
(RMW)f(op)q(erations.)166 2098 y(W)m(e)g(ma)o(y)f(prefer)j
Fi(inva)o(r)g Fj(and)e Fi(outva)o(r)i Fj(to)e(b)q(e)i(64)e(bit)h(v)n(alues)f
(on)h(systems)g(that)g(supp)q(ort)h(64)e(bit)h(in)o(tegers.)166
2147 y(W)m(e)h(ma)o(y)e(prefer)j(y)o(et)g(another)g(sp)q(ecial)f(opaque)g(ob)
r(ject,)h(rather)g(than)f(attac)o(hing)g(y)o(et)h(another)f(function)75
2197 y(to)h(comm)o(unicators.)21 b(But)16 b(theres)i(seem)e(to)f(b)q(e)i(no)e
(harm,)g(since)h(no)g(other)g(MPI)g(call)f(need)i(b)q(e)f(a)o(w)o(are)g(of)f
(the)75 2247 y(sync)o(hronizer)g(attac)o(hed)g(with)e(the)i(comm)o(unicator.)
166 2297 y(Ma)o(y)g(w)o(an)o(t)g(a)g(sw)o(ap)g(op)q(eration)g(that)h(also)e
(supp)q(orts)j(a)e(nondestructiv)o(e)i(read.)23 b(E.g.,)14
b(if)g Fi(inva)o(r)i Fj(has)g(v)n(alue)75 2347 y Fi(MPI)p 152
2347 V 14 w(UNDEFINED)p Fj(,)j(then)i(sw)o(ap)e(returns)j(the)f(old)e(v)n
(alue)g(of)h(the)g(sync)o(hronization)g(ob)r(ject,)i(but)e(do)q(es)h(not)75
2396 y(up)q(date)15 b(it.)166 2453 y(Am)d(not)i(sure)h(that)f(w)o(e)g(need)h
Fi(MPI)p 732 2453 V 14 w(BAND)p Fj(.)f(Ma)o(y)f(b)q(e)h(only)g(t)o(w)o(o)f
(op)q(erations?)189 2642 y Fh(R)n(ationale.)58 b Ft(E\016cien)o(t)18
b(sync)o(hronization)i(requires)e(the)h(use)f(of)g(RMW)g(op)q(erations,)h
(suc)o(h)f(as)189 2698 y(de\014ned)d(here;)g(it)f(requires,)h(in)f(a)g
(preemptiv)o(e)h(execution)g(en)o(vironmen)o(t,)f(in)o(tegration)g(with)h
(the)1967 46 y Fn(1)1967 103 y(2)1967 159 y(3)1967 215 y(4)1967
272 y(5)1967 328 y(6)1967 385 y(7)1967 441 y(8)1967 498 y(9)1959
554 y(10)1959 611 y(11)1959 667 y(12)1959 724 y(13)1959 780
y(14)1959 836 y(15)1959 893 y(16)1959 949 y(17)1959 1006 y(18)1959
1062 y(19)1959 1119 y(20)1959 1175 y(21)1959 1232 y(22)1959
1288 y(23)1959 1345 y(24)1959 1401 y(25)1959 1457 y(26)1959
1514 y(27)1959 1570 y(28)1959 1627 y(29)1959 1683 y(30)1959
1740 y(31)1959 1796 y(32)1959 1853 y(33)1959 1909 y(34)1959
1966 y(35)1959 2022 y(36)1959 2078 y(37)1959 2135 y(38)1959
2191 y(39)1959 2248 y(40)1959 2304 y(41)1959 2361 y(42)1959
2417 y(43)1959 2474 y(44)1959 2530 y(45)1959 2587 y(46)1959
2643 y(47)1959 2699 y(48)p eop
%%Page: 29 29
bop 75 -100 a Fl(4.9.)34 b(FUR)l(THER)16 b(EXAMPLES)1153 b
Ft(29)189 49 y(sc)o(heduler.)29 b(W)l(e)18 b(ha)o(v)o(e)g(not)g(attempted)f
(to)g(handle)j(the)e(later,)g(as)g(MPI)f(has)h(so)g(far)f(refrained)189
106 y(from)h(pro)o(viding)j(sc)o(heduling)g(services,)g(or)e(from)f(making)i
(assumptions)f(on)h(the)f(underlying)189 162 y(sc)o(heduling)e(p)q(olicies.)
22 b(\()p Fh(End)16 b(of)g(r)n(ationale.)p Ft(\))75 305 y Fq(4.9)59
b(F)n(urther)20 b(Examples)75 407 y Ft(NEED)15 b(TO)g(CHECK)g(ALL)h(EXAMPLES)
g(ANEW)-32 46 y Fn(1)-32 103 y(2)-32 159 y(3)-32 215 y(4)-32
272 y(5)-32 328 y(6)-32 385 y(7)-32 441 y(8)-32 498 y(9)-40
554 y(10)-40 611 y(11)-40 667 y(12)-40 724 y(13)-40 780 y(14)-40
836 y(15)-40 893 y(16)-40 949 y(17)-40 1006 y(18)-40 1062 y(19)-40
1119 y(20)-40 1175 y(21)-40 1232 y(22)-40 1288 y(23)-40 1345
y(24)-40 1401 y(25)-40 1457 y(26)-40 1514 y(27)-40 1570 y(28)-40
1627 y(29)-40 1683 y(30)-40 1740 y(31)-40 1796 y(32)-40 1853
y(33)-40 1909 y(34)-40 1966 y(35)-40 2022 y(36)-40 2078 y(37)-40
2135 y(38)-40 2191 y(39)-40 2248 y(40)-40 2304 y(41)-40 2361
y(42)-40 2417 y(43)-40 2474 y(44)-40 2530 y(45)-40 2587 y(46)-40
2643 y(47)-40 2699 y(48)p eop
%%Trailer
end
userdict /end-hook known{end-hook}if
%%EOF

• Next message: (no name): "Re: YASM"
• Previous message: Richard Frost: "info, hints, commandline in dynamic/io/1sided (fwd)"
• Next in thread: Steve Huss-Lederman: "Re: darft of chapter four"