(no subject)

• Messages sorted by: [ date ][ thread ][ subject ][ author ]
• Next message: Marc Snir: "Re: new version of chapter 4"
• Next message: David C. DiNucci: "Re: new version of chapter 4"
• Previous message: Marc Snir: "new version of chapter 4"

end
save
-108.0 495.0 translate
1 -1 scale

/clp {closepath} bind def
/ef {eofill} bind def
/gr {grestore} bind def
/gs {gsave} bind def
/l {lineto} bind def
/m {moveto} bind def
/n {newpath} bind def
/s {stroke} bind def
/slc {setlinecap} bind def
/slj {setlinejoin} bind def
/slw {setlinewidth} bind def
/srgb {setrgbcolor} bind def
/rot {rotate} bind def
/sc {scale} bind def
/tr {translate} bind def
/tnt {dup dup currentrgbcolor
4 -2 roll dup 1 exch sub 3 -1 roll mul add
4 -2 roll dup 1 exch sub 3 -1 roll mul add
4 -2 roll dup 1 exch sub 3 -1 roll mul add srgb}
bind def
/shd {dup dup currentrgbcolor 4 -2 roll mul 4 -2 roll mul
4 -2 roll mul srgb} bind def
/$F2psBegin {$F2psDict begin /$F2psEnteredState save def} def
/$F2psEnd {$F2psEnteredState restore end} def

$F2psBegin
10 setmiterlimit
0.06000 0.06000 sc
15.000 slw
[100.0] 0 setdash
n 4950 2625 m 3300 3150 l gs col-1 s gr [] 0 setdash
n 3546.89 3134.41 m 3300.00 3150.00 l 3510.51 3020.06 l 3529.20 3077.73 l 3546.89 3134.41 l clp gs 0.00 setgray ef gr gs col-1 s gr
[100.0] 0 setdash
n 6450 2625 m 8025 3150 l gs col-1 s gr [] 0 setdash
n 7816.29 3017.18 m 8025.00 3150.00 l 7778.34 3131.03 l 7797.82 3074.61 l 7816.29 3017.18 l clp gs 0.00 setgray ef gr gs col-1 s gr
[100.0] 0 setdash
n 8025 5175 m 6600 6000 l gs col-1 s gr [] 0 setdash
n 6837.76 5931.68 m 6600.00 6000.00 l 6777.64 5827.83 l 6808.20 5880.25 l 6837.76 5931.68 l clp gs 0.00 setgray ef gr gs col-1 s gr
[100.0] 0 setdash
n 3300 5325 m 4950 6000 l gs col-1 s gr [] 0 setdash
n 4750.59 5853.60 m 4950.00 6000.00 l 4705.15 5964.66 l 4728.37 5909.63 l 4750.59 5853.60 l clp gs 0.00 setgray ef gr gs col-1 s gr
n 7875 4050 m 5700 7200 l gs col-1 s gr
n 5885.74 7036.60 m 5700.00 7200.00 l 5786.99 6968.41 l 5836.87 7003.00 l 5885.74 7036.60 l clp gs 0.00 setgray ef gr gs col-1 s gr
n 3075 4125 m 5025 6750 l gs col-1 s gr
n 4930.05 6521.56 m 5025.00 6750.00 l 4833.72 6593.12 l 4882.38 6557.84 l 4930.05 6521.56 l clp gs 0.00 setgray ef gr gs col-1 s gr
/Times-Bold findfont 270.00 scalefont setfont
1800 3930 m
gs 1 -1 sc (put) col-1 show gr
/Times-Bold findfont 270.00 scalefont setfont
1800 4560 m
gs 1 -1 sc (put) col-1 show gr
/Times-Bold findfont 270.00 scalefont setfont
5100 1380 m
gs 1 -1 sc (load) col-1 show gr
/Times-Bold findfont 270.00 scalefont setfont
5100 2010 m
gs 1 -1 sc (load) col-1 show gr
/Times-Bold findfont 270.00 scalefont setfont
8100 3930 m
gs 1 -1 sc (put) col-1 show gr
/Times-Bold findfont 270.00 scalefont setfont
8100 4560 m
gs 1 -1 sc (put) col-1 show gr
/Times-Bold findfont 270.00 scalefont setfont
5100 6930 m
gs 1 -1 sc (load) col-1 show gr
/Times-Bold findfont 270.00 scalefont setfont
5100 7560 m
gs 1 -1 sc (load) col-1 show gr
/Times-Bold findfont 270.00 scalefont setfont
5100 8190 m
gs 1 -1 sc (RMA_clear) col-1 show gr
/Times-Bold findfont 270.00 scalefont setfont
5100 750 m
gs 1 -1 sc (RMA_wait) col-1 show gr
/Times-Bold findfont 270.00 scalefont setfont
5100 2640 m
gs 1 -1 sc (RMA_clear) col-1 show gr
/Times-Bold findfont 270.00 scalefont setfont
1800 3300 m
gs 1 -1 sc (RMA_start) col-1 show gr
/Times-Bold findfont 270.00 scalefont setfont
1800 5190 m
gs 1 -1 sc (RMA_complete) col-1 show gr
/Times-Bold findfont 270.00 scalefont setfont
8100 3300 m
gs 1 -1 sc (RMA_start) col-1 show gr
/Times-Bold findfont 270.00 scalefont setfont
8100 5190 m
gs 1 -1 sc (RMA_complete) col-1 show gr
/Times-Bold findfont 270.00 scalefont setfont
5100 6300 m
gs 1 -1 sc (RMA_wait) col-1 show gr
$F2psEnd
restore
%%EndDocument
75 4 a
endTexFig
75 1218 a Fu(Figure)19 b(1.1:)27 b(2-part)o(y)18 b(comm)o(unication.)31
b(Dashed)19 b(arro)o(ws)f(represen)o(t)h(sync)o(hronizations)h(and)f(solid)75
1274 y(arro)o(ws)14 b(represen)o(t)h(comm)o(unication)h(\(data)e(dep)q
(endencies\).)75 1414 y(In)20 b(a)e(similar)i(manner,)g(w)o(e)e(can)h
(replace)h(in)g(Example)f(1.2,)g(page)f(9,)i(the)f(call)h(to)e(get)g(b)o(y)h
(a)g(call)h(to)75 1471 y(accum)o(ulate,)e(th)o(us)f(p)q(erforming)h(the)g
(computation)f(with)h(only)g(one)f(comm)o(unication)i(b)q(et)o(w)o(een)e(an)o
(y)75 1527 y(t)o(w)o(o)d(pro)q(cesses.)75 1670 y Fr(1.4)59
b(Synchronization)20 b(calls)75 1772 y Fu(Sync)o(hronization)e(calls)g
(accept)f(as)f(parameter)g(a)h(windo)o(w)g(iden)o(ti\014er)h(\(comm)o
(unicator,)e(or)g(comm)o(u-)75 1828 y(nicator)g(and)g(rank\).)21
b(These)16 b(calls)h(are)f(used)g(for)g(ordering)g(correctly)g(con\015icting)
h(accesses)g(to)e(RMA)75 1885 y(windo)o(ws,)h(and)h(making)f(sure)h(that)e
(they)h(are)g(not)g(concurren)o(t.)23 b(These)17 b(sync)o(hronization)g
(calls)h(ma)o(y)75 1941 y(not,)12 b(in)g(general,)g(b)q(e)h(used)f(to)f(sync)
o(hronize)i(op)q(erations)e(other)h(than)f(con\015icting)i(accesses)f(to)f
(an)h(RMA)75 1997 y(windo)o(w.)166 2054 y(Figure)k(1.1)e(illustrates)j(the)f
(general)g(sync)o(hronization)g(pattern,)f(for)g(2-part)o(y)g(comm)o
(unication,)75 2110 y(i.e.,)f(for)g(con\015icts)g(b)q(et)o(w)o(een)h(RMA)f
(acceses)h(and)f(lo)q(cal)h(accesses.)20 b(The)15 b(\014gure)f(illustrates)h
(the)f(case)g(of)75 2167 y(m)o(ultiple)h(pro)q(ducers)f(to)f(a)h(windo)o(w,)f
(that)g(use)h(put)f(calls;)i(the)f(other)f(RMA)h(calls)g(are)f(dealt)h
(similarly)l(.)75 2223 y(Accesses)j(to)e(the)i(windo)o(w)f(is)h(preceded)g
(and)f(follo)o(w)o(ed)h(b)o(y)f(sync)o(hronization)h(calls)g(on)g(this)f
(windo)o(w.)75 2280 y(The)e(preceeding)i(sync)o(hronization)f(call)h(\()p
Fo(MPI)p 902 2280 14 2 v 15 w(RMA)p 1016 2280 V 17 w(ST)l(ART)f
Fu(and)g Fo(MPI)p 1360 2280 V 15 w(RMA)p 1474 2280 V 17 w(W)l(AIT)p
Fu(\))f(ensures)h(that)75 2336 y(the)j(windo)o(w)g(access)h(do)q(es)f(not)g
(start)e(b)q(efore)j(other)e(con\015icting)j(op)q(erations)e(ha)o(v)o(e)g
(completed;)i(the)75 2393 y(ensuing)f(sync)o(hronization)g(call)h(\()p
Fo(MPI)p 761 2393 V 15 w(RMA)p 875 2393 V 17 w(COMPLETE)e Fu(and)g
Fo(MPI)p 1324 2393 V 16 w(RMA)p 1439 2393 V 17 w(CLEAR)p Fu(\))g(ensures)h
(that)75 2449 y(the)c(windo)o(w)h(access)f(completes)h(b)q(efore)f(other)g
(con\015icting)i(op)q(erations)e(ha)o(v)o(e)g(started.)166
2506 y(Sync)o(hronization)h(calls)h(that)d(protect)h(RMA)g(accesses)h(need)g
(a\013ect)f(only)g(the)h(brac)o(k)o(eted)f(RMA)75 2562 y(calls)h(to)e(the)h
(relev)m(an)o(t)g(windo)o(w\(s\).)20 b(The)15 b(preceding)h(sync)o
(hronization)g(has)e(to)g(b)q(e)i(completed)g(b)q(efore)75
2618 y(the)f(RMA)h(call)g(a\013ects)e(the)h(target)f(memory;)h(and)g(the)g
(data)g(transfers)f(e\013ected)h(b)o(y)g(the)h(RMA)f(calls)75
2675 y(m)o(ust)k(complete)h(b)q(efore)g(the)f(ensuing)i(sync)o(hronization)f
(call)g(starts.)31 b(On)20 b(the)g(other)f(hand,)h(other)-32
46 y Fj(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: 14 14
bop 75 -100 a Fu(14)723 b Fl(CHAPTER)15 b(1.)35 b(ONE)16 b(SIDED)f(COMMUNICA)
l(TIONS)75 4 y
18945146 18566234 0 0 30851645 30391173 startTexFig
75 4 a
%%BeginDocument: sync2.eps
/$F2psDict 200 dict def
$F2psDict begin
$F2psDict /mtrx matrix put
/col-1 {} def
/col0 {0.000 0.000 0.000 srgb} bind def
/col1 {0.000 0.000 1.000 srgb} bind def
/col2 {0.000 1.000 0.000 srgb} bind def
/col3 {0.000 1.000 1.000 srgb} bind def
/col4 {1.000 0.000 0.000 srgb} bind def
/col5 {1.000 0.000 1.000 srgb} bind def
/col6 {1.000 1.000 0.000 srgb} bind def
/col7 {1.000 1.000 1.000 srgb} bind def
/col8 {0.000 0.000 0.560 srgb} bind def
/col9 {0.000 0.000 0.690 srgb} bind def
/col10 {0.000 0.000 0.820 srgb} bind def
/col11 {0.530 0.810 1.000 srgb} bind def
/col12 {0.000 0.560 0.000 srgb} bind def
/col13 {0.000 0.690 0.000 srgb} bind def
/col14 {0.000 0.820 0.000 srgb} bind def
/col15 {0.000 0.560 0.560 srgb} bind def
/col16 {0.000 0.690 0.690 srgb} bind def
/col17 {0.000 0.820 0.820 srgb} bind def
/col18 {0.560 0.000 0.000 srgb} bind def
/col19 {0.690 0.000 0.000 srgb} bind def
/col20 {0.820 0.000 0.000 srgb} bind def
/col21 {0.560 0.000 0.560 srgb} bind def
/col22 {0.690 0.000 0.690 srgb} bind def
/col23 {0.820 0.000 0.820 srgb} bind def
/col24 {0.500 0.190 0.000 srgb} bind def
/col25 {0.630 0.250 0.000 srgb} bind def
/col26 {0.750 0.380 0.000 srgb} bind def
/col27 {1.000 0.500 0.500 srgb} bind def
/col28 {1.000 0.630 0.630 srgb} bind def
/col29 {1.000 0.750 0.750 srgb} bind def
/col30 {1.000 0.880 0.880 srgb} bind def
/col31 {1.000 0.840 0.000 srgb} bind def

end
save
-108.0 494.0 translate
1 -1 scale

/clp {closepath} bind def
/ef {eofill} bind def
/gr {grestore} bind def
/gs {gsave} bind def
/l {lineto} bind def
/m {moveto} bind def
/n {newpath} bind def
/s {stroke} bind def
/slc {setlinecap} bind def
/slj {setlinejoin} bind def
/slw {setlinewidth} bind def
/srgb {setrgbcolor} bind def
/rot {rotate} bind def
/sc {scale} bind def
/tr {translate} bind def
/tnt {dup dup currentrgbcolor
4 -2 roll dup 1 exch sub 3 -1 roll mul add
4 -2 roll dup 1 exch sub 3 -1 roll mul add
4 -2 roll dup 1 exch sub 3 -1 roll mul add srgb}
bind def
/shd {dup dup currentrgbcolor 4 -2 roll mul 4 -2 roll mul
4 -2 roll mul srgb} bind def
/$F2psBegin {$F2psDict begin /$F2psEnteredState save def} def
/$F2psEnd {$F2psEnteredState restore end} def

$F2psBegin
10 setmiterlimit
0.06000 0.06000 sc
15.000 slw
[100.0] 0 setdash
n 4950 2625 m 3300 3150 l gs col-1 s gr [] 0 setdash
n 3546.89 3134.41 m 3300.00 3150.00 l 3510.51 3020.06 l 3529.20 3077.73 l 3546.89 3134.41 l clp gs 0.00 setgray ef gr gs col-1 s gr
[100.0] 0 setdash
n 6450 2625 m 8025 3150 l gs col-1 s gr [] 0 setdash
n 7816.29 3017.18 m 8025.00 3150.00 l 7778.34 3131.03 l 7797.82 3074.61 l 7816.29 3017.18 l clp gs 0.00 setgray ef gr gs col-1 s gr
[100.0] 0 setdash
n 8025 5175 m 6600 6000 l gs col-1 s gr [] 0 setdash
n 6837.76 5931.68 m 6600.00 6000.00 l 6777.64 5827.83 l 6808.20 5880.25 l 6837.76 5931.68 l clp gs 0.00 setgray ef gr gs col-1 s gr
[100.0] 0 setdash
n 3300 5325 m 4950 6000 l gs col-1 s gr [] 0 setdash
n 4750.59 5853.60 m 4950.00 6000.00 l 4705.15 5964.66 l 4728.37 5909.63 l 4750.59 5853.60 l clp gs 0.00 setgray ef gr gs col-1 s gr
n 4500 1800 m 2925 4125 l gs col-1 s gr
n 3109.28 3959.95 m 2925.00 4125.00 l 3009.93 3892.65 l 3060.10 3926.80 l 3109.28 3959.95 l clp gs 0.00 setgray ef gr gs col-1 s gr
n 6000 1725 m 7500 4050 l gs col-1 s gr
n 7420.31 3815.80 m 7500.00 4050.00 l 7319.47 3880.86 l 7370.39 3848.83 l 7420.31 3815.80 l clp gs 0.00 setgray ef gr gs col-1 s gr
/Times-Bold findfont 270.00 scalefont setfont
5100 2010 m
gs 1 -1 sc (put) col-1 show gr
/Times-Bold findfont 270.00 scalefont setfont
5100 1380 m
gs 1 -1 sc (put) col-1 show gr
/Times-Bold findfont 270.00 scalefont setfont
1800 3930 m
gs 1 -1 sc (get) col-1 show gr
/Times-Bold findfont 270.00 scalefont setfont
1800 4560 m
gs 1 -1 sc (get) col-1 show gr
/Times-Bold findfont 270.00 scalefont setfont
8100 3930 m
gs 1 -1 sc (get) col-1 show gr
/Times-Bold findfont 270.00 scalefont setfont
8100 4560 m
gs 1 -1 sc (get) col-1 show gr
/Times-Bold findfont 270.00 scalefont setfont
5100 6930 m
gs 1 -1 sc (put) col-1 show gr
/Times-Bold findfont 270.00 scalefont setfont
5100 7560 m
gs 1 -1 sc (put) col-1 show gr
/Times-Bold findfont 270.00 scalefont setfont
5100 750 m
gs 1 -1 sc (lock\(exclusive\)) col-1 show gr
/Times-Bold findfont 270.00 scalefont setfont
5100 2640 m
gs 1 -1 sc (unlock) col-1 show gr
/Times-Bold findfont 270.00 scalefont setfont
1800 3300 m
gs 1 -1 sc (lock\(shared\)) col-1 show gr
/Times-Bold findfont 270.00 scalefont setfont
1800 5190 m
gs 1 -1 sc (unlock) col-1 show gr
/Times-Bold findfont 270.00 scalefont setfont
8100 3300 m
gs 1 -1 sc (lock\(shared\)) col-1 show gr
/Times-Bold findfont 270.00 scalefont setfont
8100 5190 m
gs 1 -1 sc (unlock) col-1 show gr
/Times-Bold findfont 270.00 scalefont setfont
5100 6300 m
gs 1 -1 sc (lock\(exclusive\)) col-1 show gr
/Times-Bold findfont 270.00 scalefont setfont
5100 8190 m
gs 1 -1 sc (unlock) col-1 show gr
$F2psEnd
restore
%%EndDocument
75 4 a
endTexFig
75 1278 a Fu(Figure)k(1.2:)27 b(3-part)o(y)18 b(comm)o(unication.)31
b(Dashed)19 b(arro)o(ws)f(represen)o(t)h(sync)o(hronizations)h(and)f(solid)75
1334 y(arro)o(ws)14 b(represen)o(t)h(comm)o(unication)h(\(data)e(dep)q
(endencies\).)75 1474 y(op)q(erations)i(brac)o(k)o(eted)g(b)o(y)g(these)g(t)o
(w)o(o)f(sync)o(hronization)i(calls)g(ma)o(y)f(not)f(b)q(e)i(dela)o(y)o(ed)g
(un)o(til)g(the)f(\014rst)75 1531 y(sync)o(hronization)i(call)g(completes,)g
(and)f(ma)o(y)f(not)h(complete)g(b)q(efore)h(the)f(ensuing)h(sync)o
(hronization)75 1587 y(completes.)166 1643 y(Figure)e(1.2)e(illustrates)j
(the)f(general)g(sync)o(hronization)g(pattern,)f(for)g(3-part)o(y)g(comm)o
(unication,)75 1700 y(i.e.,)20 b(for)e(con\015icts)i(b)q(et)o(w)o(een)g(RMA)f
(accesses.)32 b(The)19 b(\014gure)g(illustrates)i(m)o(ultiple)g(readers)e
(and)g(one)75 1756 y(writer,)d(whic)o(h)i(access)f(memory)f(at)g(a)g(pro)q
(cess)h(that)f(is)h(not)f(in)o(v)o(olv)o(ed)i(in)f(the)g(comm)o(unication.)25
b(Ac-)75 1813 y(cesses)16 b(are)e(brac)o(k)o(eted)h(b)o(y)g(sync)o
(hronization)i(calls)f(that)e(ensure)i(that)f(con\015icts)h(are)f(a)o(v)o
(oided.)166 1869 y(MPI)10 b(pro)o(vides)h(three)g(sync)o(hronization)h(mec)o
(hanisms:)18 b(The)11 b(most)e(elab)q(orate)i(is)g(pro)o(vided)g(b)o(y)g(the)
75 1926 y(four)f(functions)h Fo(MPI)p 441 1926 14 2 v 15 w(RMA)p
555 1926 V 17 w(ST)l(ART,)g(MPI)p 820 1926 V 16 w(RMA)p 935
1926 V 17 w(COMPLETE,)e(MPI)p 1297 1926 V 16 w(RMA)p 1412 1926
V 17 w(W)l(AIT,)h Fu(and)g Fo(MPI)p 1730 1926 V 16 w(RMA)p
1845 1926 V 17 w(CLEAR)p Fu(,)75 1982 y(whic)o(h)17 b(directly)g(supp)q(orts)
f(the)g(2-part)o(y)f(sync)o(hronization)i(pattern)e(illustrated)j(in)e
(Figure)h(1.1.)k(The)75 2039 y Fo(MPI)p 160 2039 V 16 w(BARRIER)16
b Fu(collectiv)o(e)h(comm)o(unication)f(call)g(supp)q(orts)g(a)f
(simpli\014ed)i(sync)o(hronization)f(pattern)75 2095 y(that)j(is)h(often)f
(used)i(in)f(parallel)h(computations:)29 b(namely)20 b(a)g(lo)q(osely-sync)o
(hronous)g(mo)q(del,)i(where)75 2152 y(global)15 b(computation)f(phases)g
(alternate)g(with)g(global)g(comm)o(unication)h(phases.)20
b(Finaly)15 b(shared)f(and)75 2208 y(exclusiv)o(e)i(lo)q(c)o(ks)g(are)e(pro)o
(vided)i(b)o(y)e(the)h(t)o(w)o(o)e(function)j Fo(MPI)p 1128
2208 V 16 w(LOCK)e Fu(and)h Fo(MPI)p 1447 2208 V 16 w(UNLOCK)p
Fu(,)f(in)i(supp)q(ort)75 2264 y(of)f(3-part)o(y)f(comm)o(unication)i(sync)o
(hronization)g(of)f(the)g(t)o(yp)q(e)g(illustrated)i(in)f(Figure)f(1.2.)1967
46 y Fj(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: 15 15
bop 75 -100 a Fl(1.4.)34 b(SYNCHR)o(ONIZA)l(TION)17 b(CALLS)1031
b Fu(15)75 49 y Fe(1.4.1)49 b(General)16 b(2-pa)o(rt)o(y)g(synchronization)75
230 y Fo(MPI)p 160 230 14 2 v 16 w(RMA)p 275 230 V 17 w(ST)l(ART\(comm)m(,)c
(rank\))117 307 y Fk(IN)155 b Fo(comm)466 b Fk(comm)o(unicator)11
b(asso)q(ciated)k(with)f(windo)o(w)f(\(handle\))117 382 y(IN)155
b Fo(rank)504 b Fk(rank)14 b(of)f(pro)q(cess)j(\(in)o(teger\))75
506 y Fq(int)23 b(MPI)p 245 506 15 2 v 17 w(RMA)p 334 506 V
17 w(start\(MPI)p 567 506 V 16 w(Comm)g(comm,)g(int)h(rank\))75
593 y(MPI)p 150 593 V 17 w(RMA)p 239 593 V 17 w(START\(COMM,)e(RANK,)h
(IERROR\))170 649 y(INTEGER)g(COMM,)h(RANK,)f(IERROR)166 736
y Fu(Starts)13 b(a)h(sequence)i(of)e(RMA)g(op)q(erations)h(targetted)e(to)g
(the)i(windo)o(w)f(asso)q(ciated)h(with)g Fo(\(com)o(m)m(,)75
792 y(rank\))p Fu(.)k(The)d(succeeding)g(RMA)f(accesses)h(to)e(this)h(windo)o
(w)g(will)i(b)q(e)e(dela)o(y)o(ed,)h(if)f(necessary)l(,)g(un)o(til)h(the)75
849 y(windo)o(w)f(is)h(mark)o(ed)f(clear)h(b)o(y)f(a)g(call)h(to)e
Fo(MPI)p 881 849 14 2 v 16 w(RMA)p 996 849 V 17 w(CLEAR)p Fu(.)75
1000 y Fo(MPI)p 160 1000 V 16 w(RMA)p 275 1000 V 17 w(COMPLETE\(com)o(m)n(,)d
(rank\))117 1077 y Fk(IN)155 b Fo(comm)466 b Fk(comm)o(unicator)11
b(asso)q(ciated)k(with)f(windo)o(w)f(\(handle\))117 1152 y(IN)155
b Fo(rank)504 b Fk(rank)14 b(of)f(pro)q(cess)j(\(in)o(teger\))75
1276 y Fq(int)23 b(MPI)p 245 1276 15 2 v 17 w(RMA)p 334 1276
V 17 w(complete\(MPI)p 639 1276 V 16 w(Comm)g(comm,)g(int)g(rank\))75
1363 y(MPI)p 150 1363 V 17 w(RMA)p 239 1363 V 17 w(START\(COMM,)f(RANK,)h
(IERROR\))170 1419 y(INTEGER)g(COMM,)h(RANK,)f(IERROR)166 1506
y Fu(Complete)15 b(\(at)e(the)i(origin\))f(a)h(sequence)g(of)f(RMA)h(op)q
(erations)f(targetted)g(to)f(the)i(windo)o(w)g(asso-)75 1562
y(ciated)h(with)f Fo(\(comm)m(,)d(rank\))p Fu(.)166 1619 y(Consider)k(a)f
(sequence)h(of)f(calls)75 1712 y Fq(MPI_RMA_START\(comm,)21
b(rank\))75 1769 y(...)75 1825 y(MPI_PUT\(...,rank,...,comm\))75
1882 y(...)75 1938 y(MPI_RMA_COMPLETE\(comm,)g(rank\))75 2032
y Fu(The)e(call)h(to)f Fo(MPI)p 404 2032 14 2 v 15 w(RMA)p
518 2032 V 17 w(COMPLETE)g Fu(blo)q(c)o(ks)h(un)o(til)g(the)f(put)g(call)h
(has)f(completed)h(at)f(the)g(origin;)75 2089 y(and)13 b(the)g(target)f
(windo)o(w)h(will)i(b)q(e)e(accessed)h(b)o(y)f(the)g(put)g(op)q(eration)g
(only)h(if)f(it)h(is)f(clear.)20 b(On)13 b(the)g(other)75 2145
y(hand,)j(it)f(is)h(up)g(to)f(the)g(implemen)o(tation)i(whether)f(the)f
(\014rst)g(t)o(w)o(o)f(calls)j(blo)q(c)o(k.)k(An)16 b(implemen)o(tation)75
2201 y(ma)o(y)e(ha)o(v)o(e)g(the)h(call)g(to)f Fo(MPI)p 577
2201 V 16 w(RMA)p 692 2201 V 17 w(ST)l(ART)h Fu(blo)q(c)o(k)h(un)o(til)f(the)
g(windo)o(w)g(is)g(clear;)f(or)g(it)h(ma)o(y)f(allo)o(w)h(the)75
2258 y(call)j(to)f Fo(MPI)p 303 2258 V 16 w(RMA)p 418 2258
V 17 w(ST)l(ART)h Fu(to)e(pro)q(ceed,)j(and)e(blo)q(c)o(k)h(the)f(call)i(to)d
Fo(MPI)p 1353 2258 V 16 w(PUT)p Fu(,)h(un)o(til)i(the)e(windo)o(w)g(is)75
2314 y(clear.)j(or,)12 b(it)i(ma)o(y)e(allo)o(w)i(b)q(oth)f(calls)h(to)e(pro)
q(ceed)i(and)g(dela)o(y)f(the)h(access)f(to)f(the)i(target)d(windo)o(w)j(un)o
(til)75 2371 y(the)h(call)i(to)d Fo(MPI)p 377 2371 V 16 w(RMA)p
492 2371 V 17 w(COMPLETE)h Fu(o)q(ccurred)h(and)f(un)o(til)i(the)e(target)f
(windo)o(w)h(is)h(clear.)166 2427 y(The)i(lo)q(cal)i(comm)o(unication)f
(bu\013er)f(of)g(the)h(put)f(call)i(should)f(not)f(b)q(e)h(accessed)g(after)e
(the)i(put)75 2484 y(call)d(started,)e(un)o(til)j(the)e(ensuing)h(call)h(to)d
Fo(MPI)p 899 2484 V 16 w(RMA)p 1014 2484 V 17 w(COMPLETE)h
Fu(completed.)166 2540 y Fo(MPI)p 251 2540 V 16 w(RMA)p 366
2540 V 17 w(COMPLETE)h Fu(enforces)h(completion)h(of)e(preceeding)i(RMA)f
(calls)h(at)d(the)i(origin.)25 b(A)75 2597 y(put)15 b(or)f(accum)o(ulate)h
(call)g(ma)o(y)f(not)g(ha)o(v)o(e)g(completed)i(at)e(the)g(target)f(when)i
(it)g(has)g(completed)g(at)f(the)75 2653 y(origin.)-32 46 y
Fj(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: 16 16
bop 75 -100 a Fu(16)723 b Fl(CHAPTER)15 b(1.)35 b(ONE)16 b(SIDED)f(COMMUNICA)
l(TIONS)166 49 y Fu(Eac)o(h)10 b(call)h(to)f Fo(MPI)p 488 49
14 2 v 15 w(RMA)p 602 49 V 17 w(ST)l(ART)h Fu(has)f(to)g(b)q(e)h(matc)o(hed)f
(b)o(y)g(a)g(unique)h(follo)o(wing)g(call)g(to)f Fo(MPI)p 1765
49 V 16 w(RMA)p 1880 49 V 17 w(COMPLETE)p Fu(,)75 106 y(and)15
b(vice)i(v)o(ersa.)75 257 y Fo(MPI)p 160 257 V 16 w(RMA)p 275
257 V 17 w(W)l(AIT\(com)o(m)m(,)12 b(count\))117 334 y Fk(IN)155
b Fo(comm)466 b Fk(comm)o(unicator)11 b(asso)q(ciated)k(with)f(windo)o(w)f
(\(handle\))117 409 y(IN)155 b Fo(count)482 b Fk(n)o(um)o(b)q(er)13
b(of)h(of)f(RMA)h(op)q(erations)g(to)f(w)o(ait)g(for)h(\(in)o(teger\))75
533 y Fq(int)23 b(MPI)p 245 533 15 2 v 17 w(RMA)p 334 533 V
17 w(wait\(MPI)p 543 533 V 16 w(Comm)g(comm,)h(int)f(count\))75
620 y(MPI)p 150 620 V 17 w(RMA)p 239 620 V 17 w(WAIT\(COMM,)f(COUNT,)h
(IERROR\))170 676 y(INTEGER)g(COMM,)h(COUNT,)f(IERROR)166 763
y Fu(Blo)q(c)o(ks)15 b(un)o(til)g Fo(count)g Fu(calls)g(to)e
Fo(MPI)p 775 763 14 2 v 16 w(RMA)p 890 763 V 17 w(COMPLETE\(comm)m(,)e
(rank\))j Fu(o)q(ccurred,)g(where)g Fo(rank)g Fu(is)75 819
y(the)h(rank)h(of)f(the)g(caller.)22 b(I.e.,)14 b Fo(MPI)p
703 819 V 16 w(RMA)p 818 819 V 17 w(W)l(AIT)h Fu(blo)q(c)o(ks)h(un)o(til)h
Fo(count)g Fu(pro)q(cesses)e(accessed)h(the)g(lo)q(cal)75 876
y(windo)o(w)h(asso)q(ciated)h(with)f Fo(comm)11 b Fu(with)18
b(RMA)f(calls.)26 b(When)18 b(the)f(call)h(returns,)f(all)h(RMA)g(accesses)75
932 y(that)e(where)g(started)g(b)q(efore)h(the)f(matc)o(hing)h(calls)g(to)f
Fo(MPI)p 1124 932 V 16 w(RMA)p 1239 932 V 17 w(COMPLETE)g Fu(ha)o(v)o(e)g
(completed)i(at)75 989 y(the)d(target)f(windo)o(w.)75 1140
y Fo(MPI)p 160 1140 V 16 w(RMA)p 275 1140 V 17 w(CLEAR\(comm)m(\))117
1217 y Fk(IN)155 b Fo(comm)466 b Fk(comm)o(unicator)11 b(asso)q(ciated)k
(with)f(windo)o(w)f(\(handle\))75 1341 y Fq(int)23 b(MPI)p
245 1341 15 2 v 17 w(RMA)p 334 1341 V 17 w(clear\(MPI)p 567
1341 V 16 w(Comm)g(comm\))75 1428 y(MPI)p 150 1428 V 17 w(RMA)p
239 1428 V 17 w(clear\(COMM,)f(IERROR\))170 1484 y(INTEGER)h(COMM,)h(IERROR)
166 1571 y Fu(Marks)13 b(the)h(lo)q(cal)g(windo)o(w)g(asso)q(ciated)g(with)g
Fo(comm)8 b Fu(as)14 b(clear,)g(th)o(us)f(allo)o(wing)i(RMA)f(op)q(erations)
75 1627 y(to)h(pro)q(ceed.)166 1684 y(Eac)o(h)10 b(call)h(to)f
Fo(MPI)p 488 1684 14 2 v 15 w(RMA)p 602 1684 V 17 w(W)l(AIT)g
Fu(has)g(to)g(b)q(e)h(matc)o(hed)f(b)o(y)g(a)g(unique)h(follo)o(wing)g(call)g
(to)f Fo(MPI)p 1735 1684 V 16 w(RMA)p 1850 1684 V 17 w(CLEAR)p
Fu(,)75 1740 y(and)15 b(vice)i(v)o(ersa.)166 1872 y Fh(Alternativ)o(es)o(:)34
b Fk(Rolf)12 b(argues)i(for)f(an)g(optimization)d(where)15
b(the)f(sync)o(hronization)f(implied)e(b)o(y)i(the)h(call)75
1922 y(to)f Fi(MPI)p 202 1922 13 2 v 14 w(RMA)p 307 1922 V
15 w(ST)m(ART)f Fk(is)h(a)o(v)o(oided)f(when)h(the)g(origin)f(pro)q(cess)j
(\\kno)o(ws")d(that)h(the)g(target)g(windo)o(w)f(is)h(clear,)g(as)75
1972 y(a)f(side-e\013ect)h(of)f(another)g(comm)o(unicatio)o(n)d(\(this)j(is)g
(analogue)f(to)h(the)g(\\ready-receiv)o(e")h(optimization\).)i(There)75
2022 y(are)f(three)h(c)o(hoices:)134 2097 y(1.)22 b(F)m(orget)13
b(ab)q(out)h(that)g(optimization.)134 2171 y(2.)22 b(Allo)o(w)14
b(RMA)h(op)q(erations)h(to)g(start)g(with)f(no)h(call)e(to)i
Fi(MPI)p 1144 2171 V 14 w(RMA)p 1249 2171 V 15 w(ST)m(ART)p
Fk(.)f(The)h(disadv)n(an)o(tage)e(is)i(that)189 2221 y(w)o(e)g(lo)q(ose)g
(the)h(brac)o(k)o(eting)f(pro)o(vided)g(b)o(y)f(the)i(start{complete)f
(pairs,)g(that)g(is)g(a)o(v)n(ailable)e(for)h(all)g(RMA)189
2271 y(op)q(erations.)27 b(In)17 b(particular,)f(one)i(gets)f(to)g(the)g
(put/get)g(call)f(without)h(kno)o(wing)f(whic)o(h)g(t)o(yp)q(e)i(of)e(syn-)
189 2321 y(c)o(hronization)d(will)g(b)q(e)h(used.)19 b(This)14
b(ma)o(y)e(restrict)j(implemen)o(tations.)134 2396 y(3.)22
b(Pro)o(vide)15 b(a)g(\\ready-start")h(v)n(arian)o(t)e(of)h(the)h
Fi(MPI)p 977 2396 V 14 w(RMA)p 1082 2396 V 15 w(ST)m(ART)f
Fk(call.)21 b(E.g.,)15 b(one)g(could)g(ha)o(v)o(e)h(an)f(addi-)189
2445 y(tional)10 b Fi(check/no)q(che)q(ck)16 b Fk(argumen)o(t)11
b(\015ag)g(to)h(tell)g(MPI)g(whether)h(it)f(can)g(assume)g(that)g(the)g
(target)h(windo)o(w)189 2495 y(is)g(clear,)h(without)g(c)o(hec)o(king.)k(An)c
(implemen)o(tation)c(could)k(ignore)g(this)g(additional)e(information.)166
2576 y(One)j(ma)o(y)c(w)o(an)o(t)j(a)g(non)o(blo)q(c)o(king)e(v)o(ersion)i
(of)f Fi(MPI)p 975 2576 V 15 w(RMA)p 1081 2576 V 15 w(W)m(AIT)g
Fk(and/or)g Fi(MPI)p 1426 2576 V 14 w(RMA)p 1531 2576 V 15
w(ST)m(ART)p Fk(.)1967 46 y Fj(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: 17 17
bop 75 -100 a Fl(1.4.)34 b(SYNCHR)o(ONIZA)l(TION)17 b(CALLS)1031
b Fu(17)75 49 y Fe(1.4.2)49 b(Ba)o(rrier)75 135 y Fu(The)16
b(MPI)f(call)i Fo(MPI)p 442 135 14 2 v 16 w(BARRIER\(comm)n(\))p
Fu(,)12 b(when)k(executed)h(on)e(a)h(comm)o(unicator)f(that)g(is)h(asso)q
(ciated)75 192 y(with)i(RMA)h(windo)o(ws,)f(has)g(the)g(additional)i
(e\013ect)d(of)h(sync)o(hronizing)h(RMA)g(op)q(erations)f(on)g(these)75
248 y(windo)o(ws.)27 b(All)18 b(RMA)g(op)q(erations)g(that)e(w)o(ere)i
(started)e(on)i(a)f(windo)o(w)g(asso)q(ciated)h(with)g Fo(comm)11
b Fu(will)75 304 y(b)q(e)19 b(completed)g(b)q(efore)g(the)f(barrier)h
(completes.)30 b(RMA)18 b(comm)o(unications)h(initiated)h(b)o(y)e(the)h
(caller)75 361 y(b)q(efore)e(the)g(barrier)f(call)i(will)g(complete)g(at)e
(the)g(origin)i(b)q(efore)f(the)f(barrier)h(call)h(returns;)f(they)f(will)75
417 y(complete)g(at)e(the)i(target)e(b)q(efore)h(the)g(barrier)h(call)g
(returns)f(at)g(the)g(target)f(pro)q(cess.)75 539 y Fe(1.4.3)49
b(Lo)q(ck)75 719 y Fo(MPI)p 160 719 V 16 w(RMA)p 275 719 V
17 w(LOCK\(com)o(m)m(,)12 b(rank,)j(lo)q(ck)p 759 719 V 16
w(t)o(yp)q(e\))117 796 y Fk(IN)155 b Fo(comm)466 b Fk(comm)o(unication)11
b(sp)q(ecifying)i(the)i(lo)q(c)o(k)o(ed)f(windo)o(w)117 871
y(IN)155 b Fo(rank)504 b Fk(rank)14 b(of)f(windo)o(w)g(o)o(wner)117
946 y(IN)155 b Fo(lo)q(ck)p 398 946 V 17 w(t)o(yp)q(e)413 b
Fk(either)10 b Fi(MPI)p 1095 946 13 2 v 15 w(LOCK)p 1219 946
V 14 w(EX)o(CLUSIVE)d Fk(or)i Fi(MPI)p 1579 946 V 15 w(LOCK)p
1703 946 V 14 w(SHARED)75 1071 y Fq(int)23 b(MPI)p 245 1071
15 2 v 17 w(RMA)p 334 1071 V 17 w(lock\(MPI)p 543 1071 V 16
w(Comm)g(comm,)h(int)f(rank,)g(int)h(lock)p 1252 1071 V 16
w(type\))75 1157 y(MPI)p 150 1157 V 17 w(RMA)p 239 1157 V 17
w(LOCK\(COMM,)e(RANK,)h(LOCK)p 757 1157 V 17 w(TYPE,)g(IERROR\))170
1214 y(INTEGER)g(COMM,)h(RANK,)f(LOCK)p 747 1214 V 16 w(TYPE,)h(IERROR)166
1300 y Fu(Starts)14 b(a)i(sequence)g(or)f(RMA)h(accesses)g(to)e(the)i(windo)o
(w)g(asso)q(ciated)f(with)h Fo(\(comm)m(,)c(rank\))p Fu(.)21
b(The)75 1357 y(call)e(also)g(ensures)f(that)g(an)o(y)g(RMA)g(access)h(to)e
(the)i(windo)o(w)f(asso)q(ciated)h(with)f Fo(\(comm)m(,)e(rank\))i
Fu(that)75 1413 y(w)o(as)c(protected)i(b)o(y)f(a)g(previously)h(released,)g
(con\015icting)h(lo)q(c)o(k)e(will)i(ha)o(v)o(e)e(completed)h(at)f(the)g
(target.)75 1564 y Fo(MPI)p 160 1564 14 2 v 16 w(WINDO)o(W)p
372 1564 V 16 w(UNLOCK\(comm)m(,)d(rank,)i(lo)q(ck)p 918 1564
V 17 w(t)o(yp)q(e\))117 1641 y Fk(IN)155 b Fo(comm)466 b Fk(comm)o(unication)
11 b(sp)q(ecifying)i(the)i(lo)q(c)o(k)o(ed)f(windo)o(w)117
1716 y(IN)155 b Fo(rank)504 b Fk(rank)14 b(of)f(windo)o(w)g(o)o(wner)166
1841 y Fu(Completes)19 b(at)f(the)h(origin)g(a)f(sequence)i(of)e(RMA)h
(accesses)g(to)f(the)h(windo)o(w)g(asso)q(ciated)g(with)75
1897 y Fo(\(comm)m(,)12 b(rank\))p Fu(.)166 1953 y(RMA)19 b(accesses)g(that)f
(are)h(protected)f(b)o(y)h(an)g(exclusiv)o(e)h(lo)q(c)o(k)g(will)g(not)e(b)q
(e)i(concurren)o(t)f(at)f(the)75 2010 y(target)c(with)h(other)f(RMA)h
(accesses)g(to)f(the)h(same)g(windo)o(w.)20 b(RMA)15 b(accesses)g(that)f(are)
g(protected)h(b)o(y)75 2066 y(a)h(shared)h(lo)q(c)o(k)g(will)h(not)e(b)q(e)i
(concurren)o(t)e(at)g(the)h(target)e(with)i(accesses)g(protected)f(b)o(y)h
(an)f(exclusiv)o(e)75 2123 y(lo)q(c)o(k)g(to)e(the)i(same)e(windo)o(w.)166
2179 y(Consider)i(a)f(sequence)h(of)f(calls)75 2272 y Fq(MPI_LOCK\(comm,)22
b(rank,)h(MPI_EXCLUSIVE\))75 2329 y(MPI_PUT\(...,)f(rank,)h(...,)h(comm\))75
2385 y(MPI_UNLOCK\(comm,)e(rank\))75 2478 y Fu(The)c(call)i(to)e
Fo(MPI)p 402 2478 V 15 w(UNLOCK)h Fu(will)h(blo)q(c)o(k)f(un)o(til)g(the)f
(put)h(transfer)e(completed)i(at)f(the)g(origin.)30 b(It)18
b(is)75 2534 y(up)h(to)f(the)h(implemen)o(tation)g(whether)g(the)g(other)f
(calls)h(are)g(blo)q(c)o(king.)31 b(An)19 b(implemen)o(tation)h(ma)o(y)75
2591 y(ha)o(v)o(e)g(the)g(call)h(to)e Fo(MPI)p 501 2591 V 16
w(LOCK)h Fu(blo)q(c)o(k)h(un)o(til)g(an)f(exclusiv)o(e)i(lo)q(c)o(k)e(on)g
(the)h(windo)o(w)f(is)g(acquired.)36 b(It)75 2647 y(could)15
b(p)q(ostp)q(one)f(lo)q(c)o(k)h(acquisition)g(un)o(til)g(the)f(call)h(to)e
Fo(MPI)p 1113 2647 V 16 w(PUT)h Fu(o)q(ccurs;)h(or)e(it)h(could)h(p)q(ostp)q
(one)f(lo)q(c)o(k)75 2704 y(acquisition)j(and)e(the)g(up)q(date)h(to)f(the)g
(windo)o(w)g(un)o(til)i(the)e(call)h(to)f Fo(MPI)p 1332 2704
V 16 w(UNLOCK)g Fu(o)q(ccurs.)-32 46 y Fj(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: 18 18
bop 75 -100 a Fu(18)723 b Fl(CHAPTER)15 b(1.)35 b(ONE)16 b(SIDED)f(COMMUNICA)
l(TIONS)166 49 y Fh(Alternativ)o(es)o(:)166 99 y Fk(W)m(e)e(ma)o(y)f
(consider)j(the)g(same)e(v)n(arian)o(ts)g(that)h(where)h(discussed)g(for)f
Fi(MPI)p 1353 99 13 2 v 14 w(RMA)p 1458 99 V 15 w(ST)m(ART)p
Fk(:)134 180 y(1.)22 b(Supp)q(ort)10 b(for)g(the)g(optimization)d(where)k
(the)g(sync)o(hronization)f(implied)d(b)o(y)j(the)h(call)e(to)h
Fi(MPI)p 1649 180 V 14 w(RMA)p 1754 180 V 15 w(LOCK)189 230
y Fk(is)19 b(a)o(v)o(oided)f(when)i(the)g(origin)e(pro)q(cess)j(\\kno)o(ws")e
(that)g(no)g(con\015icting)g(lo)q(c)o(k)g(is)g(held)h(on)f(the)g(target)189
280 y(windo)o(w.)e(Here,)d(to)q(o,)f(w)o(e)g(can)h(ha)o(v)o(e)f(a)g
(\\ready-lo)q(c)o(k")g(v)n(arian)o(t)f(of)h(the)h Fi(MPI)p
1391 280 V 14 w(RMA)p 1496 280 V 15 w(LOCK)f Fk(call,)f(pro)o(vided)189
330 y(b)o(y)h(an)h(additional)e Fi(check/no)q(che)q(ck)17 b
Fk(argumen)o(t)c(\015ag.)134 401 y(2.)22 b(One)14 b(ma)o(y)e(w)o(an)o(t)i(a)f
(non)o(blo)q(c)o(king)g(v)o(ersion)h(of)f Fi(MPI)p 998 401
V 15 w(RMA)p 1104 401 V 14 w(LOCK)p Fk(.)75 657 y Fr(1.5)59
b(Semantics)18 b(and)i(Co)n(rrectness)75 758 y Fu(An)c(access)g(b)o(y)f(pro)q
(cess)h(A)f(to)g(a)h(windo)o(w)f Fo(\(comm)m(,)d(rank\))k Fp(pr)n(e)n(c)n(e)n
(des)e Fu(an)h(access)h(b)o(y)f(pro)q(cess)h(B)g(to)f(that)75
815 y(same)g(windo)o(w)g(if)h(one)f(of)g(the)g(follo)o(wing)h(conditions)h
(hold:)131 901 y(1.)22 b(A)10 b(executed)i(a)f(call)g(to)g
Fo(MPI)p 666 901 14 2 v 15 w(BARRIER\(comm)o(\))c Fu(after)j(its)h(access)g
(completed,)h(and)f(B)g(completed)189 957 y(a)k(matc)o(hing)g(call)h(b)q
(efore)g(its)f(access)g(started.)131 1048 y(2.)22 b(A)12 b(released)h(its)f
(lo)q(c)o(k)g(on)g(the)g(windo)o(w,)h(calling)h Fo(MPI)p 1092
1048 V 15 w(WINDO)o(W)p 1303 1048 V 17 w(UNLOCK\(com)o(m)m(,)c(rank\))p
Fu(,)i(after)189 1104 y(its)20 b(access,)i(B)f(acquired)g(that)f(lo)q(c)o(k,)
i(completing)g(a)e(call)i(to)e Fo(MPI)p 1386 1104 V 16 w(WINDO)o(W)p
1598 1104 V 16 w(LOCK\(com)o(m)m(,)189 1161 y(rank\))p Fu(,)14
b(after)g(A)h(released)h(it,)f(and)g(b)q(efore)g(its)g(o)o(wn)g(access,)f
(and)i(at)e(least)h(one)g(of)g(the)g(pro)q(cesses)189 1217
y(held)h(the)f(lo)q(c)o(k)h(exclusiv)o(ely)l(..)131 1308 y(3.)22
b(The)12 b(access)h(b)o(y)g(A)f(is)i(a)e(load)h(or)f(store,)g(the)h(access)f
(b)o(y)h(B)g(is)g(an)f(RMA)h(op)q(eration,)g(A)g(executed)g(a)189
1365 y(call)h(to)e Fo(MPI)p 408 1365 V 16 w(RMA)p 523 1365
V 17 w(CLEAR)h Fu(after)f(its)i(access,)f(and)g(B)g(completed)h(a)f(call)h
(to)e Fo(MPI)p 1603 1365 V 16 w(RMA)p 1718 1365 V 17 w(ST)l(ART)189
1421 y Fu(after)i(A)h(cleared)h(the)g(windo)o(w)f(and)h(b)q(efore)f(its)g(o)o
(wn)g(access.)131 1512 y(4.)22 b(The)12 b(access)g(b)o(y)g(A)g(is)h(an)f(RMA)
g(access,)g(the)g(access)g(b)o(y)g(B)h(is)f(a)g(load)g(or)g(store,)f(A)h
(executed)h(a)f(call)189 1568 y(to)h Fo(MPI)p 328 1568 V 16
w(RMA)p 443 1568 V 17 w(COMPLETE)h Fu(after)f(its)h(access,)g(and)g(B)g
(completed)h(a)f(call)h(to)e Fo(MPI)p 1633 1568 V 16 w(RMA)p
1748 1568 V 17 w(W)l(AIT)189 1625 y Fu(b)q(efore)i(its)h(o)o(wn)e(access)i
(that)e(w)o(aited)h(on)g(the)h(complete)g(call)g(of)f(B.)166
1711 y(W)l(e)g(sa)o(y)f(that)h(u)g Fp(fol)r(lows)g Fu(v)g(if)g(v)g(precedes)h
(u;)f(u)g(and)g(v)g(are)g Fp(c)n(oncurr)n(ent)f Fu(if)i(neither)g(u)f
(precedes)h(v)75 1767 y(nor)f(v)g(precedes)h(u.)166 1824 y(A)f(correct)g(MPI)
g(program)f(m)o(ust)h(ob)q(ey)g(the)h(follo)o(wing)g(constrain)o(ts.)131
1910 y(1.)22 b(An)e(RMA)h(up)q(date)g(\(put)f(or)g(accum)o(ulate\))g(to)g(a)g
(windo)o(w)h(is)g(not)e(concurren)o(t)i(with)g(a)f(store)189
1966 y(up)q(date)15 b(to)f(that)g(same)g(windo)o(w.)20 b(Ev)o(en)14
b(up)q(dates)i(to)e(distinct)h(lo)q(cations)h(of)e(the)g(same)h(windo)o(w)189
2023 y(are)g(prohibited.)131 2113 y(2.)22 b(An)15 b(RMA)f(up)q(date)h(to)f(a)
h(windo)o(w)g(is)g(not)f(concurren)o(t)h(to)e(another)i(RMA)f(up)q(date)i(to)
e(the)g(same)189 2170 y(windo)o(w,)k(if)h(they)f(access)g(o)o(v)o(erlaping)g
(lo)q(cations.)29 b(The)18 b(only)h(exception)g(to)e(this)h(rule)h(is)g(that)
189 2226 y(the)d(same)h(v)m(ariable)h(can)f(b)q(e)g(up)q(dated)g(b)o(y)g(t)o
(w)o(o)e(concurren)o(t)i Fo(MPI)p 1348 2226 V 16 w(A)o(CCUMULA)l(TE)g
Fu(calls,)h(that)189 2283 y(use)d(the)g(same)g(accum)o(ulate)h(op)q(eration,)
f(with)h(the)f(same)g(basic)h(datat)o(yp)q(e.)166 2452 y Fh(Discussion:)33
b Fk(W)m(e)12 b(can)h(c)o(ho)q(ose)g(to)g(lab)q(el)f(programs)f(that)i(break)
g(one,)g(or)f(b)q(oth)h(rules,)g(either)h(unde\014ned)75 2508
y(or)g(erroneous.)166 2647 y Fu(If)20 b(a)g(program)e(ob)q(eys)j(these)f
(restrictions,)h(then)f(RMA)g(op)q(erations)g(will)i(ha)o(v)o(e)d(the)h
(follo)o(wing)75 2704 y(b)q(eha)o(vior:)1967 46 y Fj(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: 19 19
bop 75 -100 a Fl(1.5.)34 b(SEMANTICS)16 b(AND)f(CORRECTNESS)896
b Fu(19)166 49 y(A)16 b(read)g(r)g(\(get)g(or)g(load\))g(from)f(a)h(lo)q
(cation)h(in)g(the)f(windo)o(w)h(will)h(return)e(the)g(v)m(alue)i(written)e
(b)o(y)75 106 y(a)f(write)g(w)g(\(put,)g(accum)o(ulate)g(or)g(store\))f(to)h
(that)f(lo)q(cation)i(if)75 162 y(The)f(write)h(w)f(precedes)h(the)f(read)g
(r,)g(and)75 219 y(an)o(y)g(other)g(write)g(w')g(to)f(that)h(lo)q(cation)h
(either)g(precedes)g(the)f(write)g(w)g(or)g(folllo)o(ws)h(the)f(read)g(r.)166
275 y(The)g(v)m(alue)i(returned)f(b)o(y)f(the)g(read)h(is)f(unde\014ned)i(if)
f(no)f(write)h(satis\014es)f(this)h(condition)h(\(e.g.,)c(if)75
332 y(no)j(write)g(precedes)g(the)g(read,)g(or)f(a)g(write)h(w)g(precedes)h
(the)f(read)f(r,)h(and)g(another)f(write)h(w')f(neither)75
388 y(precedes)h(w)f(nor)g(follo)o(ws)g(r\).)189 493 y Fp(R)n(ationale.)57
b Fu(The)18 b(correctness)g(criteria)h(deal)g(only)f(with)h(con\015icts)g
(that)e(in)o(v)o(olv)o(e)i(RMA)f(op-)189 549 y(erations.)33
b(Con\015icts)21 b(b)q(et)o(w)o(een)f(b)q(et)o(w)o(een)g(load/store)f
(accesses)i(to)e(a)g(windo)o(w)h(will)i(b)q(e)f(dealt)189 606
y(according)16 b(to)g(the)g(shared)g(memory)g(seman)o(tics)g(of)g(the)g
(underlying)i(arc)o(hitecture,)e(whic)o(h)h(are)189 662 y(outside)e(the)h
(scop)q(e)f(of)g(MPI.)189 737 y(The)c(\014rst)g(constrain)o(t)g(on)g(correct)
g(RMA)g(accesses)g(ma)o(y)g(seem)g(unduly)i(restrictiv)o(e,)f(as)f(it)h
(forbids)189 793 y(concurren)o(t)22 b(accesses)h(to)f(nono)o(v)o(erlaping)i
(lo)q(cations)f(in)g(a)g(windo)o(w.)42 b(The)23 b(reason)f(for)h(this)189
850 y(constrain)o(t)14 b(is)h(that,)f(on)g(some)h(arc)o(hitectures,)f
(explicit)j(coherence)f(restoring)e(op)q(erations)h(ma)o(y)189
906 y(b)q(e)c(needed)h(at)e(sync)o(hronization)h(p)q(oin)o(ts.)19
b(A)11 b(di\013eren)o(t)f(op)q(eration)h(ma)o(y)f(b)q(e)h(needed)h(for)e(lo)q
(cations)189 963 y(that)g(where)i(lo)q(cally)h(up)q(dated)f(b)o(y)f(stores)f
(and)i(for)f(lo)q(cations)h(that)e(where)i(remotely)f(up)q(dated)h(b)o(y)189
1019 y(put)k(or)f(accum)o(ulate)h(op)q(erations.)23 b(Without)15
b(this)i(constrain)o(t,)e(the)h(MPI)g(library)h(will)g(ha)o(v)o(e)f(to)189
1076 y(trac)o(k)9 b(precisely)k(whic)o(h)f(lo)q(cations)f(in)h(a)e(windo)o(w)
h(where)g(up)q(dated)h(b)o(y)e(a)h(put)g(or)f(accum)o(ulate)h(call.)189
1132 y(The)20 b(additional)h(o)o(v)o(erhead)e(of)g(main)o(taining)i(suc)o(h)f
(information)f(is)i(considered)g(prohibitiv)o(e.)189 1189 y(\()p
Fp(End)15 b(of)i(r)n(ationale.)p Fu(\))75 1310 y Fe(1.5.1)49
b(Implem)o(e)o(ntation)14 b(mo)q(del)75 1396 y Fo(Distributed)j(Memo)o(r)o(y)
75 1482 y Fu(The)h(data)f(transfer)g(e\013ected)h(b)o(y)g(a)f(put)h(can)g(b)q
(e)g(implemen)o(ted)i(in)e(a)g(distributed)h(memory)e(system)75
1538 y(b)o(y)h(a)f(matc)o(hing)h(pair)h(of)e(send)i(and)f(receiv)o(e.)29
b(The)18 b(origin)g(pro)q(cess)g(sends)h(the)f(put)g(request,)g(whic)o(h)75
1595 y(includes)23 b(the)e(target)f(displacemen)o(t,)k(coun)o(t)d(and)g
(datat)o(yp)q(e)f(argumen)o(ts,)i(to)e(the)h(target)f(pro)q(cess.)75
1651 y(The)d(origin)g(next)f(sends)h(the)f(data,)g(using)h(a)f(standard)f
Fo(MPI)p 1157 1651 14 2 v 16 w(SEND)i Fu(call,)g(with)g(the)f(origin)h
(address,)75 1708 y(coun)o(t,)h(and)g(datat)o(yp)q(e)g(argumen)o(ts.)27
b(The)18 b(target)f(pro)q(cess)h(receiv)o(es)h(the)f(put)g(request)g(and)h(p)
q(osts)e(a)75 1764 y(corresp)q(onding)j(receiv)o(e,)i(using)e(the)f(target)g
(displacemen)o(t,)i(coun)o(t)e(and)h(datat)o(yp)q(e)f(argumen)o(ts.)32
b(A)75 1820 y(similar)16 b(sc)o(heme)g(w)o(orks)e(for)h(the)g(other)g(RMA)g
(transfers.)166 1877 y(In)j(2-part)o(y)d(comm)o(unication,)j(it)f(is)h
(su\016cien)o(t)g(that)e(the)h(target)f(pro)q(cess)h(handles)h(p)q(ending)h
(re-)75 1933 y(quests)h(when)h(the)f(pro)q(cess)h(executes)g(RMA)f(sync)o
(hronization)h(calls)h(that)d(precede)i(load)g(or)f(store)75
1990 y(accesses)h(to)f(the)h(windo)o(w)f(\()p Fo(MPI)p 671
1990 V 16 w(BARRIER,)h Fu(and)42 b Fo(MPI)p 1105 1990 V 16
w(W)l(AIT)p Fu(\).)20 b(In)h(3-part)o(y)e(comm)o(unication,)k(it)75
2046 y(b)q(ecomes)e(necessary)f(to)f(ha)o(v)o(e)h(an)g(async)o(hronous)f
(agen)o(t)h(\(handler,)h(thread,)g(daemon,)g(etc.\))34 b(that)75
2103 y(pro)q(cesses)16 b(p)q(ending)h(requests)e(ev)o(en)g(if)h(no)f(MPI)g
(calls)h(o)q(ccur)g(on)f(the)g(target)f(pro)q(cess.)75 2223
y Fo(Sha)o(red)i(Memo)n(ry)75 2308 y Fu(Put)23 b(can)h(b)q(e)g(implemen)o
(ted)h(in)f(a)g(homogeneous)f(shared)g(memory)g(en)o(vironmen)o(t)h(b)o(y)f
(co)q(de)h(exe-)75 2365 y(cuted)d(b)o(y)f(the)g(calling)i(pro)q(cess.)35
b(Supp)q(ose)21 b(one)f(has,)h(in)g(this)f(en)o(vironmen)o(t,)i(an)e
(implemen)o(tation)75 2421 y(of)f(send/receiv)o(e,)i(using)f
(memory-to-memory)e(cop)o(y)l(.)32 b(The)20 b(put)f(op)q(eration)h(can)f(b)q
(e)h(implemen)o(ted)75 2478 y(as)g(a)g(\\send")h(from)e(origin)i(memory)f(to)
g(an)g(in)o(termediate)h(bu\013er,)h(implemen)o(ted)g(as)e(memory-to-)75
2534 y(memory)c(cop)o(y)l(,)h(follo)o(w)o(ed)g(b)o(y)g(a)f(\\receiv)o(e")i
(from)e(the)h(in)o(termediate)g(bu\013er)g(to)f(the)h(target)f(memory)l(,)75
2591 y(implemen)o(ted)h(again)f(using)g(memory-to-memory)f(cop)o(y)l(.)21
b(The)16 b(in)o(termediate)g(cop)o(y)g(can)f(b)q(e)i(a)o(v)o(oided)75
2647 y(\(or)c(pip)q(elined\))j(for)d(simple)i(\(con)o(tiguous\))e(bu\013ers)g
(or)g(with)h(a)f(more)g(elab)q(orate)h(implemen)o(tation.)20
b(On)75 2704 y(the)c(other)g(hand,)h(additional)h(cop)o(ying)f(ma)o(y)e(b)q
(e)i(required)h(to)d(a)o(v)o(oid)i(alignmen)o(t)g(problems.)23
b(Similar)-32 46 y Fj(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: 20 20
bop 75 -100 a Fu(20)723 b Fl(CHAPTER)15 b(1.)35 b(ONE)16 b(SIDED)f(COMMUNICA)
l(TIONS)75 49 y Fu(tec)o(hniques)i(can)f(b)q(e)h(used)f(for)f(other)h
(systems)f(that)g(supp)q(ort)h(some)g(form)f(of)g(cop)o(ying)i(from)e(lo)q
(cal)i(to)75 106 y(remote)e(memory)f(\(no)q(de-to-no)q(de)i(DMA,)e
(re\015ectiv)o(e)j(memory)l(,)d(etc.\).)166 162 y(An)c(additional)i
(memory-to-memory)d(cop)o(y)h(b)o(y)g(the)g(target)f(pro)q(cess)h(is)h
(required)g(if)g(b)q(oth)f(pro)q(cesses)75 219 y(ha)o(v)o(e)15
b(access)g(to)g(shared)g(memory)l(,)f(but)i(the)f(origin)h(pro)q(cess)f
(cannot)g(access)h(the)f(arget)f(windo)o(w.)166 275 y(V)l(arious)h(w)o(eak)g
(memory)g(mo)q(dels)h(in)o(tro)q(duce)g(the)f(follo)o(wing)h(complexities:)75
360 y Fn(Ordering)23 b Fu(If)15 b(memory)g(is)h(not)f(serially)i(consisten)o
(t,)e(explicit)i(calls)g(are)e(needed)i(to)d(ensure)i(that)f(ac-)189
417 y(cesses)e(b)o(y)h(di\013eren)o(t)f(pro)q(cessors)g(are)g(ordered)h
(correctly)l(,)g(at)e(p)q(oin)o(ts)i(that)f(indicate)i(completion)189
473 y(of)f(RMA)i(op)q(erations.)75 564 y Fn(Coherence)22 b
Fu(In)g(man)o(y)e(of)g(the)h(target)e(systems)h(lo)q(cal)i(load/store)e
(accesses)h(are)f(not)h(necessarily)189 620 y(coheren)o(t)13
b(with)h(remote)f(put/get)g(accesses.)19 b(It)14 b(ma)o(y)e(b)q(e)j(the)e
(case)h(that)e(put/get)h(access)h(shared)189 676 y(memory)l(,)j(and)h(that)f
(lo)q(cal)h(cac)o(hes)g(are)f(not)g(main)o(tained)i(coheren)o(t)e(with)h
(memory)l(.)27 b(Or,)18 b(that)189 733 y(a)g(windo)o(w)h(is)h(in)f(priv)m
(ate)h(memory)l(,)f(and)g(a)f(\(noncoheren)o(t\))h(mirror)f(cop)o(y)h(is)g(k)
o(ept)g(in)h(shared)189 789 y(memory)l(,)14 b(where)i(it)f(can)g(b)q(e)h
(accessed)g(b)o(y)f(other)g(pro)q(cesses.)75 880 y Fn(A)o(tomicit)o(y)23
b Fu(The)15 b(implemen)o(tation)h(m)o(ust)f(guaran)o(tee)g(that)f(accum)o
(ulates)i(are)f(atomic.)75 970 y Fn(Gran)o(ularit)o(y)23 b
Fu(Man)o(y)11 b(target)g(systems)g(do)h(not)g(supp)q(ort)g(atomic)g(stores)f
(of)g(arbitrarily)i(sized)g(en)o(tries.)189 1027 y(Load)j(and)h(stores)e(ma)o
(y)h(b)q(e)h(restricted)f(to)g(w)o(ords;)g(and)g(DMA)g(engines)h(ma)o(y)f(b)q
(e)h(restricted)g(to)189 1083 y(cac)o(he)e(line)i(or)e(page)g(transfers.)166
1168 y(The)g(seman)o(tics)f(of)h(RMA)f(op)q(erations)h(allo)o(w)g(us)g(to)f
(asso)q(ciate)g(coherence)i(restoring)e(and)h(mem-)75 1225
y(ory)k(sync)o(hronization)i(op)q(erations)f(with)g(barrier)f(and)h(lo)q(c)o
(k)g(sync)o(hronization)h(calls.)34 b(Th)o(us,)21 b(RMA)75
1281 y(op)q(erations)d(can)g(b)q(e)h(implemen)o(ted)h(e\016cien)o(tly)f(on)f
(shared)g(memory)f(systems)h(with)g(w)o(eak)g(memory)75 1338
y(seman)o(tics.)166 1394 y(W)l(e)g(\014rst)g(describ)q(e)i(a)e(generic)h
(implemen)o(tation)h(mo)q(del,)f(and)g(then)f(map)g(this)h(generic)g(imple-)
75 1451 y(men)o(tation)c(on)o(to)f(concrete)i(examples.)166
1507 y(In)h(our)g(generic)h(mo)q(del,)g(there)f(are)f(t)o(w)o(o)g(copies)h
(of)g(eac)o(h)g(RMA)g(windo)o(w:)23 b(a)17 b Fp(private)g Fu(cop)o(y)g(and)75
1563 y(an)i Fp(RMA)f Fu(cop)o(y)l(.)31 b(A)19 b(load)g(acccesses)g(the)g
(priv)m(ate)h(cop)o(y;)g(a)e(store)h(up)q(dates)g(the)g(priv)m(ate)g(cop)o(y)
l(,)h(and)75 1620 y(p)q(ossibly)e(the)e(RMA)g(cop)o(y;)g(a)f(get)h(accesses)g
(the)g(RMA)g(cop)o(y;)g(a)g(put)g(up)q(dates)h(the)f(RMA)g(cop)o(y)l(,)g(and)
75 1676 y(p)q(ossibly)h(the)e(priv)m(ate)h(cop)o(y)l(.)k(This)c(is)f
(illustrated)i(in)f(Figure)f(1.3.)75 1795 y Fo(Ordering:)45
b Fu(The)23 b(implemen)o(tation)g(m)o(ust)f(ensure)h(that)f(RMA)g(accesses)h
(ha)o(v)o(e)e(completed)j(in)f(the)75 1851 y(RMA)16 b(windo)o(w)g(cop)o(y)g
(at)f(those)g(p)q(oin)o(ts)i(in)f(the)g(execution)h(where)f(RMA)g(op)q
(erations)g(are)g(said)g(to)f(b)q(e)75 1908 y(complete)h(at)e(the)i(target.)
75 2026 y Fo(Coherence:)46 b Fu(The)21 b(rules)g(on)f(correct)g(use)h(of)f
(RMA)h(calls)g(imply)h(that)e(accesses)g(to)g(windo)o(ws)h(can)75
2083 y(b)q(e)c(divided)h(in)o(to)f Fp(private)g(ep)n(o)n(chs)p
Fu(,)f(where)h(the)f(windo)o(w)h(is)f(accessed)h(b)o(y)f(loads)h(or)f(stores)
f(and)i Fp(glob)n(al)75 2139 y(ep)n(o)n(chs)p Fu(,)d(where)i(the)f(windo)o(w)
h(is)g(accessed)g(b)o(y)f(RMA)g(calls.)22 b(A)15 b(priv)m(ate)h(ep)q(o)q(c)o
(h)g(starts)e(\(and)i(an)f(RMA)75 2195 y(ep)q(o)q(c)o(h)f(ends\))g(with)g(a)g
(call)g(to)f Fo(MPI)p 682 2195 14 2 v 16 w(BARRIER)i Fu(or)e
Fo(MPI)p 1036 2195 V 16 w(RMA)p 1151 2195 V 17 w(W)l(AIT)p
Fu(;)g(a)g(priv)m(ate)i(ep)q(o)q(c)o(h)f(ends)h(\(and)e(an)75
2252 y(RMA)i(ep)q(o)q(c)o(h)h(starts\))e(with)i(a)e(call)j(to)d
Fo(MPI)p 839 2252 V 16 w(BARRIER)j Fu(or)d Fo(MPI)p 1196 2252
V 16 w(RMA)p 1311 2252 V 17 w(CLEAR)p Fu(.)166 2308 y(Coherence)i(is)f(ac)o
(hiev)o(ed)h(b)o(y)f(cop)o(ying)h(the)f(RMA)g(cop)o(y)g(in)o(to)g(the)g(priv)
m(ate)h(cop)o(y)f(when)g(a)g(priv)m(ate)75 2365 y(ep)q(o)q(c)o(h)21
b(starts;)g(and)f(cop)o(ying)h(the)f(priv)m(ate)h(cop)o(y)f(in)o(to)h(the)f
(RMA)g(cop)o(y)g(when)h(the)f(priv)m(ate)h(ep)q(o)q(c)o(h)75
2421 y(ends.)f(There)13 b(is)h(one)f(di\016cult)o(y:)20 b(a)13
b(call)h(to)e Fo(MPI)p 911 2421 V 16 w(BARRIER)j Fu(can)e(signify)h(either)g
(the)f(b)q(eginning)i(or)e(the)75 2478 y(end)j(of)f(a)h(priv)m(ate)g(ep)q(o)q
(c)o(h.)21 b(In)c(order)e(to)g(distinguish)i(b)q(et)o(w)o(een)f(the)g(t)o(w)o
(o,)e(one)h(ma)o(y)g(main)o(tain)h(a)f Fp(dirty)75 2534 y(bit)j
Fu(that)g(is)h(set)g(when)g(the)f(windo)o(w)h(is)g(up)q(dated)h(b)o(y)e(a)g
(put)h(or)f(accum)o(ulate)h(call,)h(and)f(reset)f(when)75 2591
y Fo(MPI)p 160 2591 V 16 w(BARRIER)h Fu(is)f(called.)30 b(A)17
b(call)i(to)e Fo(MPI)p 862 2591 V 16 w(BARRIER)i Fu(is)g(assumed)f(to)f
(complete)h(an)g(RMA)g(ep)q(o)q(c)o(h)75 2647 y(if)f(the)f(bit)h(is)f(set,)g
(a)g(priv)m(ate)h(ep)q(o)q(c)o(h,)g(otherwise.)23 b(It)16 b(is)h(not)e
(necessary)i(to)e(cop)o(y)h(the)g(en)o(tire)h(windo)o(w:)75
2704 y(only)f(lo)q(cations)g(that)e(ha)o(v)o(e)h(b)q(een)h(up)q(dated)g(need)
g(b)q(e)g(copied.)1967 46 y Fj(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: 21 21
bop 75 -100 a Fl(1.5.)34 b(SEMANTICS)16 b(AND)f(CORRECTNESS)896
b Fu(21)75 4 y
28417720 14208860 0 0 42692362 21707980 startTexFig
75 4 a
%%BeginDocument: newindow.eps
/$F2psDict 200 dict def
$F2psDict begin
$F2psDict /mtrx matrix put
/col-1 {} def
/col0 {0.000 0.000 0.000 srgb} bind def
/col1 {0.000 0.000 1.000 srgb} bind def
/col2 {0.000 1.000 0.000 srgb} bind def
/col3 {0.000 1.000 1.000 srgb} bind def
/col4 {1.000 0.000 0.000 srgb} bind def
/col5 {1.000 0.000 1.000 srgb} bind def
/col6 {1.000 1.000 0.000 srgb} bind def
/col7 {1.000 1.000 1.000 srgb} bind def
/col8 {0.000 0.000 0.560 srgb} bind def
/col9 {0.000 0.000 0.690 srgb} bind def
/col10 {0.000 0.000 0.820 srgb} bind def
/col11 {0.530 0.810 1.000 srgb} bind def
/col12 {0.000 0.560 0.000 srgb} bind def
/col13 {0.000 0.690 0.000 srgb} bind def
/col14 {0.000 0.820 0.000 srgb} bind def
/col15 {0.000 0.560 0.560 srgb} bind def
/col16 {0.000 0.690 0.690 srgb} bind def
/col17 {0.000 0.820 0.820 srgb} bind def
/col18 {0.560 0.000 0.000 srgb} bind def
/col19 {0.690 0.000 0.000 srgb} bind def
/col20 {0.820 0.000 0.000 srgb} bind def
/col21 {0.560 0.000 0.560 srgb} bind def
/col22 {0.690 0.000 0.690 srgb} bind def
/col23 {0.820 0.000 0.820 srgb} bind def
/col24 {0.500 0.190 0.000 srgb} bind def
/col25 {0.630 0.250 0.000 srgb} bind def
/col26 {0.750 0.380 0.000 srgb} bind def
/col27 {1.000 0.500 0.500 srgb} bind def
/col28 {1.000 0.630 0.630 srgb} bind def
/col29 {1.000 0.750 0.750 srgb} bind def
/col30 {1.000 0.880 0.880 srgb} bind def
/col31 {1.000 0.840 0.000 srgb} bind def

end
save
-90.0 425.0 translate
1 -1 scale

/clp {closepath} bind def
/ef {eofill} bind def
/gr {grestore} bind def
/gs {gsave} bind def
/l {lineto} bind def
/m {moveto} bind def
/n {newpath} bind def
/s {stroke} bind def
/slc {setlinecap} bind def
/slj {setlinejoin} bind def
/slw {setlinewidth} bind def
/srgb {setrgbcolor} bind def
/rot {rotate} bind def
/sc {scale} bind def
/tr {translate} bind def
/tnt {dup dup currentrgbcolor
4 -2 roll dup 1 exch sub 3 -1 roll mul add
4 -2 roll dup 1 exch sub 3 -1 roll mul add
4 -2 roll dup 1 exch sub 3 -1 roll mul add srgb}
bind def
/shd {dup dup currentrgbcolor 4 -2 roll mul 4 -2 roll mul
4 -2 roll mul srgb} bind def
/$F2psBegin {$F2psDict begin /$F2psEnteredState save def} def
/$F2psEnd {$F2psEnteredState restore end} def

$F2psBegin
10 setmiterlimit
0.06000 0.06000 sc
7.500 slw
n 6000 5100 m 7200 5100 l 7200 5400 l 6000 5400 l clp gs col-1 s gr
/Times-Roman findfont 180.00 scalefont setfont
6150 5325 m
gs 1 -1 sc (? ? ? ? ?) col-1 show gr
[1 50.0] 50.000000 setdash
n 8700 4500 m 8700 4200 l 5700 4200 l 5700 4500 l clp gs col-1 s gr [] 0 setdash
30.000 slw
n 6300 3000 m 6600 3900 l gs col-1 s gr
n 6573.91 3513.41 m 6600.00 3900.00 l 6388.92 3575.08 l gs col-1 s gr
7.500 slw
n 6000 4200 m 7200 4200 l 7200 4500 l 6000 4500 l clp gs col7 0.50 shd ef gr gs col-1 s gr
[1 50.0] 50.000000 setdash
n 8700 5400 m 8700 5100 l 5700 5100 l 5700 5400 l clp gs col-1 s gr [] 0 setdash
15.000 slw
[50.0] 0 setdash
n 6600 4575 m 6600 5025 l gs col-1 s gr [] 0 setdash
7.500 slw
n 6630.00 4905.00 m 6600.00 5025.00 l 6570.00 4905.00 l gs col-1 s gr
/Times-Bold findfont 240.00 scalefont setfont
6075 2775 m
gs 1 -1 sc (PUT) col-1 show gr
n 9600 4200 m 10800 4200 l 10800 4500 l 9600 4500 l clp gs col-1 s gr
/Times-Roman findfont 180.00 scalefont setfont
9750 4425 m
gs 1 -1 sc (? ? ? ? ?) col-1 show gr
n 9600 5100 m 10800 5100 l 10800 5400 l 9600 5400 l clp gs col7 0.50 shd ef gr gs col-1 s gr
[1 50.0] 50.000000 setdash
n 12300 5400 m 12300 5100 l 9300 5100 l 9300 5400 l clp gs col-1 s gr [] 0 setdash
[1 50.0] 50.000000 setdash
n 12300 4500 m 12300 4200 l 9300 4200 l 9300 4500 l clp gs col-1 s gr [] 0 setdash
30.000 slw
n 9900 6600 m 10200 5700 l gs col-1 s gr
n 9988.92 6024.92 m 10200.00 5700.00 l 10173.91 6086.59 l gs col-1 s gr
15.000 slw
[50.0] 0 setdash
n 10200 5025 m 10200 4575 l gs col-1 s gr [] 0 setdash
7.500 slw
n 10170.00 4695.00 m 10200.00 4575.00 l 10230.00 4695.00 l gs col-1 s gr
/Times-Bold findfont 240.00 scalefont setfont
9525 6975 m
gs 1 -1 sc (STORE) col-1 show gr
30.000 slw
n 1800 3000 m 2100 3900 l gs col-1 s gr
n 2073.91 3513.41 m 2100.00 3900.00 l 1888.92 3575.08 l gs col-1 s gr
n 4200 3900 m 4500 3000 l gs col-1 s gr
n 4288.92 3324.92 m 4500.00 3000.00 l 4473.91 3386.59 l gs col-1 s gr
n 1800 6600 m 2100 5700 l gs col-1 s gr
n 1888.92 6024.92 m 2100.00 5700.00 l 2073.91 6086.59 l gs col-1 s gr
n 4200 5700 m 4500 6600 l gs col-1 s gr
n 4473.91 6213.41 m 4500.00 6600.00 l 4288.92 6275.08 l gs col-1 s gr
7.500 slw
[1 50.0] 50.000000 setdash
n 4800 4500 m 4800 4200 l 1800 4200 l 1800 4500 l clp gs col-1 s gr [] 0 setdash
[1 50.0] 50.000000 setdash
n 4800 5400 m 4800 5100 l 1800 5100 l 1800 5400 l clp gs col-1 s gr [] 0 setdash
/Times-Bold findfont 240.00 scalefont setfont
1500 2775 m
gs 1 -1 sc (PUT) col-1 show gr
/Times-Bold findfont 240.00 scalefont setfont
4275 2775 m
gs 1 -1 sc (GET) col-1 show gr
/Times-Bold findfont 240.00 scalefont setfont
1500 7050 m
gs 1 -1 sc (STORE) col-1 show gr
/Times-Bold findfont 240.00 scalefont setfont
4200 7050 m
gs 1 -1 sc (LOAD) col-1 show gr
/Times-Bold findfont 240.00 scalefont setfont
2625 5325 m
gs 1 -1 sc (private copy) col-1 show gr
/Times-Bold findfont 270.00 scalefont setfont
2700 1800 m
gs 1 -1 sc (Window) col-1 show gr
/Times-Bold findfont 270.00 scalefont setfont
6225 1800 m
gs 1 -1 sc (RMA Update) col-1 show gr
/Times-Bold findfont 270.00 scalefont setfont
9900 1800 m
gs 1 -1 sc (Local Update) col-1 show gr
/Times-Bold findfont 240.00 scalefont setfont
2550 4425 m
gs 1 -1 sc (RMA copy) col-1 show gr
$F2psEnd
restore
%%EndDocument
75 4 a
endTexFig
533 1002 a Fu(Figure)15 b(1.3:)k(Sc)o(hematic)d(description)g(of)f(windo)o(w)
75 1139 y Fo(A)o(tomicit)o(y)75 1225 y Fu(The)h(atomicit)o(y)g(of)g(accum)o
(ulate)g(up)q(date)h(can)f(b)q(e)h(guaran)o(teed)f(b)o(y)g(using)h(an)f(in)o
(ternal)h(windo)o(w)f(lo)q(c)o(k.)75 1281 y(Since)j(puts)f(and)g(gets)f
(cannot)h(con\015ict)h(with)f(accum)o(ulates,)g(there)g(is)h(no)e(need)i(to)e
(c)o(hec)o(k)h(this)g(lo)q(c)o(k)75 1338 y(for)g(regular)h(put)g(and)g(get)g
(accesses.)31 b(Finer)19 b(lo)q(c)o(king)h(is)g(p)q(ossible,)h(with)e(a)g
(more)f(complex)i(lo)q(c)o(king)75 1394 y(proto)q(col.)75 1514
y Fo(Granula)o(rit)o(y)75 1600 y Fu(Supp)q(ose)15 b(that)e(RMA)h(accesses)g
(are)f(restricted)i(to)e(aligned)i(w)o(ords,)e(and)h(up)q(dates)g(of)g
(smaller)g(en)o(tities)75 1657 y(require)20 b(a)f(load-mo)q(dify-store)h
(cycle.)33 b(Then,)21 b(correct)d(b)q(eha)o(vior)i(can)g(b)q(e)g(ac)o(hiev)o
(ed)g(b)o(y)f(protecting)75 1713 y(partial)12 b(w)o(ord)e(accesses)i(with)f
(an)g(in)o(ternal)h(windo)o(w)g(lo)q(c)o(k)f(\(can)g(b)q(e)h(the)g(same)f(as)
f(for)h(accum)o(ulate\).)19 b(F)l(ull)75 1769 y(w)o(ord)c(accesses)g(need)h
(not)f(b)q(e)h(lo)q(c)o(k)g(protected.)k(Finer)c(lo)q(c)o(k)f(gran)o(ularit)o
(y)g(\(and)g(higher)h(concurrency\))75 1826 y(is)g(p)q(ossible)i(b)o(y)e
(using)g(di\013eren)o(t)h(lo)q(c)o(ks)f(for)f(di\013eren)o(t)h(parts)f(of)h
(the)g(windo)o(w.)22 b(A)16 b(shared)g(lo)q(c)o(k)g(can)g(b)q(e)75
1882 y(used)g(for)e(gets,)h(and)g(an)g(exclusiv)o(e)i(lo)q(c)o(k)f(for)e
(puts.)166 1939 y(Indeed,)19 b(con\015icting)g(accesses)f(can)f(only)h(b)q(e)
g(due)g(to)f Fp(false)g(sharing)p Fu(,)g(i.e.)27 b(concurren)o(t)18
b(accesses)75 1995 y(to)h(di\013eren)o(t)h(parts)f(of)h(the)g(same)f(w)o
(ord.)34 b(This,)21 b(b)q(ecause)g(RMA)f(accesses)g(cannot)f(b)q(e)i
(concurren)o(t)75 2052 y(with)14 b(lo)q(cal)g(accesses,)f(and)h(\(truely\))f
(con\015icting)h(RMA)g(accesses)f(cannot)g(b)q(e)h(concurren)o(t.)19
b(Th)o(us,)13 b(t)o(w)o(o)75 2108 y(con\015icting)20 b(accesses)g(m)o(ust)e
(b)q(e)i(partial)f(w)o(ord)f(accesses,)i(and)f(there)g(comp)q(ete)h(for)e
(the)h(same)g(lo)q(c)o(k.)75 2165 y(P)o(artial)g(w)o(ord)f(accum)o(ulates)i
(are)e(also)h(protected)g(b)o(y)g(that)g(lo)q(c)o(k)g(\(this)g(is)h(the)f
(case)g(an)o(yho)o(w)f(if)i(the)75 2221 y(same)15 b(lo)q(c)o(k)h(is)f(used)h
(for)f(accum)o(ulates)g(and)h(partial)f(w)o(ord)g(up)q(dates.\))166
2278 y(An)e(additional)g(problem)g(o)q(ccurs)g(if)g(the)f(windo)o(w)h(b)q
(oundaries)g(are)f(not)g(w)o(ord)g(aligned.)20 b(If)13 b(a)f(w)o(ord)75
2334 y(o)o(v)o(erlaps)g(the)g(windo)o(w)h(b)q(oundary)g(than)f(a)g(lo)q(cal)h
(access)g(to)e(the)i(part)f(of)f(the)i(w)o(ord)f(that)f(is)i(outside)g(the)75
2390 y(windo)o(w)j(can)f(b)q(e)i(concurren)o(t)e(with)h(an)g(RMA)f(access)h
(to)f(the)h(part)f(of)g(that)g(w)o(ord)g(that)f(is)i(inside)i(the)75
2447 y(windo)o(w.)h(This)14 b(con\015ict)f(is)g(a)o(v)o(oided)g(if)g(t)o(w)o
(o)f(copies)h(\(lo)q(cal)h(and)f(public\))h(of)f(this)g(w)o(ord)f(are)g(main)
o(tained)75 2503 y({)j(using)h(the)f(windo)o(w)h(cop)o(y)f(proto)q(col)g
(describ)q(ed)i(ab)q(o)o(v)o(e.)166 2560 y(An)j(implemen)o(tation)h(that)d(c)
o(hec)o(ks)i(for)f(out)g(of)g(windo)o(w)h(b)q(oundary)g(accesses)g(will)h
(not)e(su\013er)75 2616 y(additional)e(o)o(v)o(erheads,)d(as)h(long)g(as)g
(the)g(o)o(v)o(erlapping)h(w)o(ord)f(is)g(not)g(up)q(dated.)166
2673 y(W)l(e)20 b(sp)q(ecialize)i(no)o(w)d(this)h(generic)g(algorithm)g(to)f
(sp)q(eci\014c)i(arc)o(hitectures,)g(with)f(noncoheren)o(t)-32
46 y Fj(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: 22 22
bop 75 -100 a Fu(22)723 b Fl(CHAPTER)15 b(1.)35 b(ONE)16 b(SIDED)f(COMMUNICA)
l(TIONS)75 49 y Fu(cac)o(hes.)20 b(There)c(are)e(three)i(cases)f(to)f
(consider.)131 143 y(1.)22 b(RMA)16 b(is)g(implemen)o(ted)h(using)g
(non-coheren)o(t)f(shared)g(memory)l(,)f(with)h(write-through)g(nonco-)189
200 y(heren)o(t)i(cac)o(hes.)29 b(Both)18 b(lo)q(cal)h(and)f(remote)g(writes)
g(go)f(directly)j(to)d(memory)l(,)h(but)h(b)q(oth)f(lo)q(cal)189
256 y(and)d(remote)g(reads)g(ma)o(y)f(go)h(to)f(cac)o(he)i(\(Cra)o(y)e(T90,)g
(Cra)o(y)g(T3D?\).)131 350 y(2.)22 b(RMA)11 b(is)g(implemen)o(ted)h(using)g
(non-coheren)o(t)f(shared)g(memory)l(,)g(with)g(write-bac)o(k)g(noncoheren)o
(t)189 406 y(cac)o(hes.)18 b(Both)10 b(lo)q(cal)i(and)f(remote)f(accesses)g
(ma)o(y)g(go)g(to)g(cac)o(he,)h(rather)f(than)g(memory)g(\(Mercury)189
463 y(RA)o(CE)15 b(i860?\).)131 557 y(3.)22 b(RMA)13 b(is)i(implemen)o(ted)g
(using)f(a)f(non-coheren)o(t)h(DMA)f(engine.)21 b(Remote)14
b(accesses)g(go)f(directly)189 613 y(to)19 b(memory)l(,)g(whereas)h(lo)q(cal)
h(accesses)e(\(b)q(oth)h(reads)f(and)h(writes\))f(ma)o(y)g(go)g(to)g(cac)o
(he)g(\(NEC)189 669 y(Cenju-3,)c(Cra)o(y)f(T3D?\).)166 763
y(In)j(all)h(cases,)f(the)g(RMA)h(cop)o(y)e(of)h(the)g(windo)o(w)g(is)h(in)f
(memory)l(,)g(whereas)g(the)g(priv)m(ate)h(cop)o(y)e(of)75
820 y(the)f(windo)o(w)f(consists)h(of)f(the)h(v)m(alues)g(in)h(the)e(lo)q
(cal)i(pro)q(cessor)e(cac)o(he,)h(or)f(the)h(v)m(alues)g(in)h(memory)l(,)e
(for)75 876 y(lo)q(cations)i(that)e(are)h(not)g(cac)o(hed.)166
933 y(Completing)g(an)g(RMA)f(up)q(date)h(is)g(ac)o(hiev)o(ed)h(b)o(y)e
(\015ushing)i(the)f(cac)o(he)g(of)f(the)g(calling)j(pro)q(cessor,)75
989 y(so)g(that)f(the)h(up)q(date)h(b)q(ecomes)g(visible)h(in)f(the)f(RMA)g
(windo)o(w.)26 b(Suc)o(h)18 b(\015ush)f(call)i(will)f(b)q(e)g(executed)75
1046 y(as)j(part)g(of)g(calls)h(to)f Fo(MPI)p 554 1046 14 2
v 15 w(BARRIER,)i(MPI)p 874 1046 V 16 w(RMA)p 989 1046 V 16
w(COMPLETE)f Fu(and)f Fo(MPI)p 1444 1046 V 16 w(RMA)p 1559
1046 V 17 w(UNLOCK)p Fu(.)g(It)g(is)75 1102 y(su\016cien)o(t)16
b(to)e(\015ush)i(dirt)o(y)g(cac)o(he)f(lines)i(that)d(o)o(v)o(erlap)h(the)h
(windo)o(w.)166 1158 y(Cop)o(ying)f(of)g(the)g(priv)m(ate)h(cop)o(y)f(to)g
(the)g(RMA)g(cop)o(y)g(is)h(ac)o(hiev)o(ed)g(b)o(y)f(\015ushing)h(the)g(cac)o
(he)f(of)g(the)75 1215 y(lo)q(cal)h(pro)q(cessor.)k(It)15 b(is)h(su\016cien)o
(t)g(to)e(\015ush)i(dirt)o(y)f(cac)o(he)h(lines)h(that)d(o)o(v)o(erlap)h
(with)h(the)f(windo)o(w.)166 1271 y(Cop)o(ying)h(of)g(the)g(RMA)h(cop)o(y)f
(in)o(to)g(the)h(priv)m(ate)g(cop)o(y)f(is)g(ac)o(hiev)o(ed)i(b)o(y)e(in)o(v)
m(alidating)i(the)f(cac)o(he)75 1328 y(of)f(the)g(lo)q(cal)h(pro)q(cessor.)22
b(It)16 b(is)h(su\016cien)o(t)f(to)g(in)o(v)m(alidate)i(the)e(lines)i(that)d
(o)o(v)o(erlap)h(with)g(the)g(windo)o(w,)75 1384 y(and)j(are)f(dirt)o(y)h(in)
h(the)e(public)j(cop)o(y)l(.)31 b(On)19 b(some)g(systems,)f(\015ush)i(ma)o(y)
e(b)q(e)h(faster)f(than)g(in)o(v)m(alidate.)75 1441 y(Since)f(the)f(lo)q(cal)
i(cop)o(y)d(is)i(clean,)f(all)h(the)f(windo)o(w)h(cac)o(he)f(lines)h(are)f
(clean,)h(and)f(a)f(\015ush)i(call)g(ma)o(y)e(b)q(e)75 1497
y(used,)j(instead)h(of)e(an)g(in)o(v)m(alidate)j(\(assuming)d(that)g(the)h
(cac)o(he)g(con)o(troller)g(is)g(smart)e(enough)i(not)g(to)75
1554 y(write)d(bac)o(k)g(clean)i(cac)o(he)e(lines.\))166 1610
y(Tw)o(o)h(windo)o(w)h(copies)g(ma)o(y)f(also)h(b)q(e)g(used)h(in)f(a)f
(shared)h(memory)f(implemen)o(tation)i(where)f(not)75 1667
y(all)f(pro)q(cesses)g(can)g(access)g(all)g(memory)l(.)21 b(E.g.,)14
b(in)i(the)g(case)f(where)h(a)f(windo)o(w)h(is)g(in)g(pro)q(cess)g(priv)m
(ate)75 1723 y(memory)l(,)22 b(one)f(ma)o(y)f(use)h(an)g(RMA)h(cop)o(y)e
(that)h(is)g(in)h(shared)f(memory)g(and)g(is)g(accessible)i(b)o(y)e(all)75
1779 y(pro)q(cesses)16 b(in)h(the)f(comm)o(unication)g(group.)22
b(Or,)15 b(the)h(RMA)g(cop)o(y)g(could)g(b)q(e)h(in)g(nonpaged)f(memory)l(,)
75 1836 y(that)j(is)i(accessible)h(to)d(DMA,)g(whereas)h(the)g(priv)m(ate)h
(cop)o(y)f(is)g(in)h(pageable)g(memory)l(.)34 b(Then,)21 b(the)75
1892 y(cop)o(ying)13 b(of)g(one)g(cop)o(y)g(to)f(another)h(actually)h
(requires)f(memory)g(to)f(memory)g(cop)o(y)l(,)i(that)e(can)h(b)q(e)h(done)75
1949 y(b)o(y)j(the)g(windo)o(w)g(o)o(wning)g(pro)q(cess.)25
b(F)l(ull)18 b(windo)o(w)f(cop)o(ying)g(can)g(b)q(e)h(a)o(v)o(oided)f(b)o(y)g
(main)o(taining)g(\014ner)75 2005 y(dirt)o(y)d(bits.)19 b(\(Dirt)o(y)13
b(bits)h(for)f(the)h(public)i(cop)o(y)d(are)h(main)o(tained)g(b)o(y)g(trac)o
(king)f(the)h(lo)q(cations)g(accessed)75 2062 y(b)o(y)h(put)h(or)f(accum)o
(ulate)h(calls)g({)g(this)f(can)h(b)q(e)g(part)f(of)g(the)h(co)q(de)g(for)f
(these)g(calls.)22 b(Dirt)o(y)15 b(bits)h(for)f(the)75 2118
y(lo)q(cal)k(cop)o(y)f(are)g(harder)g(to)g(main)o(tain.)29
b(One)19 b(needs)g(to)f(use)g(tec)o(hniques)h(similar)h(to)d(those)h(used)h
(b)o(y)75 2175 y(soft)o(w)o(are)12 b(implemen)o(tations)i(of)f(distributed)i
(shared)f(memory:)19 b(e.g.,)12 b(use)i(page)f(protection)h(to)f(detect)75
2231 y(lo)q(cal)j(up)q(dates)g(at)f(page)g(gran)o(ularit)o(y)l(.)75
2353 y Fe(1.5.2)49 b(Registers)15 b(and)i(Compiler)e(Optimizations)75
2439 y Fu(Another)e(coherence)g(problem)h(exists)f(b)q(et)o(w)o(een)g(v)m
(ariables)h(k)o(ept)e(in)i(registers)e(and)h(the)g(memory)f(v)m(alue)75
2495 y(of)19 b(these)g(v)m(ariables.)32 b(An)20 b(RMA)f(call)h(ma)o(y)e
(access)h(a)g(v)m(ariable)h(in)g(memory)f(\(or)f(cac)o(he\),)h(while)i(the)75
2552 y(up-to-date)12 b(v)m(alue)h(of)f(this)h(v)m(ariable)g(is)g(in)g
(register.)18 b(A)12 b(get)g(will)i(not)e(return)g(the)g(latest)g(v)m
(ariable)h(v)m(alue,)75 2608 y(and)i(a)g(put)h(ma)o(y)e(b)q(e)i(o)o(v)o
(erwritten)e(when)i(the)f(register)g(is)h(stored)f(bac)o(k)g(in)h(memory)l(.)
166 2664 y(The)f(problem)h(is)g(illustrated)g(b)o(y)g(the)f(follo)o(wing)h
(co)q(de:)1967 46 y Fj(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(1.6.)34 b(A)15 b(F)o(OR)l(TRAN)h(PR)o(OBLEM)g(WITH)f
(REGISTER)i(OPTIMIZA)l(TION)361 b Fu(23)110 49 y(Source)16
b(of)f(Pro)q(cess)g(1)122 b(Source)16 b(of)f(Pro)q(cess)g(2)122
b(Executed)16 b(in)g(Pro)q(cess)f(2)110 114 y(bbbb)i(=)e(777)272
b(bu\013)15 b(=)h(999)296 b(reg)p 1190 114 14 2 v 16 w(A:=999)110
170 y(call)17 b(MPI)p 286 170 V 16 w(PUT\(bbbb)611 b(stop)15
b(appl.)8 b(thread)110 227 y(in)o(to)16 b(bu\013)f(of)g(pro)q(cess)g(2\))580
b(bu\013:=777)15 b(in)h(PUT)f(handler)1126 283 y(con)o(tin)o(ue)h(appl.)8
b(thread)110 340 y(call)17 b(MPI)p 286 340 V 16 w(BARRIER\(...\))32
b(call)17 b(MPI)p 794 340 V 16 w(BARRIER\(...\))618 396 y(ccc)f(=)g(bu\013)
304 b(ccc:=reg)p 1298 396 V 17 w(A)166 499 y(In)18 b(this)f(example,)h(v)m
(ariable)g(bu\013)f(is)h(allo)q(cated)g(in)g(the)f(register)g(reg)p
1380 499 V 16 w(A)g(and)g(therefore)g(ccc)g(will)75 556 y(ha)o(v)o(e)e(the)g
(old)h(v)m(alue)g(of)f(bu\013)g(and)h(not)e(the)i(new)f(v)m(alue)i(777.)166
612 y(This)11 b(problem,)h(whic)o(h)f(also)g(a\017icts)f(in)i(some)e(cases)h
(send/receiv)o(e)g(comm)o(unication,)h(is)f(discussed)75 669
y(more)k(at)f(length)i(in)g(Section)g(1.6.)166 725 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.)75 782 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)
75 838 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
894 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)75 951
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 1007 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 1064 y(to)h(v)m(ariables)h(stored)f(in)h Fi(COMMON)f Fu(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
Fu(\(while)75 1120 y Fi(V)o(OLA)m(TILE)e Fu(is)h(no)g(a)f(standard)h(F)l
(ortran)f(declaration,)h(it)h(is)f(supp)q(orted)g(b)o(y)g(man)o(y)g(F)l
(ortran)e(compilers\).)75 1177 y(Details,)k(and)h(an)f(additional)h(solution)
g(are)f(discussed)i(in)f(Section)g(1.6)e(on)i(page)f(23.)75
1319 y Fr(1.6)59 b(A)20 b(F)n(o)n(rtran)g(Problem)e(with)h(Register)h
(Optimization)75 1504 y Fh(Discussion:)15 b Fk(This)f(section)h(should)f(b)q
(e)g(included)g(in)o(to)f(MISC)189 1689 y Fp(A)n(dvic)n(e)i(to)i(users.)189
1764 y Fu(MPI)d(pro)o(vides)g(op)q(erations)g(whic)o(h)h(ma)o(y)e(b)q(e)i
(hidden)h(from)d(the)h(user)g(co)q(de)h(and)f(run)h(in)f(parallel)189
1820 y(with)g(it,)h(accessing)g(the)f(same)g(memory)g(as)g(user)g(co)q(de.)20
b(Examples)15 b(include)i(the)d(data)g(transfer)189 1877 y(for)c(an)h
Fo(MPI)p 398 1877 V 16 w(RECV)h Fu(The)f(optimizer)i(of)d(a)h(compiler)i
(will)f(assume)f(that)g(it)g(can)h(recognize)g(p)q(erio)q(ds)189
1933 y(when)j(a)g(cop)o(y)g(of)g(a)f(v)m(ariable)j(can)e(b)q(e)h(k)o(ept)f
(in)h(a)e(register)h(without)g(reloading)h(from)f(or)f(storing)189
1990 y(to)e(memory)l(.)18 b(When)13 b(the)g(user)g(co)q(de)g(is)g(w)o(orking)
f(with)h(a)f(register)h(cop)o(y)f(of)g(some)h(v)m(ariable)g(while)189
2046 y(the)i(hidden)i(op)q(eration)f(reads)f(or)g(writes)g(the)h(memory)f
(cop)o(y)l(,)g(problems)h(o)q(ccur.)k(This)c(section)189 2103
y(discusses)g(register)f(optimization)h(pitfalls.)189 2177
y(When)f(a)f(v)m(ariable)i(is)f(lo)q(cal)h(to)e(a)g(F)l(ortran)f(subroutine)j
(\(i.e.)k(not)14 b(in)h(a)g(COMMON)f(blo)q(c)o(k\),)h(the)189
2234 y(compiler)g(will)i(assume)d(that)g(it)h(cannot)f(b)q(e)h(mo)q(di\014ed)
h(b)o(y)f(a)f(called)i(subroutine)f(unless)h(it)f(is)g(an)189
2290 y(actual)g(argumen)o(t)g(of)h(the)f(call.)23 b(In)16 b(the)g(most)e
(common)i(link)m(age)h(con)o(v)o(en)o(tion,)e(the)h(subroutine)189
2347 y(is)d(exp)q(ected)i(to)d(sa)o(v)o(e)h(and)g(restore)g(certain)h
(registers.)19 b(Th)o(us,)13 b(the)g(optimizer)i(will)g(assume)e(that)189
2403 y(a)k(register)h(whic)o(h)h(held)g(a)e(v)m(alid)j(cop)o(y)d(of)h(suc)o
(h)g(a)g(v)m(ariable)h(b)q(efore)f(the)g(call)h(will)g(still)h(hold)e(a)189
2460 y(v)m(alid)e(cop)o(y)g(on)f(return.)189 2534 y(Normally)22
b(users)g(are)g(not)g(a\017icted)g(with)h(this.)41 b(But)22
b(the)g(user)g(should)h(pa)o(y)f(atten)o(tion)g(to)189 2591
y(this)f(section)g(if)g(in)h(his/her)f(program)e(a)i(bu\013er)f(argumen)o(t)g
(to)g(an)h Fo(MPI)p 1498 2591 V 16 w(SEND)p Fu(,)f Fo(MPI)p
1746 2591 V 16 w(RECV)189 2647 y Fu(etc.)25 b(uses)18 b(a)f(name)g(whic)o(h)h
(hides)g(the)f(actual)h(v)m(ariables)g(in)o(v)o(olv)o(ed.)27
b Fo(MPI)p 1493 2647 V 16 w(BOTTOM)18 b Fu(with)f(an)189 2704
y(MPI)p 281 2704 V 16 w(Datat)o(yp)q(e)i(con)o(taining)i(absolute)g
(addresses)f(is)h(one)f(example.)36 b(Creating)20 b(a)f(datat)o(yp)q(e)-32
46 y Fj(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 Fu(24)723 b Fl(CHAPTER)15 b(1.)35 b(ONE)16 b(SIDED)f(COMMUNICA)
l(TIONS)189 49 y Fu(whic)o(h)f(uses)f(one)g(v)m(ariable)i(as)e(an)g(anc)o
(hor)f(and)i(brings)f(along)h(others)e(b)o(y)h(using)h Fo(MPI)p
1659 49 14 2 v 16 w(ADDRESS)189 106 y Fu(to)h(determine)j(their)f(o\013sets)e
(from)g(the)i(anc)o(hor)f(is)h(another.)22 b(The)17 b(anc)o(hor)f(v)m
(ariable)i(w)o(ould)e(b)q(e)189 162 y(the)g(only)h(one)f(men)o(tioned)h(in)g
(the)g(call.)24 b(Also)16 b(atten)o(tion)g(m)o(ust)g(b)q(e)h(pa)o(y)o(ed)f
(if)h(MPI)f(op)q(erations)189 219 y(are)f(used)g(that)g(run)g(in)h(parallel)h
(with)f(the)f(user's)g(application.)189 293 y(The)g(follo)o(wing)h(example)g
(sho)o(ws,)e(what)h(F)l(ortran)f(compilers)i(are)f(allo)o(w)o(ed)h(to)e(do:)
224 414 y(This)i(source)f(...)663 b(can)15 b(b)q(e)h(compiled)h(as:)224
479 y(call)f(MPI)p 399 479 V 17 w(ADDRESS\(buf,bufaddr,ierr\))189
b(call)16 b(MPI)p 1344 479 V 17 w(ADDRESS\(buf,...\))224 535
y(call)g(MPI)p 399 535 V 17 w(TYPE)p 545 535 V 16 w(STR)o(UCT\(1,1,bufaddr,)
163 b(call)16 b(MPI)p 1344 535 V 17 w(TYPE)p 1490 535 V 16
w(STR)o(UCT\(...\))590 592 y(MPI)p 682 592 V 17 w(REAL,t)o(yp,ierr\))224
648 y(call)g(MPI)p 399 648 V 17 w(TYPE)p 545 648 V 16 w(COMMIT\(t)o(yp\))308
b(call)16 b(MPI)p 1344 648 V 17 w(TYPE)p 1490 648 V 16 w(COMMIT\(...\))224
705 y(v)m(al)p 283 705 V 17 w(old)g(=)g(buf)681 b(register)15
b(=)h(buf)1169 761 y(v)m(al)p 1228 761 V 17 w(old)g(=)f(register)224
818 y(call)h(MPI)p 399 818 V 17 w(RECV\(MPI)p 654 818 V 16
w(BOTTOM,1,t)o(yp,...\))107 b(call)16 b(MPI)p 1344 818 V 17
w(RECV\(MPI)p 1599 818 V 16 w(BOTTOM,...\))224 874 y(v)m(al)p
283 874 V 17 w(new)g(=)f(buf)664 b(v)m(al)p 1228 874 V 17 w(new)16
b(=)f(register)189 995 y(The)j(compiler)g(do)q(es)g(not)g(in)o(v)m(alidate)h
(the)f(register)g(b)q(ecause)g(it)g(cannot)f(see)h(that)f Fo(MPI)p
1746 995 V 16 w(RECV)189 1051 y Fu(c)o(hanges)h(the)g(v)m(alue)i(of)e
Fo(buf)p Fu(.)30 b(The)19 b(access)f(of)g Fo(buf)i Fu(is)f(hidden)h(b)o(y)e
(the)h(use)f(of)g Fo(MPI)p 1659 1051 V 16 w(ADDRESS)189 1108
y Fu(and)d Fo(MPI)p 362 1108 V 16 w(BOTTOM)p Fu(.)189 1182
y(The)g(next)g(example)h(sho)o(ws)f(extreme,)g(but)g(allo)o(w)o(ed)h(p)q
(ossibilities!)-1480 b Fm(>)15 b Fj(\(July\))224 1303 y Fu(Source)424
b(compiled)17 b(as)322 b(or)15 b(compiled)i(as)224 1368 y(call)f(MPI)p
399 1368 V 17 w(IRECV\(buf,..req\))i(call)f(MPI)p 955 1368
V 16 w(IRECV\(buf,..req\))h(call)f(MPI)p 1510 1368 V 16 w(IRECV\(buf,..req\))
779 1424 y(register)e(=)h(buf)278 b(b1)15 b(=)h(buf)224 1481
y(call)g(MPI)p 399 1481 V 17 w(W)-5 b(AIT\(req,..\))104 b(call)17
b(MPI)p 955 1481 V 16 w(W)-5 b(AIT\(req,..\))104 b(call)17
b(MPI)p 1510 1481 V 16 w(W)-5 b(AIT\(req,..\))224 1537 y(b1)15
b(=)h(buf)377 b(b1)15 b(:=)g(register)189 1658 y Fo(MPI)p 274
1658 V 15 w(W)l(AIT)e Fu(or)f(a)g(parallel)i(thread)e(mo)q(di\014es)i
Fo(buf)f Fu(b)q(et)o(w)o(een)g(the)f(in)o(v)o(o)q(cation)h(of)f
Fo(MPI)p 1648 1658 V 16 w(IRECV)h Fu(and)189 1715 y(the)j(\014nish)h(of)f
Fo(MPI)p 528 1715 V 16 w(W)l(AIT)p Fu(.)g(But)g(the)g(compiler)i(cannot)e
(see)g(an)o(y)g(p)q(ossibilit)o(y)i(that)e Fo(buf)h Fu(can)f(b)q(e)189
1771 y(c)o(hanged)g(after)e Fo(MPI)p 557 1771 V 16 w(IRECV)i
Fu(has)g(\014nished.)23 b(And)16 b(therefore)f(it)h(can)g(sc)o(hedule)h(the)f
(load)g(of)f Fo(buf)189 1828 y Fu(earlier)f(than)g(t)o(yp)q(ed)g(in)h(the)f
(source.)20 b(It)13 b(has)h(no)g(reason)f(to)h(a)o(v)o(oid)f(using)i(a)e
(register)h(to)f(hold)i Fo(buf)189 1884 y Fu(across)f(the)h(call)i(to)d
Fo(MPI)p 625 1884 V 16 w(W)l(AIT)p Fu(.)h(It)g(also)g(can)h(reorder)f(the)g
(instructions)h(as)f(in)h(the)f(left)h(case.)189 1959 y(T)l(o)c(prev)o(en)o
(t)h(instruction)h(reordering)g(or)e(the)h(allo)q(cation)h(of)f(a)g(v)m
(ariable)h(or)f(bu\013er)g(in)h(a)e(register,)189 2015 y(there)j(are)g(three)
g(p)q(ossibilities)k(in)d(p)q(ortable)f(F)l(ortran)f(co)q(de:)243
2108 y Fm(\017)23 b Fu(One)d(can)f(put)h(the)g(v)m(ariables)g(and)g
(bu\013ers)f(in)o(to)h(common)f(blo)q(c)o(ks;)j(then)e(they)f(will)i(b)q(e)
289 2164 y(allo)q(cated)d(in)g(the)g(memory)e(while)j(an)o(y)e(library)h
(subroutine)h(\(e.g.)25 b(an)18 b(MPI)f(routine\))g(is)289
2221 y(called)c(pro)o(vided)g(that)e(the)h(compiler)h(cannot)e(analyse)h
(that)f(the)h(library)h(subroutine)g(do)q(es)289 2277 y(not)h(use)i(these)f
(common)g(blo)q(c)o(ks,)h(or)243 2349 y Fm(\017)23 b Fu(one)11
b(can)h(declare)g(them)f(as)g(V)o(OLA)l(TILE,)h(but)g(this)g(is)f(not)g(part)
g(of)g(the)g(standards)g(F)l(ortran)289 2406 y(77)f(and)h(90,)g(and)g(y)o
(our)f(co)q(de)i(will)h(not)d(b)q(e)i(p)q(ortable)f(b)q(ecause)h(there)f(are)
g(compilers)h(without)289 2462 y(V)o(OLA)l(TILE,)k(and)f(normally)h(it)g
(prev)o(en)o(ts)e(an)o(y)h(optimization,)h(or)243 2534 y Fm(\017)23
b Fu(one)14 b(can)g(call)h(a)f(dumm)o(y)g(routine)g(with)g(the)h(v)m(ariable)
g(or)e(bu\013er)h(as)g(an)g(actual)g(argumen)o(t;)289 2591
y(if)22 b(the)h(dumm)o(y)f(routine)h(is)f(not)g(part)g(of)g(the)g
(application)i(then)f(the)f(compiler)i(m)o(ust)289 2647 y(allo)q(cate)16
b(the)f(v)m(ariable)i(or)e(bu\013er)h(in)g(the)f(memory)g(while)i(that)e
(subroutine)h(is)g(executed;)289 2704 y Fo(MPI)p 374 2704 V
15 w(ADDRESS)h Fu(can)e(b)q(e)h(used)g(as)e(dumm)o(y)h(routine.)1967
46 y Fj(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(1.6.)34 b(A)15 b(F)o(OR)l(TRAN)h(PR)o(OBLEM)g(WITH)f
(REGISTER)i(OPTIMIZA)l(TION)361 b Fu(25)189 49 y(High)13 b(qualit)o(y)g(F)l
(ortran)e(implemen)o(tations)j(ma)o(y)e(pro)o(vide)h(a)f(general)h(solution)h
(to)e(this)h(problem.)189 106 y(Ho)o(w)o(ev)o(er,)h(co)q(de)j(that)e(do)q(es)
i(not)e(use)i(one)f(of)g(the)g(three)g(solutions)g(describ)q(ed)i(ab)q(o)o(v)
o(e)e(ma)o(y)f(not)189 162 y(b)q(e)h(p)q(ortable.)189 237 y(In)h(C,)f
(subroutines)i(whic)o(h)g(mo)q(dify)f(v)m(ariables)h(that)e(are)h(not)f(in)i
(the)f(parameter)f(list)i(will)g(not)189 294 y(cause)e(register)h
(optimization)g(problems.)25 b(This)17 b(is)g(b)q(ecause)g(taking)g(p)q(oin)o
(ters)g(to)f(storage)f(ob-)189 350 y(jects)g(b)o(y)g(using)h(the)f(&)h(op)q
(erator)e(and)h(later)h(referencing)g(the)f(ob)s(jects)g(b)o(y)g(w)o(a)o(y)f
(of)h(the)g(p)q(oin)o(ter)189 407 y(is)i(an)g(in)o(tegral)g(part)f(of)g(the)h
(language.)25 b(A)17 b(C)g(compiler)h(understands)f(the)g(implications,)i
(but)189 463 y(with)c(aggressiv)o(e)g(optimization)h(compilers)g(can)g(p)q
(ossibly)h(generate)e(incorrect)g(co)q(de.)189 538 y(\()p Fp(End)g(of)i
(advic)n(e)f(to)g(users.)p Fu(\))p 75 645 827 2 v 75 701 a(The)f(follo)o
(wing)h(advice)h(m)o(ust)d(b)q(e)i(added)g(in)75 769 y(-)f(MPI)g(2,)g(c)o
(hapter)g(?.?)21 b(Handlers,)75 826 y(-)15 b Fo(MPI)p 190 826
14 2 v 16 w(ADDRESS)h Fu(\(MPI)f(1.1,)f(page)h(70,)f(line)j(16\),)d(and)h(in)
75 882 y(-)g Fo(MPI)p 190 882 V 16 w(IRECV)h Fu(\(MPI)f(1.1,)e(\\Non)o(blo)q
(c)o(king)k(comm)o(unication",)e(end)h(of)e(page)h(40\))189
988 y Fp(A)n(dvic)n(e)i(to)i(users.)53 b Fu(T)l(o)17 b(prev)o(en)o(t)g
(troubles)h(with)g(the)f(register)g(optimization)i(of)e(the)g(F)l(ortran)189
1045 y(compilers)i(please)g(pa)o(y)e(atten)o(tion)h(to)f(the)h(hin)o(ts)g(in)
h(section)g(1.6)e(\\A)g(F)l(ortran)g(Problem)i(with)189 1101
y(register)c(optimization")h(on)f(page)g(23.)k(\()p Fp(End)d(of)g(advic)n(e)g
(to)h(users.)p Fu(\))-32 46 y Fj(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: Marc Snir: "Re: new version of chapter 4"
• Next message: David C. DiNucci: "Re: new version of chapter 4"
• Previous message: Marc Snir: "new version of chapter 4"