1/%*2.&34.&%56+74.&%*8"(%&,.9%*!"#$%&'()*)+,'-.&.' & /* '012*)314'-5'"16*1+713'

Transcription

1 1/%*2.&34.&%56+74.&%*8"(%&,.9%*!"#$%&'()*)+,'-.&.' & /* '012*)314'-5'"16*1+713' 8"'(&:9(+&;'' 89:';13<=,/' 1%.9/#"0*<''#'(."(';* A* F$%&$#%4*! =+:&'%*(/%C%';*G#0*."3*H#IH%*! J+:&*#CK+&(."(*&%.H#L%'*! 2+4*(/%*9+:&'%*M('*#"(+*(/%*=6>*9:&&#9:H:C*! N+0#'L9'* D*

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

3 N#IH%*1/%C%*D;*1&."'H.L+"*! 1/%&%*#'*.*G#0*0.K*G%(4%%"*/+4*4%*(/#"[*.G+:(*K&+0&.C'*."3* 3.(.*."3*(/%*E'*."3*A'*+,*9+CK:(%&'*! R%%3*H."0:.0%'*(+*3%'9&#G%*4/.(*4%*C%."*! N."0:.0%'*"%%3*(+*G%*(&."'H.(%3*+"%*'(%K*.(*.*LC%*! DF3EU7CUGF3E'! V39++93'! -%*["+4*W.$.*.'*.*K&+0&.CC#"0*H."0:.0%*! N#IH%*1/%C%*?;*=+"(&+H*JH+4*! 2+4*3+*9+CK:(%&'*+&9/%'(&.(%*(/%*C.")*(/#"0'*(/%)*.&%* 3+#"0*^*'%%C#"0H)*#"*K.&.HH%H*! -/.(*3+*4%*/.$%*(+*[%%K*(&.9[*+,*4/%"*4%*9.HH*.*C%(/+3T*."3*(/%"*."+(/%&T*."3*(/%"*."+(/%&T*."3*'+*+"*! 2+4*3+*4%*["+4*4/.(*(+*3+*:K+"*_&%(:&"`*!!'%&*K&+0&.C'*."3*+K%&.L"0*')'(%C'*! `):J6:1')/13'63FK39+/'! df:),*93c'f3'7c'mf321i' a*

4 =+:&'%*F:(9+C%'*! J+:"3.L+";*G.'#9'*+,*/#0/BH%$%H*K&+0&.CC#"0*SW.$.V*!!"3%&'(."3#"0*+,*'+C%*+,*(/%*.G'(&.9L+"'*(/.(*%U#'(* G%(4%%"*K&+0&.C'*."3*(/%*/.&34.&%*(/%)*&:"*+"T*4/)*(/%)* %U#'(T*."3*/+4*(/%)*G:#H3*:K+"*%.9/*+(/%&*! b"+4h%30%*+,*'+c%*+,*(/%*3%(.#h'*+,*:"3%&h)#"0* #CKH%C%"(.L+"'*! O%9+C%*C+&%*%c%9L$%*K&+0&.CC%&'*! `F31'1e2=1,*'9*'N,E=,K'9,E'1:=+=,9J,K'7)K/'! 63FK39+/'9,E'E9*9'! d&%k.&%*,+&*h.(%&*9h.''%'*#"*=6>* e* P%.H#()*A;*8"('*f*8"(%0%&'*g*JH+.('*f*P%.H'*! P%K&%'%"(.L+"'*.&%*M"#(%*! >U.CKH%*A;*8'*U D *h*ex*! >:F9*/f'c1/X'!?,*/f'! 'g...'h'g...''uui'&^...'! '%...'h'%...''uui'ii'! >U.CKH%*D;*8'*SU*i*)V*i*j**k**U*i*S)*i*jVX*! R,/=K,1E'j'"=K,1E'?,*/f'c1/X'! >:F9*/f ''! '[&1-.'k'U&1-._'k'$l&g'UUi'$l&g'! '&1-.'k'[U&1-.'k'$l&g_'UUi'II' l*

5 =+3%*6%9:&#()*>U.CKH%* /* Kernel memory region holding user-accessible data */ #define KSIZE 1024 char kbuf[ksize]; int len = KSIZE; /* Copy at most maxlen bytes from kernel region to user buffer */ int copy_from_kernel(void *user_dest, int maxlen) { /* Byte count len is minimum of buffer size and maxlen */ if (KSIZE > maxlen) len = maxlen; memcpy(user_dest, kbuf, len); return len;! 6#C#H.&*(+*9+3%*,+:"3*#"*J&%%O6QY'*#CKH%C%"(.L+"*+,* 0%(K%%&".C%*! 1/%&%*.&%*H%0#+"'*+,*'C.&(*K%+KH%*(&)#"0*(+*M"3*$:H"%&.G#H#L%'* #"*K&+0&.C'* m* 1)K#9.H*!'.0%* /* Kernel memory region holding user-accessible data */ #define KSIZE 1024 char kbuf[ksize]; int len = KSIZE; /* Copy at most maxlen bytes from kernel region to user buffer */ int copy_from_kernel(void *user_dest, int maxlen) { /* Byte count len is minimum of buffer size and maxlen */ if (KSIZE > maxlen) len = maxlen; memcpy(user_dest, kbuf, len); return len; #define MSIZE 528 void getstuff() { char mybuf[msize]; copy_from_kernel(mybuf, MSIZE); printf( %s\n, mybuf); AE*

6 n.h#9#+:'*!'.0%* /* Kernel memory region holding user-accessible data */ #define KSIZE 1024 char kbuf[ksize]; int len = KSIZE; /* Copy at most maxlen bytes from kernel region to user buffer */ int copy_from_kernel(void *user_dest, int maxlen) { /* Byte count len is minimum of buffer size and maxlen */ if (KSIZE > maxlen) len = maxlen; memcpy(user_dest, kbuf, len); return len; #define MSIZE 528 void getstuff() { char mybuf[msize]; copy_from_kernel(mybuf, -MSIZE);... AA* P%.H#()*oD;*Z+:Y$%*p+(*(+*b"+4*<''%CGH)*! =/."9%'*.&%T*)+:YHH*"%$%&*4&#(%*.*K&+0&.C*#"*.''%CGH)*9+3%*!!F+6=:13/'931'+)2@'71B13'9,E'+F31'69J1,*'*@9,'CF)'931'! O:(;*!"3%&'(."3#"0*.''%CGH)*#'*(/%*[%)*(+*(/%*C.9/#"%BH%$%H* %U%9:L+"*C+3%H*! A1@9H=F3'FM'63FK39+/'=,'631/1,21'FM'7)K/'! 8=K@U:1H1:':9,K)9K1'+FE1:'7319</'EFG,'! W),=,K'63FK39+'613MF3+9,21'! R,E13/*9,E'F6J+=m9JF,/'EF,1b,F*'EF,1'7C'*@1'2F+6=:13'! R,E13/*9,E=,K'/F)321/'FM'63FK39+'=,1e2=1,2C'!?+6:1+1,J,K'/C/*1+'/FnG931'! a6139j,k'/c/*1+/'+)/*'+9,9k1'63f21//'/*9*1'!!319j,k'b'nk@j,k'+9:g931'! L]^'9//1+7:C'=/'*@1':9,K)9K1'FM'2@F=21' AD*

7 <''%CGH)*=+3%*>U.CKH%*! 1#C%*6(.CK*=+:"(%&*! <KKH#9.L+"*! `19/)31'J+1'[=,'2:F2<'2C2:1/_'31o)=31E'7C'63F21E)31' double t; start_counter(); P(); t = get_counter(); printf("p required %f clock cycles\n", t); A?* =+3%*(+*P%.3*=+:"(%&*! -&#(%*'C.HH*.C+:"(*+,*.''%CGH)*9+3%*:'#"0*p==Y'*.'C*,.9#H#()*! 8"'%&('*.''%CGH)*9+3%*#"(+*C.9/#"%*9+3%*0%"%&.(%3*G)* 9+CK#H%&* /* Set *hi and *lo to the high and low order bits of the cycle counter. */ void access_counter(unsigned *hi, unsigned *lo) { asm("rdtsc; movl %%edx,%0; movl %%eax,%1" : "=r" (*hi), "=r" (*lo) /* output */ : /* input */ : "%edx", "%eax"); /* clobbered */ A]*

8 P%.H#()*o?;*n%C+&)*n.I%&'*! n%c+&)*#'*"+(*:"g+:"3%3*!?*'+)/*'71'9::f29*1e'9,e'+9,9k1e'! `9,C'966:=29JF,/'931'+1+F3CUEF+=,9*1E'! n%c+&)*&%,%&%"9#"0*g:0'*.&%*%'k%9#.hh)*k%&"#9#+:'*! 613MF3+9,21'! +9SF3'/611E'=+63FH1+1,*/' n%c+&)*p%,%&%"9#"0*o:0*>u.ckh%* double fun(int i) { volatile double d[1] = {3.14; volatile long int a[2]; a[i] = ; /* Possibly out of bounds */ return d[0]; fun(0) > 3.14 fun(1) > 3.14 fun(2) > fun(3) > fun(4) > 3.14, then segmentation fault Aa*

9 n%c+&)*p%,%&%"9#"0*o:0*>u.ckh%* double fun(int i) { volatile double d[1] = {3.14; volatile long int a[2]; a[i] = ; /* Possibly out of bounds */ return d[0]; fun(0) > 3.14 fun(1) > 3.14 fun(2) > fun(3) > fun(4) > 3.14, then segmentation fault >UKH.".L+";* Saved State 4 d7 d4 3 d3 d0 2 a[1] 1 a[0] 0 N+9.L+"*.99%''%3*G)* fun(i) Ae* n%c+&)*p%,%&%"9#"0*>&&+&'*! =*S."3*=iiV*3+*"+(*K&+$#3%*.")*C%C+&)*K&+(%9L+"*! a)*'fm'7f),e/'9339c'31m131,21/'!?,h9:=e'6f=,*13'h9:)1/'! (7)/1/'FM'+9::F2bM311'! =."*H%.3*(+*".'()*G:0'*! D@1*@13'F3',F*'7)K'@9/'9,C'1p12*'E161,E/'F,'/C/*1+'9,E'2F+6=:13'! (2JF,'9*'9'E=/*9,21'!!F33)6*1E'F7S12*':FK=29::C'),31:9*1E'*F'F,1'71=,K'9221//1E'! #p12*'fm'7)k'+9c'71'n3/*'f7/13h1e':f,k'9n13'=*'=/'k1,139*1e'! 2+4*9."*8*3%.H*4#(/*(/#'X*! ;3FK39+'=,'Z9H9'[F3'!q4'F3'`04'F3'r_'! R,E13/*9,E'G@9*'6F//=7:1'=,*1392JF,/'+9C'F22)3'! R/1'F3'E1H1:F6'*FF:/'*F'E1*12*'31M131,2=,K'133F3/' Al*

10 n%c+&)*6)'(%c*d%&,+&c."9%*>u.ckh%!! 2#%&.&9/#9.H*C%C+&)*+&0."#j.L+"*! void copyij(int src[2048][2048], int dst[2048][2048]) { int i,j; for (i = 0; i < 2048; i++) for (j = 0; j < 2048; j++) dst[i][j] = src[i][j]; void copyji(int src[2048][2048], int dst[2048][2048]) { int i,j; for (j = 0; j < 2048; j++) for (i = 0; i < 2048; i++) dst[i][j] = src[i][j]; DA*LC%'*'H+4%&* Sd%"L:C*]V* Am* 1/%*n%C+&)*n+:"(.#"* P%.3*(/&+:0/K:(*SnO5'V* L1 ;1,J)+'???'\1F,* %%.'`8m* &^'sa'f,u2@=6'0&'eu292@1* &^'sa'f,u2@=6'0&'=u292@1* %&-'sa''fpu2@=6'),=n1e* 0-'292@1* L s1 s3 s5 s7 6(&#3%*S4+&3'V* s9 s11 Mem s13 s15 8m 2m 512k 128k 32k 8k 2k -+&[#"0*'%(*'#j%*SG)(%'V* DE*

11 P%.H#()*o];*d%&,+&C."9%*#'"Y(*9+:"L"0*+K'*! >U.9(*+K*9+:"(*3+%'*"+(*K&%3#9(*K%&,+&C."9%*! 63F21E)31/4'9,E':FF6/'! n:'(*:"3%&'(."3*')'(%c*(+*+klc#j%*k%&,+&c."9%*! 8FG'63FK39+/'2F+6=:1E'9,E'1L12)*1E'! 8FG'+1+F3C'/C/*1+'=/'F3K9,=m1E'! 8FG'*F'+19/)31'63FK39+'613MF3+9,21'9,E'=E1,JMC'7FB:1,12</'! K1,139:=*C' DA* >U.CKH%*n.(&#U*n:HLKH#9.L+"*! "*9,E93E'E1/<*F6'2F+6)*134'H1,EF3'2F+6=:134')/=,K'F6J+=m9JF,'Q9K/'! $ _' Best code (K. Goto) Triple loop 160x DD*

12 nnn*dh+(;*<".h)'#'* Multiple threads: 4x Vector instructions: 4x Memory hierarchy and other optimizations: 20x! T19/F,'MF3'-.Lf'7:F2<=,K'F3'J:=,K4':FF6'),3F::=,K4'9339C'/29:93=m9JF,4' D?* d&%b&%q:#'#(%'*! &g-'9,e'&g$f'?,*3f';3fk39++=,k'?'9,e'??'! F"%*+,*a*9+&%*9+:&'%'*! $&&f'>f),e9jf,/'?'! $&-f'>f),e9jf,/'??'! $$&f'"d't1/=k,'9,e'?+6:1+1,*9jf,'! $$-f't9*9'(7/*392jf,/'! $%&f'8db"d'?,*13m921'! D]*

13 =6>]ee5]lA* =.K'(+"%'* 2-*Q%'#0"* =6>???* 6)'(%C'*d&+0* FK*6)'(%C'* =6>]EA* =+CK#H%&'* =6>]aA* R%(4+&['* =6>]l]* 6%9:&#()* =6>]aa* >CG*6)'(%C'* d%&,+&c."9%* =+CK\*<&9/\* =+"9:&&%"9)* n.9/#"%* =+3%* Q#'(&#G:(%3* 6)'(%C'* >U%9:L+"** n+3%h* P%.HB1#C%* =+"(&+H* "#$!%&'(&!)*+$,-./$! 89:#*(;,9+'-*,9$,-(#)'(,9<,9+' 43*:=3*#'39:')>?=3*#' =6*A]?* 8"(&+*d&+0*88* =+:&'%*d%&'k%9l$%*! t1/=k,'6=61:=,1e'63f21//f3'=,'d13=:fk'! a6139j,k'"c/*1+/'!?+6:1+1,*':93k1'6f3jf,/'fm'f6139j,k'/c/*1+'!!f+6=:13/'! D3=*1'2F+6=:13'MF3'/=+6:1':9,K)9K1'! O1*GF3<=,K'!?+6:1+1,*'9,E'/=+):9*1',1*GF3<'63F*F2F:/' Da*

14 =+:&'%*d%&'k%9l$%*s=+"(\v*! 1/#'*9+:&'%*#'*d&+0&.CC%&B=%"(&#9*! F,1'29,'71'+F31'1p12JH1'9/'9'63FK39++13'! A1B13'E17)KK=,K'! A1B13'79/=/'MF3'1H9:)9J,K'613MF3+9,21'! 8FG'+):J6:1'92JH=J1/'GF3<'=,'2F,213*'[1lKl4'a"'9,E')/13'63FK39+/_'! De* 1%U(G++['*! =+CK:(%&*6)'(%C';*<*d&+0&.CC%&Y'*d%&'K%9L$%T*D "3 *>3#L+"*! T9,E9:'#l'A3C9,*'9,E't9H=E'Tl'aY89::93F,''! 8FG'*F'/F:H1':97/'! ;392J21'63F7:1+/'*C6=29:'FM'1L9+'63F7:1+/'! (/ *>3#L+"*! "9+)1:';l'8937=/F,'???'9,E'V)C'0l'"*11:14'Z3l'! ;31,J21U89::4'-..-'! "F:=E'!'63FK39++=,K':9,K)9K1'31M131,21'! Dl*

15 =+:&'%*=+CK+"%"('*! N%9(:&%'*Sr?EV*! 6%9L+"'*SrAEV*! (66:=1E'2F,216*/4'=+6F3*9,*'*FF:/'9,E'/<=::/'MF3':97/4'2:93=N29JF,'FM' :12*)31/4'1L9+'31H=1G'9,E' JF,'! -&#I%"*.''#0"C%"('*S]V*! ;3F7:1+/'M3F+'*1L*'*F'/F:=E=MC'),E13/*9,E=,K'! N.G'*S]V*! >U.C'*SC#3(%&C*i*M".HV*! W1/*'CF)3'),E13/*9,E=,K'FM'2F,216*/'9,E'63=,2=6:1/' Dm* P%'+:&9%' **! =+:&'%*-%G*d.0%*! =+:&'%*Q#'9:''#+"*O+.&3*! "*9p'G=::'+F,=*F3'9,E'2F,*3=7)*1'! =+:&'%*n.#h#"0*n#'(*! 0FG'*39e2'P'+F/*:C'9,,F),21+1,*/u'CF)'931'9:319EC'/)7/23=71E'! 6(.c*%C.#H*! <"+")C+:'*J%%3G.9[*S4#HH*G%*H#"[%3*,&+C*/+C%K.0%V*!

16 d+h#9#%';*p&.3#"0*! >U.C';*4%#0/(%3*A5?*SC#3(%&CVT*D5?*SM".HV*! -&#I%"*.''#0"C%"(';*4%#0/(%3*.99+&3#"0*(+*%c+&(*! N.G'*.''#0"C%"(';*4%#0/(%3*.99+&3#"0*(+*%c+&(*! p&.3#"0;*! -%w'g3=b1,'9//=k,+1,*/'! $%w':97'9//=k,+1,*/'! g.w'1l9+/'?a* N%(Y'*/.$%*,:"*! N%(Y'*H%.&"*^*(+0%(/%&*! N%(Y'*9+CC:"#9.(%*! N%(Y'*'%(*(/%*G.&*,+&*.*:'%,:H*."3*#"(%&%'L"0*9H.''*! n.")*(/."['*(+*(/%*c.")*#"'(&:9(+&'*4/+*/.$%*'/.&%3*(/%#&* H%9(:&%*"+(%'*^*8*4#HH*G%*G+&&+4#"0*H#G%&.HH)*(/&+:0/*(/%*q(&*^* (/%)*3%'%&$%*.HH*(/%*9&%3#(T*(/%*%&&+&'*.&%*.HH*C#"%*!