From 3d217e784060de58d1dc570cca767c59f6a87867 Mon Sep 17 00:00:00 2001 From: Nguyen Quy Anh <23906-nguyenquy@users.noreply.gitlab.inf.ethz.ch> Date: Fri, 24 Apr 2026 16:31:42 +0200 Subject: [PATCH] Added measurements --- board.hpp | 15 +++++++-- forces.cpp | 22 +++++++++++- main | Bin 0 -> 31912 bytes main.cpp | 30 +++++++++++++++++ questions.md | 6 ++++ tsc_x86.h | 93 +++++++++++++++++++++++---------------------------- vector2d.hpp | 2 ++ 7 files changed, 113 insertions(+), 55 deletions(-) create mode 100755 main create mode 100644 questions.md diff --git a/board.hpp b/board.hpp index e76b5ea..33da163 100644 --- a/board.hpp +++ b/board.hpp @@ -7,16 +7,27 @@ struct Obstacle { Obstacle() : start(0, 0), end(0, 0) {} Obstacle(Vector2d s, Vector2d e) : start(s), end(e) {} - // find the point of the obstacle that is nearest to p + + // Find the point of the obstacle that is nearest to p + // # 15 flops Vector2d find_nearest_obstacle_point(Vector2d p) { + + // # 2 ADDs Vector2d obstacle_v = end - start; + + // # 9 flops = 2 ADDs, 2 MULTs, 1 ADD, 1 DIV, 2 MULTs, 1 ADD double projection = (p - start).dot(obstacle_v) / obstacle_v.dot(obstacle_v); + if (projection < 0.0) projection = 0.0; - if (projection > 1.0) + else if (projection > 1.0) projection = 1.0; + + // # 4 flops = 2 ADDs, 2 MULTs return start + obstacle_v * projection; + } + }; struct Board { diff --git a/forces.cpp b/forces.cpp index 24dfa5f..5911909 100644 --- a/forces.cpp +++ b/forces.cpp @@ -1,6 +1,7 @@ #include "forces.hpp" #include +// # 8 flops = 5 MULTs, 2 ADD, 1 SQRT double w(Vector2d e, Vector2d f) { if (e.dot(f) >= f.norm() * cos_phi) return 1.0; @@ -8,6 +9,7 @@ double w(Vector2d e, Vector2d f) { return c; } +// # 2 DIVs double g(double v_max_alpha, double w_alpha_norm) { if (w_alpha_norm <= v_max_alpha) { return 1; @@ -15,12 +17,16 @@ double g(double v_max_alpha, double w_alpha_norm) { return v_max_alpha / w_alpha_norm; } +// # 8 flops void compute_acceleration_term(Pedestrian *p) { + // # 2 flops Vector2d desired_velocity = p->e * p->desired_v0; Vector2d actual_velocity = p->curr_vel; + // # 6 flops = 4 ADDs, 2 DIVs p->f_i += (desired_velocity - actual_velocity) / tau_alpha; } +// # 55(N - 1) = 55N - 55 flops void compute_repulsive_effects_from_pedestrians(Pedestrian *pedestrians, Pedestrian *p, int N) { for (int i = 0; i < N; i++) { @@ -34,28 +40,39 @@ void compute_repulsive_effects_from_pedestrians(Pedestrian *pedestrians, Pedestr // continue; // } + // # 2 ADDs Vector2d r_ab = p->curr_pos - pedestrians[i].curr_pos; + // # 4 flops = 1 SQRT, 2 MULTs, 1 ADD double r_ab_norm = r_ab.norm(); Vector2d e_b = pedestrians[i].e; + // # 2 MULTs double s_b = pedestrians[i].desired_v0 * delta_t; + // # 4 flops = 2 ADDs, 2 MULTs Vector2d r_ab2 = r_ab - (e_b * s_b); + // # 4 flops = 1 SQRT, 2 MULTs, 1 ADD double r_ab2_norm = r_ab2.norm(); - double b = sqrt(pow(r_ab_norm + r_ab2_norm, 2) - pow(s_b, 2)) / 2; + // # 7 flops = 1 SQRT, 3 ADDs, 2 MULTs, 1 DIV + double b = sqrt((r_ab_norm + r_ab2_norm) * (r_ab_norm + r_ab2_norm) - s_b * s_b) / 2; + // # 12 flops = 6 DIVs, 3 ADDs, 3 MULTs Vector2d b_grad = (r_ab / r_ab_norm + r_ab2 / r_ab2_norm) * (r_ab_norm + r_ab2_norm) / (4 * b); + // # 6 flops = 3 MULTs, 2 DIVs, 1 EXP Vector2d f_ped = b_grad * ((V0_AB / SIGMA) * (exp(-b / SIGMA))); + // # 14 flops: 9 MULTs, 4 ADDs, 1 SQRT p->f_i += f_ped * w(p->e, f_ped * -1); } } +// # Assume that the four borders are also 4 obstacles +// # 36 n_obstacles flops = 144 flops void compute_repulsive_effects_from_borders(Board &board, Pedestrian *p) { Obstacle *obstacles = board.obstacles; @@ -65,10 +82,13 @@ void compute_repulsive_effects_from_borders(Board &board, Pedestrian *p) { Obstacle obstacle = obstacles[i]; + // # 17 flops Vector2d r_aB = p->curr_pos - obstacle.find_nearest_obstacle_point(p->curr_pos); + // # 17 flops = 2 DIVs, 1 SQRT, 2 MULTs, 1 ADD, 2 MULTs, 1 DIV, 2 MULTs, 1 EXP, 1 SQRT, 2 MULTs, 1 ADD, 1 DIV Vector2d f_obst = (r_aB / r_aB.norm()) * (U0_AB / R) * std::exp(-r_aB.norm() / R); + // # 2 ADDs p->f_i += f_obst; } } diff --git a/main b/main new file mode 100755 index 0000000000000000000000000000000000000000..6df50769a8b160144d31a3d4aa10357823305fe1 GIT binary patch literal 31912 zcmeHw4_uU0y8k=C;F!V;t`=n$)lk^3$N(C&c>}|Ir^rw&p(RIN7=?|0%?ze#D+ca5 zsjaoP)m`_lcdz@qwzte$yK7y$f%5tnUBDEPNJKHnKZurqg!B78=RIc}M%cgi^SPhj z@8i*V&vTyhJkR<6oadbL@}6bWQXdV_=_KI-rGJtrR1&6Pl6b@>2M6FKNeNOYo>9_$ zQZT|n45!5>X&z!VX9^tX8#tcO#OLDi{{HQpPWU2+N zp8c1HxBe#mRJUQ%FJ?6Mk64x)GqE6FnH)c{Aa_DReo^V735yaYPnaAZ?JSOtl}PNZ z_>nv<-^&4MIP&Z42YHX6(Dm-OK#jT{Ej>B}e$x>68f0KFIXi~He=`KWZV3D*L!?XH zX|Qxh3?XML@LW}a^P0;Xf<6xJZ4wua)*{rjlu;tnv_67M)m)$Y@iAM^G zi|n(rl>)mKH+al5>8?bO7v>jbyX*@k+cUWZX$uSC&qyyWa8Ao-kjoP1$S%q)E<^$g z^K0?5>5`1=ENHq=d;Ya@rj(YwomZv(k%aBS+5^WkZrJXHm9ILAHu)IoYn9 zyg?;C0RH*;Mfpr#x+~U_U+lCg*-pC+L(`G%SPbd;E+l4KuxOF3#O`nw7iAaZyB6CP z#z0`Y%QD$!D|WgZ_UuBDksM$xIeFO*o6C`%?{ZpmC}$`O=@g$^vSlg z$J66%1(24Tou(Dbzuc5d%Ez=>kIzVVJwQcUWU*LcHGxPyzi5GVkYv-dDeY-2qwzV# zr7p?7s6;BVFLGtiN99We#S04PA=#cPdLh3k*EY-UEG z;Z0tconORj7d3~!6{dF;>L*O}lsT2uTW& z9tDn^4m@TaAA_*_wO>M=50hqaT>|Qn=&`r|YJiNH>edK?*^zS^J$LI~x zRvtd|{&O&&0;Qci>>B?%Oo0IDAj0H?eIx9HG_GXd;lj`b7Vqph`X$t6wVc{QBm{eFt z1Pdz^Ce_tpV&P{f97-bC&tL zRVn80LXyfb^QV}5jGJehdv;R&SOoXU! z=0_XvtYET;9&x?E%4_r&$xG~UjX`tWOfZ^jxTnQzSx?OVJRo7?;Xv6jb2{3qC<{R%!xyDcVFPpLgyi)1+pOVX`ybel9cDK1k z%B!Z#1-^YaYN0(Bk6rp_5!OA6w7uyQf5!`mOk#{-m_4z_RX!@NmO=Z&EC8A(X|kqA6~6i(s!^vrrRLcNr+x4O!i`E-nW zV)Rk9adf?UBCDP#=0>S!Ozp_Ud&ks{(I?f$%v05^b(_wry))a?#$9K1YRBBq)k7KI zsC{8Bvq9aTc^Sl?cB%Vk9)zq)wRf(nc4WXGpJFy4ohlFpnNe~2#Sftoa{E0VWG8Yq z0%~8TzJUCuxg*U8q(ENZs7n0ys$H3>VDHTYd*)HKch?>rM5M88i2Iv>yIXhgS%l8E?G^4w$_RjzM)JVh0o z{8aMnbb0yNo*;Cc99y`&qRIGb7{(Y{t0)|gr!!JlEiT8l2Tti%mhp=$Sgqq;_TPQR}j5)sDSN$r?ZkM4KOaWkId&OAujk29zEHvV}^| z8;Z5f-ap$m{ZXd}v`;A7r#)AS2eI^$o}+YYHG<}{%khm0?0VtfHQg0dVl{`7Gciu$r+Dh)fg z%Pg1onVbfUgN|n~D^krxg(t&kmCD6_`PQ~=L1Uri++QRwKXt+5SvJjb zPTe~PI%ZM#8CKgH%`q6AMEj%2TToEBs%&;t4{J2wKWJoY!*s8sqm7FF9G}rXE@*4Vm?!*8k4F1T8ESNP> z$NYkcfy(o2STMN+uGdl{5CsqkSD3e8}ToiX0XcOZbsRlIFP4r+4B&?9A|lIVrlg$=@43ZN=biDzUSa6;)AAS z%?&UL%*fVZle5~Dx;8nhQR!MEXLTucc=jmI zzd^a1i6`&}+I#SL)Bn+b$~4H_MryEA?&iwm#+g!*XV`;sCy0aRAe#)E$uL zUR3H-c`j-Ny>#3glb{Ha$In3rdr#i6D?sV`oxJ0Sj)nfDg^ugg{^VBm+vGFqxv8~k z+f-_(eNmV3+#BUVMW;qxQZ62lH=h9fL{y(rugaU-m5b-(%@-&mn|mqy8JN)UY*V`4 zkTZ_3=W(U$TRF3enmuf_xrFr~v+z$&b)<27+NzE1W2lqi9=EhjYKM#;!rNVB;Q&a$RgL7(`d5j9gnFV8tEtB zk}7}&5K(<{u|CMGYcF7!s8n-=+C^3Z#nCips$iTk5Kc!=nGAdUpUlF+r^ zuStNV00Gq}UkZl=RB_lqL5R4uQg=Do7I zJYYQonp#idlzNXLUj~z(vu#r4=cV6@tY6Scvu3NrsSmeVtqr;8Q12ngpLZ zxmlC&XX=Cm`PWIyCI31(NSMLugxD$b?`bT&PPF#dwZ920>pT-y#8T^{c}*-l=648F zwP5_#LWFg>N_&w?*I{|?M&7*uzRS84RVnLjt65i|rX5(j=z`mMKL?0@$a{F^4oV^u zotyXb%nIJmo8@iYEcxx4E~N(2a>6sEDch@ZF;}M-!GZC{)bkcGYB=-7!}7LEERJht zLbMPf3}s=&5Txny)Wmbf6)nA}7DQ)Yj2@6Pj(O)|c@?HX>NaDmSnnBojP;$2Dr%uB zOoFhMnHAPYtZj0}K8o0OO}lQjh8MJ4~l)$Cy2s7mrn+?AINB>WT4fsk(lQAxwrO zrXRKpDGIEM0COwu>ar|YaQ&ufWK%qqgvS#Rcsx4~kFA`~w4L!)D=*!ue=&yjxZqaw z^U~lmc)k?8!$R2)&ox+Lm#rDNpf)wD?Z zQF2!Dv$ULg7B#e!x?#u7& z+Mhk|@mSrvrmgT8E1bx?^|?Rg$=lQ!XRGPIyk_QG)vfz4Ur1$}0(y6McSF+U-NRfD zd3I4p2fF<9umS72-E0?|4bn2&bkU$FKyrCyp6oeT+AJ@>a+Esz%K=C*)qPRc?IU5@ zdSht*-AQcMNE2A%6-{lqyQ~&ZNaM*J^+G)?HUUxI*d#3TphC*#0kgYogKt#XCWN%u zJuKFMd;;J&8lqaP3DEAck^!-cys>YwSSsT+-WCy?G*Z?l!FrMtYmF;HsWsMYH}63j zsqXFOZIs;#7RN;^Jl5krSatpmjbsZ* zEC=%sQjT?1C{Hz2CA!=&dCEdmy>Z1osA|lV`X-!kuSzxF{^{{8hPSC!6HtTHlI}fK z?RV<-)b%hmLZzYD4e-giD^cr|5!6tMIi|cfz_{9qRcWLCJhtMR9x}I2J>&5-Ej61O z^(`C|Lzn`y>C3K0eJya7y^VXDIvUS6?Pxys<%#CiW~xx*FRHMG@vt1H1?4r#@v?h| zJe-;#!ZplNFL!sH39{68cVYgR@CL-lLvXvZsIWYuQqVc5XE;Z$($^q@LogldC#L0WW(nHT=x51S|Nik9S~eA=SLdSP>45 zB`@!zN;bZ{1noe!O6F|yEo9SYPU;Aw|F$R$9Qy-*Gn$R3S9%=uY0 zuzum^*08l%aY2nv*b&&l_-&}on6WY>SLX~isjSCJ7IueX}py+3j?>;^ia8<2ezVyG?`-g zD7AAhL#(b@1cY2(6(KM0Z2_5aMH4OiG3G02v5%EPVtpW5*}MnAYGrYzepQqPguKFm zF>VfwTS>sl-ka-nGPSK#=I&np6~rx*hc~z{mbU_Zn$U7xryRH49D#w;eeABUWy=+4 z2Qz6B_rB$=8pB2R!Sd5!IH56oEyt}hM-ju9cb%6lM`X8_LwOYpL1H|DK4Dl|udx&7 z@O|B1etaI>avfui0kfBtO2hZ3uAn#MsaMb!7UB7B%}THU*jhsxr}|{JvEHD5(1cZ9 zo8i8F%As?H$etdx!`Xlh)y{P_GMS8Z=Z1G4QtF#Qpk7%vyJlYX$xP!RV*7$I93_TB z8bbyc5Ya4{4hyD|>XS2#pBt`Y$kQ0U6buJ6hIKV%EPqzx;p=%8d2@yQtg9higd(l4 z*~E)Z8pN3FZ^Te4=xG&0CDQ zZFpxBQeIy}Y8Aa}Bzo72GPm^zkVGt);O(M~;WkZzA8Efl@OSpG*niH0guyXNfeScU|A_lvQ2 ztE1_Pv^FyE1=C)AVYAI8vU`7Xin-)y$g>ay)s;u8>mvG^`(jn=u;s0q>bmGYQgzx$ z{g_l2`r$?BS*WgyLUkRhZtcLKL!W83InP)Ta!qwPDwx{J;%aNkR3xJ*t0V!1{F8vj zJOl`8Yc1DPn!YOF`pQN6N>fmpItvzx%6Wik4huBZQN;pH)uLDq$W84`Q|)D%s!sXv zfP7ygQ&h(_MfDXt=$B6DmjQ~3Ya2~TF@2>eDNSE79i-_i=*}a<6brQE1*REK&{j}n zH{`u1JF&g;{ChG6i>9wsYv&bcs>eEaKoLFOxsNL;Y6z~RdIY$d>#O%tJL_58y1znS zH4V^LAx-ya`fAKE4vcH$K;%&Zps$Youj#8}e~-S(W%{as>8q_mUv205DwD?N_@-*6 zj`l0*7@t)=)m7QfT5W)83JU}D3#A^qXFrB%7AdTm#xD%&J56oMA#j}V8PQB(z0-6W zJ9CT)`bT4e!m=8hp?v%pn8JDoQzU0lHHP(EVI>>Cyq-r>Sd;)$8(d*+Ekn+as z71rv1SYf?F3d?)OO6M|pcOi$D=3>-U;cuHq`hvt&M7w3v}#@B*%aEQJW zNd(}?AuKo8V0EVkhXX$13XULDYH$>eCBw3U6DX7toJ677!8ozP{szNjFeRP(!(4NW zam5O3fSh@VG&XGP9aOqUS};*Uv7fvXO8*DiNs0al0*nX*cQXRoM(iM)r5x*1$@uDc zW>fz&=*Xr%e_Ytqt2yQi!f=~#(nU{R#uOaQ}RC5NVql}{=2k&6hQCfBD z+JDTFPBT`7`b$`+4bY(5%dU{tvyTW1+FmxYiD7FUy*} zEFTS-09M$`Nu;3uA$vLL&)UnQkc_65OaclC70{TGfNt2!{{!~&e}rrLY0U3AFzy`= zL~bJBhP`~lUatFl?B%~!L0xYzC*TEnVK0AHVnj_>-V9U{7RT5dt|4Br!k~JSza+LgB z6;30tQw$EWqlwsSIsb-{{~62qFcQCJInVxEE$2k+KO2Akuh-ek#>1M;{61}zBygMg zw+INE`TXZZfSNLOG|<9E?B!nup3Z2NSJRQtwbLPN^zLqdJ5Z9&+%FCVWcN-w7eM&h zxxj9)%kJat6OsNIY;f=N3jQ^Z8G{&-^#o)MIPdW}Zz?~#h&I^wq4GW}Lv`X!No=L3 zn3HzH4o2Ha?VX3T1KOyNZ$X43x7)_NOI3|Eh1wg&cGlBAxe1gOdRs#dTZt7C9U!J* z*%6jhBClFvpryqo%G9bQCQfQ~tL<+9Wp!7lxI5dgoMJ~>mP+GqjTX!0X)8`)UwowO zo{i<$Zz`v)IA{D#RYX-;uMUTD=xpdq`rY`;judx$B(|oP$!jnbrn<4JNqw#-j}^e= zWAC$lc9%Ip>_b-{Q4m;w{BK)3_}Bk%W3?X1#iHF2bYgL9mZ# zQ}2zkmRILJk@%HyMGVSMN975rt5AH;u9U>n#uXzNa!QSb#^t1!&>26W!P5S3^&U_1 zbLs=rjPJMMP2YDBUtYN+#rS?piu-i?VZ;LaOO0y=abe3=YA*!dXmP8_?!(kTAaQCW zabCfXF;)yw>;aCnR<%WdXa$H=GN^Ll_r)UXAYSPZ4K|e82 zg5TA#jD%?TX2P@Vp8~$RD)9blY(TPOlPO=vfV^KL{}M1t`w@|22UBYQ2zaz<`==~^ zDGrENcwGAS%?Mg5@p9`TpD}mFa3=Un(3t>?q*X+T;lzlS6E8;DUijYgjI{PEwRka#iK1Hu%TBk1^PCUAe#QhC6^(-PB^+aW-+-X0TBh%BHnj|ym9iUt`nBkvZ_ga`pm zxWtCtYk&UY$IA&SgdSJ84c6oq5UbDs}Pr z-ZP7R#;5E)hVw0}n@O>*8r{IA7rqpK1h3`8(#2A|_HVLu@dnwOh9&ii@~z?HV{8h- zT6_moAwSh}xRrXGTHRu)U#ryNTxxv-^*sh|Y>)-`Zo^6f`9T%Yt!voBFayr2Xcx9l zp&j5xwgY^E?a1z9JHYj92l$v0hqJ2M9wiQERY&3Btg2;*vnm{5b#85-zh9ue;m++1 zn+TjqXH^aK%>!{(Wx-ih2Ro~3RpPLh-0nTAno0Y_4N>B(>JmGv>SJeBJE@4v;v1N^ zv_YrLL@=cxfSpk7XQ9-FK)HMej;F4)ps2#9 zluw%deEM5;)r-Yh4O_^|ZNSdhi^T;EcuD!iVpjt;+3561`Jm0v)Q?6q+;>p<1J0km z)6So&tMWV7HLUHyl=uM-pc>G?paAQ#*$tbt6DZ|_W&@6zi0KRM3`%3#z?cqG9B0K= zFlXJy&!Mhk+QOI)_%LBBm^1C6yQ24uYwe_xog6lFe~BYXYusn^(81A5WXo0b5+d(tW~*nP_EWO| zU7A~`B=;Ml8;s?@AQKbTh|nxWYX{dSGPd@xa=rvxMO!)lUHSyOwEU19E9ad)>f{xn zPH=v7L!aExCp2(w=#&2^^a;$#Iwjy!=^gr;8&WApXoi2C5Xx)sOL%owGhZ~*h0dCH z5aDj)&$e`vMY=e87gs*Uihlv7<3U=Z(7DRu@jdj0PdKfBPVd63!EC0{n35QzKL)!N z7MdY*N~M4%Z5Ggk^#Y1~MT6wbac>D|%ufXr;u4Tv0i?q6yBj_o^@MBIS9ZT4sLJVc`BbC-vnfj*~8*FK|42V(LpJ3rn z;3DX_YX&wN0Ed(DjrTK6Qw?;k765ZU&&S*+V-ED46fVOK zlZV!NgwyCGiI)Bp)?9cJ2a*odko$(q|1HPM!mnI1dh_%XFIMpQ3!ia5uQ!}SA{{CX zIauG!(LN;Na|XrWIID-Gs^T){8Wqg^C&B0V9Oc00+@2L#B8e2Db0 z*TWokE(UStq=86t((_YpYWTO3Zy(-RLO?CA2T?x*scCmy!^o9>F+3>RleH6Y*QYl) z-(SKCd4g`Jut-nDo0H~56^(x|LW-DQP+a1?Nt(@rIg4`&>MD#pLsl^(U_;;N5K!O^mrbE9}hnb z{<Bf#6W>~8mW+MFU z_(dSSyEPK_9GqVJ4kOWT*zWN}K&}*OdNefrG2@6A3}w>8xBc+`I5QJN{3-aoh4P8M zi6F~I7W}oqlN{l&A;-RfxDxqD;7G#JRu$ES$PeMx<5vls89uof{I=lt1#mMdPmhI~ zriO+mheo7?CY1-h8u;^ozz21q5y_$9Qz^!<^WdLSK}Pj$Ks`P|4Ux=vriMmL4UL)_ z8j~EFFg}_aPy#Pa`FCgKj_?Z;u!(4XuqxtnS6MRMvvSe)R_aPvL zc-@sawAilrkfwOup?K3Z{p|7RqNP_M_*I?Z7s1lc1Md2Du+$o$|5dQm7C7VWVClbu z=D!&%ZPZVHBUn14r)Fm^2v3M%`jx(S8Biuw5{_q467dwKcygo@F|=;D8#QpF25!{A zjT*R712<}b*MRtLW?Gnr5Z}cV^C^Aflbi_CH%5KK^i5LVFnv>$+(^yyeymcgUq(N! zMcm5c#dkbm6ER0~J2{v_%!%)I3Y$~dq~Bik6cfQ(zTgq-rKn(yk9*QK9xv8O$9Vpj ztp`%P(#9NKBN^a0o7B!yxn@7S#VbPCwPKwmzSWt=1q-~)6BhY4aDK6_6Y1kziKm1Q z!opqU;RH^IB`%96`-a>F9uDB~_`9nC@fy|t-cx)(-aiMEc>~jlKe?y5KcD*xx&Kq{ z|C0Ovhx`A@{axHY!u`|Szs&t$-k)#g{=M9v%zZ6&^q&#|X+FLf{9r^>ie1UiE{d>B zijIwrnP8bbo`VvP$SKG!S`e`i-)YV-E{ce?L|dX0BBJ8HuUN)QKY$>ulWA=(MwaNo z1JlESFc5l4$B1<}W0b}PPQO{}e$fbd2NLHCH~~PHuJaeGTdyF3@#*H82*w*b8h238 zFg6@?VYTVZbW{Oi;Vk-M`Gi6!$eFd#P`4#Ovo)R#TUURB*XyW%yYV)gYPOWvQJrHFS0xG za}dmSsf2SjdKHiVyCgv0u&Cg#jF+VP0^28j6i-?IfI!BtwayHRnN1k&_ zbMmPIN)R;>Y!fV#@Sl1yr$WJPljHeC01Jvrx%j`7BwMRfOEbq>I5?9JC8(r;*IYr; z@Zqny)&<>^RCd>If`)%jmnKnX7Cgg5TMso1G|6n936i=zJm~dE&FQcUA?|&M;4zfO zVDxtZAAp@2|70f)LH{79-{7Yo9^&{~Km3y%FXl@GnX}`mlg3ILxL%>LK<;_`bkfa| zn9mpzkGJq+`J~b6<2C};Nn^ag3Byn#B6OPC2q!^@IB>(v#5_nb*e8*AD&$MTz!xy;4 zJS^nQ0^YC6SX3_XCRh)m{ssMWL(p5GF(W{~(r+I48R)|#qqdm>A!6SB`0K31I=ZT{kav0e=_pZ}EKcktl^=su`>v?jd|Ys3hzy zEmcXPudokBPv4Or4FA*+_(>?oD2%TXJ}&O&36u{(Pv0~d6~H&#L}uXs3-DB4|MvR* z5ORW%uHR5((r#z@@e$0yiNI5T@NeITc>Xusr4iEJ8@XAGeyp^~uRqKgLjKRVoD012 z*ghuY>=yDlUf7EVxqScf9^>?BoRIBpLe8z|c!RaeG=>k6=JCh`$~@dtL(nfB0>6F; zyobxL>6kn%(Z#r0B3sF~xw02X1838LGcn?vu2Q`3m4j0Nam9$uRcOnh>qnfBl3Q$B zP*AL77ua%L#SW(}yL6G1Q(Rb5V0YPbarMNY6znn(TeicIz1U_iayb@D^BviRc3W;~ zVc}x1_=MPv90N&<9XWPqbPhgMPnTpcfG)bQFUql(xKJb*0J!qP4`YGYM`tBJG0iq@ zMhac1VuJ-`nNI{2N)o$aL%aJzve|R9UD@n95TDCe29j8x#62G{r);H8doCq7@D`P%YZq(q zNO1{>CWh|vzy%)ck`TIHWB?Xb?ku+D;Yt@=4wEt`+F_p`J?PCU-YZH5veQj1NR_S; z@wtq|OA{AQmzj8jxKo7gh#BlM746Cwjq=*PAp8;W+2K&O{vua;w_Wuei&R(lWV8zV6tJ5qqU3BfP>V4?=?EULZfjv zitU;BSQ;gzF1rneYiWTqf1%xGpFbZqw-enTck%cnl4glX^Cp=dW20+(EDLg}-zZL} zH+OLfe?g>D?7%2?&YCo}INOmsKq@2-eJ{^|+~W=yr2ay})5vPEBQ`gkhQQ!=1P$Or ztq*d$k54aPVg~C*|G=F*19E`NY;4cOP7;-ros(lPusgD8k;6v0e}(Wnl6+BE{UU1~C1>83zMLQQ4LRG`Z?Y*rO3_&H+8rel!{%i>u|&*WTm%W4?{a7{d_!VDz=k-7y&#(yI8=hOY)H(<6JYd$ zVgN4O)+9wU%@>VjBkAQPJ1PhW{-N%x6!cgg3~-YXx{N(R!7uhng)h>Fna7-1Pw=3W1c26Hf?w>T3ZFj4 zMh@m0)@L}6rvRMT&y)dEkp;ilmleL)hb10z{`pS>j`lnSzu4y#emvv4R&QQ9j(-+D z?V$?1*hdw7e#|wV3k1Kq!kLzE5qWlXv|5Pqe>?0FD?S+f< z1%DZElts322cI2!`}ktNnNZ|JKD3Y|Edk)mA7RpByuM$B^Js4}KlS5J;)^cfi~Nf8 zL|FJg_v07)Fv8!FC^$7=q%W{<`0fggWNx<)7b)j}|rBkGL@rT_Bd7yCEDFB62^_fP)- z@V@%T|Gw1-KYsu8Tm1ONzMJs<`Tf&B5B^a6xZ3mXV~P7s{o7llFZ?Uu#qNujU+kN0 zpn(l1=(Nzau)x!YjcDwM^u>OYxaZcy>z~FyIl(9L^$Kv*{(@iJ2e-{e)Y`jC8l*i1 zzwnJ<_T?A*SM8jCgD*2+!6*E1KYp=)v2vcqDD;F7Bltx44)7CB)V8K7t4=?zI{{znNM<%t-(=h&W1)re*;Sl_zvNehy3qtPu^G_dwe<|mm z^%wY`8iM~I=g<2K{Mnq}znqg5jj?kGey}lDFa&>Oju!t5PA5FUC;X+HU&IT48|N4E zr@#w75neq6|5?uOKmYjiQ`k?%G0}!xErmsVV!IbX>dT@(ihfP{Jy0VZ@GRpEzmoHd WNdNR{x*Ck%@Vv&c-jBgADE(g?lkXJ( literal 0 HcmV?d00001 diff --git a/main.cpp b/main.cpp index 10ee86e..ffbc725 100644 --- a/main.cpp +++ b/main.cpp @@ -2,7 +2,9 @@ #include "forces.hpp" #include "math.h" #include "pedestrian.hpp" +#include "tsc_x86.h" #include "vector2d.hpp" +#include #define BOARD_WIDTH 10.0 #define BOARD_LENGTH 50.0 @@ -50,6 +52,11 @@ int main(int argc, const char *argv[]) { int n_p = atoi(argv[1]); int n_iter = atoi(argv[2]); Pedestrian *pedestrians = init_pedestrians(n_p); + + // # Start here or right after function signature + uint64_t start = start_tsc(); + + // # (8 + 55N - 55 + 144 + 4 + 8 + 4 + 10)*n_p*n_iter for (int t = 0; t < n_iter; t++) { for (int i = 0; i < n_p; i++) { @@ -58,21 +65,44 @@ int main(int argc, const char *argv[]) { p.f_i.reset(); + // # 8 flops compute_acceleration_term(&p); + // # 55N - 55 flops compute_repulsive_effects_from_pedestrians(pedestrians, &p, n_p); + // # 144 flops compute_repulsive_effects_from_borders(board, &p); + // # 4 flops = 2 ADDs, 2 MULTs Vector2d w_alpha = p.curr_vel + p.f_i * dt; + // # 8 flops = 2 MULTs, 2 DIVs, 1 SQRT, 2 MULTs, 1 ADD p.curr_vel = w_alpha * g(p.v_max, w_alpha.norm()); + // # 4 flops = 2 ADDs, 2 MULTs p.curr_pos = p.curr_pos + p.curr_vel * dt; + // # 10 flops = 5 ADDs, 2 DIVs, 1 SQRT, 2 MULTs p.e = (p.destination - p.curr_pos) / (p.destination - p.curr_pos).norm(); } } + int W = (8 + 55 * n_p - 55 + 144 + 4 + 8 + 4 + 10) * n_p * n_iter; + std::cout << "W(n): "; + std::cout << W; + std::cout << " flops\n"; + + double T = stop_tsc(start); + std::cout << "T(n): "; + std::cout << T; + std::cout << " cycles\n"; + double P = W/T; + + std::cout << "P(n): "; + std::cout << P; + std::cout << " flops/cycle\n"; + + delete []pedestrians; return 0; } diff --git a/questions.md b/questions.md new file mode 100644 index 0000000..6de7008 --- /dev/null +++ b/questions.md @@ -0,0 +1,6 @@ +- [ ] SQRT == 1 flop? EXP == 1 flop? +- [ ] Where to start measurements, do we include measurements? +- [ ] Can Vector2d be reused? +- [ ] How precise do we have to be? How do we define "correctness"? +- [ ] Do borders count as obstacles? +- [ ] Do we use a fixed machine, frequency scaling, etc. diff --git a/tsc_x86.h b/tsc_x86.h index 4e4baf4..be144a3 100644 --- a/tsc_x86.h +++ b/tsc_x86.h @@ -1,24 +1,24 @@ /* ==================== GNU C and possibly other UNIX compilers ===================== */ - #ifndef _WIN32 +#ifndef _WIN32 - #if defined(__GNUC__) || defined(__linux__) - #define VOLATILE __volatile__ - #define ASM __asm__ - #else - /* if we're neither compiling with gcc or under linux, we can hope - * the following lines work, they probably won't */ - #define ASM asm - #define VOLATILE - #endif +#if defined(__GNUC__) || defined(__linux__) +#define VOLATILE __volatile__ +#define ASM __asm__ +#else +/* if we're neither compiling with gcc or under linux, we can hope + * the following lines work, they probably won't */ +#define ASM asm +#define VOLATILE +#endif - #define myInt64 unsigned long long - #define INT32 unsigned int +#define myInt64 unsigned long long +#define INT32 unsigned int /* ======================== WIN32 ======================= */ #else - #define myInt64 signed __int64 - #define INT32 unsigned __int32 +#define myInt64 signed __int64 +#define INT32 unsigned __int32 #endif @@ -27,68 +27,57 @@ * Time Stamp Counter. The Intel manuals contain more information. */ - #define COUNTER_LO(a) ((a).int32.lo) #define COUNTER_HI(a) ((a).int32.hi) #define COUNTER_VAL(a) ((a).int64) -#define COUNTER(a) \ - ((unsigned long long)COUNTER_VAL(a)) +#define COUNTER(a) ((unsigned long long)COUNTER_VAL(a)) -#define COUNTER_DIFF(a,b) \ - (COUNTER(a)-COUNTER(b)) +#define COUNTER_DIFF(a, b) (COUNTER(a) - COUNTER(b)) /* ==================== GNU C and possibly other UNIX compilers ===================== */ #ifndef _WIN32 - typedef union - { myInt64 int64; - struct {INT32 lo, hi;} int32; - } tsc_counter; +typedef union { + myInt64 int64; + struct { + INT32 lo, hi; + } int32; +} tsc_counter; - #define RDTSC(cpu_c) \ - ASM VOLATILE ("rdtsc" : "=a" ((cpu_c).int32.lo), "=d"((cpu_c).int32.hi)) - #define CPUID() \ - ASM VOLATILE ("cpuid" : : "a" (0) : "bx", "cx", "dx" ) +#define RDTSC(cpu_c) ASM VOLATILE("rdtsc" : "=a"((cpu_c).int32.lo), "=d"((cpu_c).int32.hi)) +#define CPUID() ASM VOLATILE("cpuid" : : "a"(0) : "bx", "cx", "dx") /* ======================== WIN32 ======================= */ #else - typedef union - { myInt64 int64; - struct {INT32 lo, hi;} int32; - } tsc_counter; +typedef union { + myInt64 int64; + struct { + INT32 lo, hi; + } int32; +} tsc_counter; - #define RDTSC(cpu_c) \ - { __asm rdtsc \ - __asm mov (cpu_c).int32.lo,eax \ - __asm mov (cpu_c).int32.hi,edx \ - } +#define RDTSC(cpu_c) {__asm rdtsc __asm mov(cpu_c).int32.lo, eax __asm mov(cpu_c).int32.hi, edx} - #define CPUID() \ - { \ - __asm mov eax, 0 \ - __asm cpuid \ - } +#define CPUID() {__asm mov eax, 0 __asm cpuid} #endif - static void init_tsc() { - ; // no need to initialize anything for x86 + ; // no need to initialize anything for x86 } static myInt64 start_tsc(void) { - tsc_counter start; - CPUID(); - RDTSC(start); - return COUNTER_VAL(start); + tsc_counter start; + CPUID(); + RDTSC(start); + return COUNTER_VAL(start); } static myInt64 stop_tsc(myInt64 start) { - tsc_counter end; - RDTSC(end); - CPUID(); - return COUNTER_VAL(end) - start; + tsc_counter end; + RDTSC(end); + CPUID(); + return COUNTER_VAL(end) - start; } - diff --git a/vector2d.hpp b/vector2d.hpp index b4eef41..b9bf1f0 100644 --- a/vector2d.hpp +++ b/vector2d.hpp @@ -15,7 +15,9 @@ struct Vector2d { *this = *this + v; return *this; } + // # 4 flops = 1 SQRT, 2 MULTs, 1 ADD double norm() { return sqrt(x * x + y * y); } + // # 3 flops = 2 MULTs, 1 ADD double dot(Vector2d v) { return x * v.x + y * v.y; } void reset() { x = 0;