From owner-reliable_computing [at] interval [dot] louisiana.edu Mon Dec 1 19:08:20 2003 Received: (from daemon@localhost) by interval.louisiana.edu (8.11.3/8.11.3/ull-interval-math-majordomo-1.3) id hB218Jl14903 for reliable_computing-outgoing; Mon, 1 Dec 2003 19:08:19 -0600 (CST) Received: from cs.utep.edu (mail.cs.utep.edu [129.108.5.3]) by interval.louisiana.edu (8.11.3/8.11.3/ull-interval-math-majordomo-1.3) with ESMTP id hB218Dw14899 for ; Mon, 1 Dec 2003 19:08:14 -0600 (CST) Received: from aragorn (aragorn [129.108.5.35]) by cs.utep.edu (8.11.7/8.11.3) with SMTP id hB217wg11983; Mon, 1 Dec 2003 18:07:58 -0700 (MST) Message-Id: <200312020107.hB217wg11983 [at] cs [dot] utep.edu> Date: Mon, 1 Dec 2003 18:07:57 -0700 (MST) From: Vladik Kreinovich Reply-To: Vladik Kreinovich Subject: FW: RESEARCH POSITION AT THE UNIVERSITY OF APPLIED SCIENCES / FH KONSTANZ To: reliable_computing [at] interval [dot] louisiana.edu, interval [at] cs [dot] utep.edu MIME-Version: 1.0 Content-Type: TEXT/plain; charset=us-ascii Content-MD5: wq3A/34uMRsF6fyfjCORhw== X-Mailer: dtmail 1.3.0 @(#)CDE Version 1.4 SunOS 5.8 sun4u sparc Sender: owner-reliable_computing [at] interval [dot] louisiana.edu Precedence: bulk forwarding, just in case ------------- Begin Forwarded Message ------------- --------------------------------------------------------------------------- --- RESEARCH POSITION AT THE UNIVERSITY OF APPLIED SCIENCES / FH KONSTANZ ---------------------------------------------------------------------------- ---------------------------------------------------------------------------- --- The Institute of Applied Research at the University of Applied Sciences Konstanz invites applications for a research position in the project 'Parameter Set Estimation'. The project covers the rigorous estimation of the parameters of some models, where data uncertainties are present. Our approach leads to structured systems of linear equations with uncertain coefficients. Such systems pose interesting research problems concerning their solution sets. This project is a joint project with the Institut de Recherche en Informatique (IRIN) at the Universite de Nantes in France and funded by the German Ministry of Education and Research and the French-German program PROCOPE. We are looking for a candidate having at least an M. Sc. degree and a background in interval computations. Also some knowledge in constrained programming would be beneficial. Proficiency with programming in C/C++ and a good command of the English or German language are highly desirable. The position offers the opportunity to work towards a doctoral degree. The appointment will be for about 14 months. The salary will be commensurate (BAT IIa, the exact amount is depending on age and personal status). Lake Constance and the nearby Alps offer excellent conditions for all kinds of water and mountaineering sports. To apply, please send your CV, including publications via email as soon as possible, preferably in PS or PDF format, to garloff@fh-konstanz.de. Review of applications begins immediately and will continue until the position is filled. Prof. Dr. Juergen Garloff University of Applied Sciences / FH Konstanz Department of Computer Science PO Box 100543 D-78405 Konstanz Germany Email: garloff@fh-konstanz.de Tel.: +49-7531-206-627 (office) +49-7533-934975 (home) Fax: +49-7531-206-559 URL: http://www-home.fh-konstanz.de/~garloff/ ------------- End Forwarded Message ------------- From owner-reliable_computing [at] interval [dot] louisiana.edu Tue Dec 2 18:50:24 2003 Received: (from daemon@localhost) by interval.louisiana.edu (8.11.3/8.11.3/ull-interval-math-majordomo-1.3) id hB30oOA16338 for reliable_computing-outgoing; Tue, 2 Dec 2003 18:50:24 -0600 (CST) Received: from fe4.cox-internet.com (fe4-cox.cox-internet.com [66.76.2.49]) by interval.louisiana.edu (8.11.3/8.11.3/ull-interval-math-majordomo-1.3) with ESMTP id hB30oJw16334 for ; Tue, 2 Dec 2003 18:50:19 -0600 (CST) Received: from Inspiron-8200 ([68.226.139.111]) by fe4.cox-internet.com (InterMail vK.4.04.00.03 201-232-140-20030416 license f018ea6efd6984189790b5f401fab223) with SMTP id <20031203005008.HGMR23448.fe4@Inspiron-8200> for ; Tue, 2 Dec 2003 18:50:08 -0600 Message-Id: <2.2.32.20031203004935.009d9098 [at] pop [dot] louisiana.edu> X-Sender: rbk5287 [at] pop [dot] louisiana.edu X-Mailer: Windows Eudora Pro Version 2.2 (32) Mime-Version: 1.0 Content-Type: text/plain; charset="us-ascii" Date: Tue, 02 Dec 2003 18:49:35 -0600 To: reliable_computing [at] interval [dot] louisiana.edu From: "R. Baker Kearfott" Subject: Global optimization paper posted Sender: owner-reliable_computing [at] interval [dot] louisiana.edu Precedence: bulk Colleagues, S. Hongthong and I have recently submitted a paper "Preprocessing for Determining the Difficulty of and Selecting a Solution Strategy for Nonconvex Optimization Problems" for publication. A preprint is available (in DVI, Postscript, and PDF) from my preprints page at: http://interval.louisiana.edu/preprints.html (Note: The DVI file does not contain included figures, but the Postscript and PDF files do.) We welcome comments, corrections, etc. on this work. Sincerely, R. Baker Kearfott --------------------------------------------------------------- R. Baker Kearfott, rbk [at] louisiana [dot] edu (337) 482-5346 (fax) (337) 482-5270 (work) (337) 993-1827 (home) URL: http://interval.louisiana.edu/kearfott.html Department of Mathematics, University of Louisiana at Lafayette Box 4-1010, Lafayette, LA 70504-1010, USA --------------------------------------------------------------- From owner-reliable_computing [at] interval [dot] louisiana.edu Tue Dec 2 20:25:42 2003 Received: (from daemon@localhost) by interval.louisiana.edu (8.11.3/8.11.3/ull-interval-math-majordomo-1.3) id hB32Pgk16579 for reliable_computing-outgoing; Tue, 2 Dec 2003 20:25:42 -0600 (CST) Received: from fe3.cox-internet.com (fe3-cox.cox-internet.com [66.76.2.40]) by interval.louisiana.edu (8.11.3/8.11.3/ull-interval-math-majordomo-1.3) with ESMTP id hB32PXw16575 for ; Tue, 2 Dec 2003 20:25:34 -0600 (CST) Received: from Inspiron-8200 ([68.226.139.111]) by fe3.cox-internet.com (InterMail vK.4.04.00.03 201-232-140-20030416 license f018ea6efd6984189790b5f401fab223) with SMTP id <20031203022526.BZED636.fe3@Inspiron-8200> for ; Tue, 2 Dec 2003 20:25:26 -0600 Message-Id: <2.2.32.20031203022452.009d848c [at] pop [dot] louisiana.edu> X-Sender: rbk5287 [at] pop [dot] louisiana.edu X-Mailer: Windows Eudora Pro Version 2.2 (32) Mime-Version: 1.0 Content-Type: text/plain; charset="us-ascii" Date: Tue, 02 Dec 2003 20:24:52 -0600 To: reliable_computing [at] interval [dot] louisiana.edu From: "R. Baker Kearfott" Subject: Another New Global Optimization Complexity Paper ... Sender: owner-reliable_computing [at] interval [dot] louisiana.edu Precedence: bulk Dear Colleagues, In addition to the preprocessing paper I just announced, Vladik Kreinovich wrote up a new analysis of the computational complexity of Global Optimization while visiting me last October, entitled: "Beyond Convex? Global Optimization Is Feasible Only for Convex Objective Functions: A Theorem," which we have submitted to the "Journal of Global Optimization." A preprint can also be found on my preprints page, at: http://interval.louisiana.edu/preprints.html Best regards, Baker --------------------------------------------------------------- R. Baker Kearfott, rbk [at] louisiana [dot] edu (337) 482-5346 (fax) (337) 482-5270 (work) (337) 993-1827 (home) URL: http://interval.louisiana.edu/kearfott.html Department of Mathematics, University of Louisiana at Lafayette Box 4-1010, Lafayette, LA 70504-1010, USA --------------------------------------------------------------- From owner-reliable_computing [at] interval [dot] louisiana.edu Tue Dec 2 22:40:05 2003 Received: (from daemon@localhost) by interval.louisiana.edu (8.11.3/8.11.3/ull-interval-math-majordomo-1.3) id hB34e4w16817 for reliable_computing-outgoing; Tue, 2 Dec 2003 22:40:04 -0600 (CST) Received: from cs.utep.edu (mail.cs.utep.edu [129.108.5.3]) by interval.louisiana.edu (8.11.3/8.11.3/ull-interval-math-majordomo-1.3) with ESMTP id hB34dtw16813 for ; Tue, 2 Dec 2003 22:39:55 -0600 (CST) Received: from aragorn (aragorn [129.108.5.35]) by cs.utep.edu (8.11.7/8.11.3) with SMTP id hB34dnJ22484 for ; Tue, 2 Dec 2003 21:39:50 -0700 (MST) Message-Id: <200312030439.hB34dnJ22484 [at] cs [dot] utep.edu> Date: Tue, 2 Dec 2003 21:39:50 -0700 (MST) From: Vladik Kreinovich Reply-To: Vladik Kreinovich Subject: FUZZ-IEEE Special Session To: reliable_computing [at] interval [dot] louisiana.edu MIME-Version: 1.0 Content-Type: MULTIPART/mixed; BOUNDARY=Murder_of_Crows_345_000 X-Mailer: dtmail 1.3.0 @(#)CDE Version 1.4 SunOS 5.8 sun4u sparc Sender: owner-reliable_computing [at] interval [dot] louisiana.edu Precedence: bulk --Murder_of_Crows_345_000 Content-Type: TEXT/plain; charset=us-ascii Content-MD5: RoP1GncBK4Xg/a/9EKUGCQ== forwarding. Vladik ------------- Begin Forwarded Message ------------- From: Bob John Dear all Hani Hagras (hani [at] essex [dot] ac.uk) and I are organising a special session for FUZZ IEEE 2004 on Type-2 Fuzzy Logic Applications. As we are aware of your work in this field we wondered if you would be interested in submitting a paper for consideration. If so, at this stage could you send me and Hani a working title and author list so we can inform the organisers. You would submit your paper through the web site as usual (deadline Jan 31) and send a copy to us so we can start the refereeing process. I attach the proposal we submitted for your information. Please feel free to forward this email to anyone else you know working with type-2. Best wishes Bob -- Dr Robert John Reader in Computer Science Head of Division of Artificial Intelligence & Computer Modelling De Montfort University The Gateway Leicester LE1 9BH Room Gateway 6.55 Tel: (44) 116 2078491 Fax: (44) 116 2541891 Web page: http://www.cse.dmu.ac.uk/~rij Secretary: Michelle Dale (44) 116 2577488 mdale [at] dmu [dot] ac.uk ------------- End Forwarded Message ------------- --Murder_of_Crows_345_000 Content-Type: APPLICATION/pdf; name="type2session.PDF" Content-Transfer-Encoding: BASE64 Content-Description: type2session.PDF Content-MD5: YJZ3t0fSBcy3HSerJqyZgg== JVBERi0xLjMKJcfsj6IKOCAwIG9iago8PC9MZW5ndGggOSAwIFIvRmlsdGVy IC9GbGF0ZURlY29kZT4+CnN0cmVhbQp4nN1dWZPcthF+318xb5bLGZoEwStP sWzZsaOULWldqUrpRV5F8qGVdURJ5F8fcMiZAdRfEx/A2cOW7KpdiGw0+j4A 8PWmLCqzKce/+x8uLs9en336sN08f3v2etMX9fhn9y/+zxeXm7vn7jH3kvvv /NlZWQzDYKvdP1abzv1qNnVftGVZb84vz+5sPj7/eX6+Kkp7fKU08ztN1Rd9 v6mHojbzS9X40r3zzYOzBw4pW3flxpZD0bbD5nL8tTr8+mL6tS6b46+l9+uP Z//YvCTXo+FnbGED9DYBegTwZlM1xdB69LIz8L4e1941zciGEfbnI2yHSmPa 3j1x/vTszpNxqHfI2c7MQy/GIYeWNW2lD+3wrN3Ept6/+GwcagrT1/uHfpXQ 34xDXVH21mGugvpuHBqKzjSdjuiraajvG7Of8F8TdFNaszDh23GoLYyt2zZj OVVruyFpOQB58NQjuR59if6MF3LoJ8kxjtUaXkNRlo3pdSQAUcEQwAsIyUu5 bIBX9oufST0AhAZDgF7zgpr2CAuwA9D+30K8smnDyTOAJUUcvHe+o1ZZtH1b z0PvU5RxpMP+xa2Y0CiqsTXObtXDZusseN3uxr8URuGdlMzfJPF/m8gztLbs 9QWAld+XKwdEfC6HACeBUBxN/eTiHn41//DmOWv4PZfSB26yqwYnD0eP0hdd 21QjOTtnv458HdyCm70x+3F60pZ2CBS9dS63Hj4Qkc4cTOzXE1f7vrb79+4J 4MqII2/ZtymgX0osZ11yVn8YdMTfyCEA68nEqKa1NgBvHVnN8IGumtLYwReN 8MUZvOkPDwHoLyQoSYfPZ1r1ZWtT5nuWSxnwFCDWBfWiXM6Me9vbptGhy/dm E1CZZk+Fd/K1Wd87U3845CP1XpJPzvdIEBS89lbKKCeQlzumGqeoB8sBYEms Ht+ZnqraumxVynw/i8xg2j2N/yl0cB5xfOhrdb6vJ9Rd2H3g1lGb+04MRRR8 68AUnYusR2PkDPfBDYRkLdUh6/7saWjlU48/ltT/k1iUXKaR0BsJfSuNAUB+ YKTrGzkhEOfZPpS9sfsXKdkFy9FJ6g8BkjLkuyv4/05oz1PV3vpovqL8EKd2 DOZ/3clpVbS2HXQ+AAP1XFIKGHhgXTUGbqu6GzHzVeMYJpSbr9z/VHCgp5xT fNA2zuMcwwPTjL/dH4OCjxyqvXU5sB8UNC5WrmufF5MG2JCmdWvK/RK/nFZd mWNQ+k6A+m0C1ZfmYEbA0PvpPVvWptYnvL83+9NaD27Gpeu2rjufZSGsnyZY lakOM15QS3w0zeji7q7WUX07waqNaQI5DWcERL3cC1xTLix7cgYu1beHGQHp AXEeyacmyznpxsKUgGBgkeBFQNY301NtaQ5xMSDFUyE5Mw5lWR1ycY6qEhRA 9L+Tfg7OM9ukCaW2AOg/MCoFiPxSChcQZzDhEwn+JYMoeE+SD+D+WjwkVR/A 5hgI8ASaAl6k2PWrGJErls+8kKDlQzOvxuR3r3CcZXsmtVJCf0OpLsU+oMxA GoHoSaHSpVgSZoSkYw7WJ4kgNUsK3rw6l281VSgZvm5P7tg2bsZmjFQHF4Bp xLlio6/bVh8WJd2c0gHoT6RXBgzRtTUCS/INaJQOPWKUpZwArFTN8PF8IecD Ev4fKQ8Ad11hAbEiJAUiAqgMSAO8GoAFDBAAD7y7zgyfOIASumH0Yb1irBmA /kwGHZRBBcpJuVZAGU5VKPMJlEBnmD/hheQhh5YkFpgQvPeOcRk6sSLWB1hx sGgurgKCy4kkJQ9qDBrhPZA/qQKcXQZU5mwBQOsoDmaBpLtM3BRdW7VNEv10 /wdY7cvIhz683/VJaRKBeYEAA0eludSATZS/BoI/Q/Jz2ux8D8BSYwYfB40j ISg1zYk4DeAhuAAf4KC5z/ApqUTSwslnZGgJZIELnDjMJQqvKGcqF8N5byWZ hXj6oGiDR4oekb6HkHL9tGbi8zmzLBzA4gI2JEVyy24amGWQDoE184FI+B4I hQH5gMwC0sjlqMldxDI9lVgBH8ZJEUUZgFYxucPWViGmyF/tCtRhOUoSA9Ss gI1cm5uAgItyYRHoetwZSTqyYZ3aRoTQZ4aPOwBalWNSfMFquHBYj5GWHYEe t/tDEtGbCtv9pyQOSVE7SAi5fIlDi6vSnkKdAKh1EWGUEAkZzSkWCISbI8Pt kdv0dBO8mK33ADzou1CMBuGCHvX6TXJALo6J2RQEeYTqYddJSqShBXADnowK JAAV1wlKmJADEZBZHKCrJJ+WA4cRD7C7p24ShwvUTMm4aaA+GEYdeR+WlpSG sMBTVP8PWByAvLSfQBxAlgiWQ/ECaJ5e9o1IqZqeUIGzqCOBoa2L5gtTDb6m a8/mDJ1PDtTfLqH2GsaNpXqvQW/MMAZz2h200EKSWHENKgCdjyHD99aGJD79 QNRFlUa1plJUYwHhqeiTY+vpdAPQiku3dVfsw5KGSy/X+hFCyjYMfz6GCkDU VgbSPgbZkVPujhVdFpa7JgAFIO3cCoErpTpt2wmU7c1ho33F8Gs2Df4+qgTT EMICVRDKR/Llz3A+an/Bst2uVBpzm6OyWzladwEFyDckJn4BJSVY9OcDQ1qw GE7I5Z2AMpplDsGvSaxjpSU9TmO6vSH0VeY0GmJchxQ1rTlYlt9b1E7phG8h MlPZq49LQsx/lHwHCsGVqwF4iltUWAeAA1VidhJyJjihdR61JyAE0KKCqKKm 6LyP1gk2fSxo1w1t+kAuUg35uJ1aoJ+YbBS5BgnAQZOwcOgHShv1FqY/BMQi oc4blpFuxK2You+rLpY0/l7cCqwFZbqVq+1UcyXMzB4hlKvlJjRX3eUsMVfE TNnzFdnioVsoJn9BJTcn7EVtq2ssuZ1Q1Q1WjtrlptXhvbWVLB9W0lEa/0V9 X7D/lFRRQAeQMgNQujVrFgUFiCZ3BmvxnMACeyTvuRJYShEncg6CO7l19aWK kDDUbhmtJhCCWuVwQyHK1cJrqAaEE568GhCCv25p0JUiUm4CsCgckrSQO1cQ 6Y2A9QACgl6iB0s/OAXoIGM5TrSeSLVQs4yVLVNO7iObzn9XkW30ZMQNdTl9 6FpPM4QOYp+UvkakXXZ7mOojenKm5qYry0lAQroSA6VvyYu0jq4jDk6m+loP zLUwsj1wpG6U1CnljoZH9upzVUowIxWr5ra29XNAWJY/PAckw4BIbsyxR22X LNXrDid5qZwD2AiArpR76jjpRmqentclnwgHfWMgllTXMbt+ClYI0NJNnv9i rp9O2s6ZvUkvvZCsH4jzVw1m1I6rwxcjBhS8CMDnXgKzsv7gg6I6EmDNgNVS N7dS5G/InqY22wCl1JDCX57O9+sOKfyHriW58SdcG1KIY1NRaw7MBXUlzcpN TbT3MB9I37W7j6VbOA5ZcErYTHUxOB9y7Ru9uaMsES2mzvTqzttfYe6BU9DH B2RICZqS99+B0JDzrZmWOSnQBZ4n0mUFZefsdEI3ZstXZHBxG3iKqzQC5jOX MenV94iurrwvx4fOnS78heI0AJ8UFAL9Td2mAgQEoAC2ByRd4xMxY+rW+ojy AEyzN1YCKie0caMysnT7mFkgIGdYkqIRassLpTHMpkZwsRLXdM6s893ug1uR w49atXg7VIVpbPsH6737D2X33rnUl5Ln7NAfCDm3cUOPb3zCc/HhiTdGRhdN XVrC3Y2Re5tnyi1a/mqSLq+IuOeEyysigrV4ecVCDHTCuysAVtzNeEkBjw8+ 8/ZHLocF8wGXwcT8wMpTmAPq6dXB9JtPqVoWt21TjzSAHfDJl3BFTkSIPEQr fb4r3c8+M4e6vCX39lZuAx/nn7i2ZsqOK1DRT04MORZxrWoQUHC1Pf3uofTA TBOL6LK57WiUaALvqopbJNbl6qX65p+ImZdWibteL3evINdv0S+kWzQSIJHi FHjljoPkeExSHSgYty2Vqwpkon4zHthHnNNmrjVOXZbNuVsuHqP2iIPwiKt9 A84vt7R0E8a5mNOXLpLu//Ivj+P26FPnBHJvS+OOPVxJxLVw0ujW3PpRVVVh TXeNZRDO4XElOS4EWlmfT66EJ+0m8KHrIcqy2AP7DyiT3WThTrnoTo/eChV1 epHCQdKejeRyeYqbOF07L10pwFO5l7iDIFMeZNHny/0KR7LGZdMKaE7uh1y4 0FD/WpKPKRVYS07IEabkrzfI0vs06y5A5PZBLPepjys2C2aBC325Na+8To/r RKvNyEhTW7+oKjVr26hRTO7J9/Rzp1x3itohAnCgNJ8rdOfeiKjfB3G72HUj +9/C91ZuqWfuQwzLZikfrsr8Zl60XnnCDtG13J6+LJB8s+mkguwPrb10PWL1 +VvXY3bnSvpWSy5SyyJDVswJnt9RTkmIRP8OecibOXzAtLGDWEv9oBrTUgsg ydUlNOcCSFzNggtdk74ZKWxpgBafgqFoNnJybNX5+BiaV+/lAgxOv8l7cbpb 8cE+rfcQYMq7juC166iUR1wxtbzT96ojga5WCYr6l5URiYLpQgh08tKpb9WJ ZvXhTNy6JN0na2aWzrW4T9c0S7oLyyfhCT+SmiC68CIgbjd+dpcnveW3qh0b DoGITT+gE0kFtOjv04dVs6ls8L30etw4ePxeurFtYe3xg+lVu9OY++New4/O OqffNiDk7jbbxvhU262+bMIhR6LKhITsrakCje/cVEEOHkx3KAZXw36v47c7 DEe8W+OLxW5HpG18lviwn0zPDG091Op0AEmwFEADgAF4cc+P2Pfr3eyF6T1+ OOXcseOLHTv8s4RAI7hmEvjqElUVAQqRDZ075ZHwib5tZ0ZcrTP6Dh+z+4cv JCpUtY7bDAls9QlvWKdar1ofLHzv84kMtZNafWcTdc889TkMbgs6wHPe3VyW dbNg7LhNTNmn3+ZOq+2Ktu93wjTltd9LWaI0JPu7ZFw/LqH4fxphujcxya8h cO4UrPp/AoeJ+r0z1rXxNZkSYRBbUBc4cV+V0SU9Uhrn4gGwwm+lyFlMsRBU LR4C5wYeSODzWYum6KuuXhb9v03WwrnPPqwIVY3zs0Pjvf2dnJu5U5g6Yf9n 1W76Ee0nB6d9uCxb0pGirLwx1UhlK8VDLQO7Fw91YqSWs0nYcn7lotfdfpKq b31FA+dQgDeUiqtzwl/jJxL6qTmxSIiWMo0MJ+Rs8pleziZBz5woy6KyntYA M5tdQgZ2UWPXvfPNg7MpHn341fzDm+dnrzd9UY9/NuOQ//PF5ebu+XJgW5X1 mFNU1cj4XWSbcrou0pz5yzjkUoKy3wtUdr5HOSgQYXM3+4EXZTT1iyKfM1tG jqxkxpRljCzxkj5jHfqmdcPVLvl4CkPn7HzjIeO8U+5fY7Z3hy7gmwkW+LaE T2spl0qZcuvI2I1kWE42uHCQ2xu1ppEHIz/mnqLwxc9UmYgwBMStYEau7blu r1y4oK+np/w6P5Wb6QeVI+enuQ3UVMq4LpvhMuO/j0ND0dumTyk9gqKVvscs c0tMyEg9OYso2u3Jzraj8e1N5ZsUcAIYYMcVm5OSHO7+2Yi5ANiDqEb9CIFv vgfx0F3pU0DhaU4nm6KtKvMHymge35EcUANpX9Qef0yxTjJFjqhJjY8oE5Sr gfNiTiBFQs9yTFH33dVmOdnc8UYWmOODX8Ucn8uSOY2UGImmnE3yK4E7zuxN n1W44dQnpI1++N23jT9PQ/6V1zIhoQ6myVxAzzQiV7msyjSm5MEcM4fRGB4T h35wxtAhfUgddpnJ+Pf/mWyIDGVuZHN0cmVhbQplbmRvYmoKOSAwIG9iago0 MzE2CmVuZG9iago3IDAgb2JqCjw8L1R5cGUvUGFnZS9NZWRpYUJveCBbMCAw IDU5NSA4NDJdCi9Sb3RhdGUgMC9QYXJlbnQgMyAwIFIKL1Jlc291cmNlczw8 L1Byb2NTZXRbL1BERiAvVGV4dF0KL0V4dEdTdGF0ZSAxNiAwIFIKL0ZvbnQg MTcgMCBSCj4+Ci9Db250ZW50cyA4IDAgUgo+PgplbmRvYmoKMyAwIG9iago8 PCAvVHlwZSAvUGFnZXMgL0tpZHMgWwo3IDAgUgpdIC9Db3VudCAxCi9Sb3Rh dGUgMD4+CmVuZG9iagoxIDAgb2JqCjw8L1R5cGUgL0NhdGFsb2cgL1BhZ2Vz IDMgMCBSCj4+CmVuZG9iago2IDAgb2JqCjw8L1R5cGUvRXh0R1N0YXRlL05h bWUvUjYvVFIvSWRlbnRpdHkvQkcgNCAwIFIvVUNSIDUgMCBSL09QTSAxL1NN IDAuMDI+PgplbmRvYmoKMTYgMCBvYmoKPDwvUjYKNiAwIFI+PgplbmRvYmoK MTcgMCBvYmoKPDwvUjEyCjEyIDAgUi9SMTUKMTUgMCBSPj4KZW5kb2JqCjUg MCBvYmoKPDwvRmlsdGVyL0ZsYXRlRGVjb2RlCi9GdW5jdGlvblR5cGUgMAov RG9tYWluWzAKMV0KL1JhbmdlWy0xCjFdCi9CaXRzUGVyU2FtcGxlIDgKL1Np emVbMjU2XS9MZW5ndGggMTI+PnN0cmVhbQp4nGtoGNkAAETAgAEKZW5kc3Ry ZWFtCmVuZG9iago0IDAgb2JqCjw8L0ZpbHRlci9GbGF0ZURlY29kZQovRnVu Y3Rpb25UeXBlIDAKL0RvbWFpblswCjFdCi9SYW5nZVswCjFdCi9CaXRzUGVy U2FtcGxlIDgKL1NpemVbMjU2XS9MZW5ndGggMTI+PnN0cmVhbQp4nGNgGNkA AAEAAAEKZW5kc3RyZWFtCmVuZG9iagoxNCAwIG9iago8PC9UeXBlL0ZvbnRE ZXNjcmlwdG9yL0ZvbnROYW1lL1JURlRYRCtUaW1lcy1Cb2xkL0ZvbnRCQm94 Wy0xNjggLTM0MSAxMDAwIDk2MF0vRmxhZ3MgMzQKL0FzY2VudCA5NjAKL0Nh cEhlaWdodCA2OTIKL0Rlc2NlbnQgLTM0MQovSXRhbGljQW5nbGUgMAovU3Rl bVYgMTY5Ci9NaXNzaW5nV2lkdGggMjUwCi9YSGVpZ2h0IDQ3MwovQ2hhclNl dCgvc3BhY2UvcGFyZW5sZWZ0L3BhcmVucmlnaHQvY29tbWEvaHlwaGVuL3pl cm8vdHdvL2ZvdXIvZml2ZS9uaW5lL0EvQi9DL0UvRi9IL0kvSi9ML08vUC9T L1QvVS9aL2EvYy9kL2UvZi9nL2gvaS9sL20vbi9vL3Avci9zL3QvdS95L3op Ci9Gb250RmlsZTMgMTMgMCBSPj4KZW5kb2JqCjEzIDAgb2JqCjw8L1N1YnR5 cGUvVHlwZTFDL0ZpbHRlci9GbGF0ZURlY29kZS9MZW5ndGggMTggMCBSPj5z dHJlYW0KeJydV3lYU9e2PxHJOYpFS3qqVE1iWxWLY2utVK3zWEFUKk4BMpKE kIQkzBBiAAE3o8wECZAgikBRFHAqzq21VzuinW6vWqu2997a9n5dh27u+94+ 2Ol+733vj/fxT8haWXvt9fut31pbQI0cQQkEgoAtEWsitq8KjtDFq62zVpgM Kv7bYG6igJs0gpvsgxfjh79k/BLhO5naeMD3CU48hmqe9NSzAYP6JwefHMvd Hfc4TpHgoKBPwAW9sSVyRnDwzJUmc6pFF6u1SeeFhIRIFanSXy3SVWqrLtYo nUY+JKkNJnO82mibLd2qVkttWrVUozOopSs3he9YH7ZWGrQ27A3pWrVRbZEb pOGJCoNOKd2oU6qNVvUMqcZkkRoe/yNVmowqnU1nMlpnS5dbpXKp1axW6siP 1ClKtZk3zJSa1ZZ4ndVKPkt1VmmsRW60qVVSm0mqMyoNiSr+ePK9xmS0Sc0W E7HHEwsJFW6y2qxKi85sk5ITw1eteZyjTSu38edadcQsNWmIp8qkTORv87vN JtcZrVKbOsXGn6NQS1U6q9kgTyXnklBmi244hUSrzhj7x+kzpRZ1rNyiMqit w3H5qvxxP+mfbi03mw2pw781DXv9fr7OZlUbNLPDdPGKRKt0iylebpSGmUKk G6WhapUuMf5/Gv6A6/8HIEVRMzcuTzUqV6SZVCvN6p2bNOGrLbFrrFqbbuu6 xIj1b2wwyOPnz3hZOnNWyNwXgyhqFvUstZPaRK2inqN2UeHUaup5ajO1hlpE zaXWUvOoadRWah01nYqg1lNvUBuo+dQMahv1MvUCtYAKpkKpFVQYxVKjqacp AfUE5U89Qz1JiahR1CxCRmokcX1LsFFwfsS4ETUjBnwm+1SNHD2ydOQPvgt8 i3wfCVXCL+kQGhg5c3vU1lH7R90bbR/9s992v2a/L8a8PubQmM+fMPtP8Y/z 7x47dmzA2IixhrF3xy0fd3DcP6HYf3AO8lraBilvwG1YMF5kHgyAZNYo1C9J NmjjFJqtqa8hJiylpkfCGWg0UFtzp4Tx4h8s9I28A+loOzOUQotK0Xa7Y4WT EYVbuAHaf1CDvLDsGFjdsNUtGFztZfOEKpTeuANGYyvXPMEkxJahPvXsrIxd zkALRHvp8BJbFWpBNeXexvN1xyZAorBjSXUsSkRJ9j3agr3Ivi+zgLHALXoo HxzsQ8h7hPN8/bl8LL/E3XYLEPj5cDZOzuJgq7CnoDoHWZGjwOm0rMURE1ZC aPJbCKHyQFRbXeEtchVO8OBiK92cX5Zf5ixGfTPq5TAGn56Qui1tR0KcDQU6 s8srClFRYbGksLSwFJUwh6xuncacoJZ16y91HKpubBQ3Nru7Ox/CDG7KhPoT lZVtxYw/dw55uIDrsZ6AcyABMUjGi/pAyye11SrszanYg+QMPkUjebYzyslY IddDa0qz6lAXA24aleyvrKxgRLeamzzVzRM7681bJXgvjTT2dF0ecc7y0Kri PdXoBAMl9O1tF0I2yUzhGrE/F555nQtsIxWY4sPNBQGrkSdkKhAzP+IOTPrx yLv9rrpsR5mk1EkKIC/cecDWhZgTLZ5jV9edfglPmj4FP4/F94Jg3K2LbZ9c ljzmwhHul8MWb8BdmA9zIQSLYdl40QuuwYWsqM5Cv5/X8CvqaHtG1opsRqS3 wAANS2DkAPj2pRzXtkm0h3V1q+sYkQGYjp7jN5/5cfYlvEgchm+zh4Ui/Y9n 38AjXlDseDU6rb5XwhWOfLOu5phYZPjgSMysZ/UmPDZfQq6FjnCz3AJuIcxn TUJnXHZqelJctN6+CDFGIaRz031bhmg6Bz/r2yL0ftviAuk+bB8KmmDhRDTh RWYbt4QvynSYB9N9oISLZ7Eo6Fk8CT/37fMwHtjvfiIYSYN/wqyk0MBebojA C/Hs5MjN4clvwTRYdaD/KkniXObFxGZu9EVY6wlAEASv8JhmwH07e7LGW3p1 /9XyCc20rNheQ2DhPsJxVvr63oocJGPw63TMyXiXDDF4RIgUT5WI2vGkm0H/ OtfuOntKnP82i3L3pGXZk21JjkTEbLO8DSJgD1653N5mVdcSFDQppwdHtZP8 V/sMToWZbHhonHVnijFLn2tC69HGTmO/8VzmR+h7BpTXYASMOdWTqu4R96ia 9Z1LGS/YzcJMrcORgpwooySxMtyzq2oXSWUXnoFn4xisgOl4Gig/vFLfdUFy pKnJ3XeWZy+WNyZ5rCC3X+BGXyIsvkgqtACCxovW/onFlTyLhwpopMh27iLg P2eFFlrkaqTDitPc6DoDU+jTu1vsvYiBkZ//RGomWfYt9gmPtW02SEQTi2Qs 2l/hqq5xNzVXNyGm37MJj8fjTOFb9Mbmoxk87pkfwaYumN8m4KbBk6xypyE1 BqmQ8UBSZ+qR7Pa89xkoFeZ+lXEq7qj+mMwdQa715CtBeBqe9mUwjPvizJEH 70lwHSxli2HMge4zqBU1pJcbSlOKTEXbmQyaJ3jnoICIkj/ksglC8yybcrde Zw3JCEbMciE3dtDX1zXUSCdhhW+r8DRIetsgqKC6sKOwq3Ao4N++E9K5WyRI Q0r74IRhfOBnghE3eXAti9WkvEFYhZWkvNNBSf6mk/srIQbPgGAsk+zTszDi Gn6CuMhD8QhMYzqUgBcFhneBgVESf7hNBGRyL+xqE1wCic+H3Mssvk2jGEeW bC+RgDYPvbooqxq9x8D1O73zwZ++UNfUeaAqJysvPyd3j9iZkedAexhDi7Wj y+vtuhJ+/JV1kdZoszhRm6NALzNbooX+HCAvN9RJuvsrgu2rfGdvhzq4yMLE wo7i3gNftPW+g24xwMz8C+lYPCAU9VnoD3Jr09FqZiiMRqH25N25+WhPgYOX 4zteenGlswJ9zIi2c5kjm+ur2//aHo2pjVnqBJvYmqzPWYWYX+cLBx3DmkIm TBB3j4jJ0CdCUc1/6ElkpmM5oZTWwn1Jd4PK1yQ0rEs2xRplmoj0pYh5nZ9B sJcGATquqsYM0wLfJAgzsFAVi6lCBjLoozUV/fDKGR0O9Iq9Ql5EPNw8T7Y7 4DzfuO+DkafxZjIhcqvsSDksxoo9juhsUtwGDx1WklWDrjKwme5v7Wkvr8hJ rxI3mOucLYjpamk9JhHN7dO1796hNq/bLeZHj+II+MOYOwfBcM7ZFoDuLobJ MAEmaj4bLwqguCTuWfYcfpoWfUu5ok1lyonY75X1M3U1+maTpMnaYD+VvNis VKNIpDiQes3GiP7u2OjUWxTPbB7QQQC8erf/n7dCT04VbxAuSt7feri1qVvS t+8U8pZ1uANFH1B9Lc3Hzz2DurMOm1qZ14a2saKPHedb0vSyiISgl2TNvY37 q1qbJaJg6mvOyZ5qUM+enRK7e1fqmw8eNnSdkjymwb9/BySc+5wH5JpQ1DUM SAaK/BMgFhjy0nNKUqvRBQYy6R8KujS1pPpe+MYiTMVCrXY64osvCu+qqXzr /pU0bZu4yVJlrt3IeAgGgZkd3MJ2QdH3IP7OBx7AZyxMTaRzM3L3ZqFctKco vVRfZSshKjh3fegr64/LHsVIbsS3WlAco4pPiFot67uTJDbhEK+wqrFsfwUq ReUFlXkHs845WhHzzbtX732o6gk6IVl7yORFHUx7q/doy0FHUpO4MbkqtS76 96FMFFzw61T2ARW5LI6wCk8692cjBTO0//+cyU3NZCAj5n+byLHF9jp+IhcO T+RtmuTNkWLSxVkfwi+9gn7+qMuEcbNgiUUY9rIyakMEY+7Ht+gNPcr3r3cf /vK42FKVrE7PTEKBJkd9twTOP6T9oW2YtAMwJ6AX5hCdJsx9ACqeuSTlXmeF k6Q8zNycbJmD5JHtobWlWbXoKCN6tJr7mY3RauVR7fre3o723h5dZww/x4hc PdOW4g6oAAHkgt940UIuhTOxEB9P5yRkZ6eRQeEoTiwn64drd1yZeiKetPA1 HPBS/0bw2yK5HndX7klAkYFrdka9tnmH94xabO7KbE/7gDFidEhY5i2vcKFy VJXflHM2rSunh7Ah6Pu/QcCAbOD1IxI8/eNXuxO96HTg1RNHb/SdSNG8Ke7Q uozuDUQZGsieeP4UtHkFfyNylAsLfLgywhCcRqOdaZkreO498NIrK9Jr0BmG I4w863bdKCXr5wOiSnvd6WgHg5U4F5TDlg94yz0LfT33QAax8AzemWFf7SRR 7nnpFeVptSQKpNH8YH/P0MyJPQFnftvUUuA7O9tT5yl5b//V/WSqRxcRMehh YAjLrPTZvOosFMPgN2lRn16f5DBNVGS435LAfRxupY9nlzv5za6Ufuls5F/7 O11nL4uHNW/Qzy0AC7kTlMEoFi13OF7OIakMeun5ZY5KdJPhPDSMzeuX1WM/ 5pCwAGS+Q38X5mOZ7yFhPfj1n4RxJYTAgSSS0Ct4wAfy/Y5FO1Izl/FxfvTS S0tTalH/cGH6G12fFJPr/2ihP9rbmEEUld8izgwKvIIv4SUY6PUZ9OHeZdGV Wte1SuJnttCf5rqT0QoGX6Zx0NY12G9q1BWYJIY7NOoua6qrra50lZwuIq7f kFrnVNvRSgYT24zGne9En9ZeS/0YfYI+aOrp6jvachM9IInmB7cPRnnJUGQI ywiSz3AXWfRRbd0AD8u/LPQX+e5M9AYzdI4OTbFEmVNrGsqLyyqqxdV1xeWo junWdao0ep1a0Rp1zCZuyex2dhMmST+5Cc9JhrtikPEI3uKbSsH37xarsDPf lYo0w/2rSUvW5T9uBn1hegPq5vPJdHNrj5DFbRrRaTGeDJPJDODir7F79uTn 5+/biwIzcypcEqilv1p6Dvtj36Xblyk8luPdrZ6O2tyaLJfYWZlfhsqYxlZX p1j04HKLOUyygsZ+sTuTNZaY+CQd4UTYqahrl066z14QV21vSjmDelCb6+hx 0oo48hJrNiRnWRETn3Sw+/KJE583ks3uk9/eCBceM+8x+86A8rcNK7cq67cG dzweTTl8gzv4Buc1qczVUFPJiL5wNdbvr594rM62aViVFA67hi+AnTiXEAF7 7FzKOzfVE0/EHK2zhg+7yh1Zat7VSZ4URY5qnuWPnxRb1LbXI8Xx72o9G9FG JDOrN/FVRG4YBz7cPHcAFMJU/Bw8N170iDw1yllu9SI6/DX8FF60fs6mmYHY Fwrx8zRRq6nw/LvgIxQ9wIqnsVoIajjrS3RxcAFpiOng63MQL2avtrc2Hj1+ +/5J8EPwFLpv+XrXTdlHMd0zrjAm+pLGa2mMOhxaOxvhSQiLM6eZNiVHWxI2 KfmMst/hckAkgJ8+9SEPsJMstAl7QORLtpUX8D++zTme25sZ+LXp3Ba0ntka Hb3VZq4+rBTHHImvSUBMbIJFu+2y9msIgVHfw5SfXv8cC9ok6ipDSUgp409e w1Cc606FlwKAuTFeFHMJr2OBuX72L+g2877i1KYdev0OsahRdjiup7QMFZaJ RTGFxUWoaGJVudOu1smXRfJrHCwFhvRBng90QzeL8zADeWQU5tvdgPrgvw4S 20wC/VyfNjsLUvcN76dXfrh1/mcEIgaWzwFfLMVPzgvCk/HojxeB4GhLae1B cQye+epi/LyRgTa4we6z79ubX5CzNyUvAzGhtivkmcNUHz7jrXfYqiR1psrV aBUyF8TlR/H1srcOBraSM1+ESfCiDxQPrmI3oPXRipVxEYmz0EIGKx6ShTWQ PIpgBshg9wtkXWXFN4Y0LArPXuI0O4zpmTbE7Ei5AkESSKUhuOHy6TPuUDyd 34MQBHKhIBTAmzDJh7NALGsVGldm5U4lq3mwkMj67M8GOq+jm4E/v3YTUzt2 JcdqxSpVss60jGkVtl3taj2PelF7SlNsdVK5CdlQdEpE3M7tS7RystTPZWzc fPpX6iDCHLgyuIEN1ZuSVIoFc2TYD2EWzfW+eHrpyZU9yn+EM630ljctLUl9 8Vcz7yOYjEBc+6j17cZeT8vbRBKS3ZzcDZFuYftokPq1t48ZA9LmMU9Q1H8D sYt8dAplbmRzdHJlYW0KZW5kb2JqCjE4IDAgb2JqCjQ0OTgKZW5kb2JqCjEx IDAgb2JqCjw8L1R5cGUvRm9udERlc2NyaXB0b3IvRm9udE5hbWUvSkFIVVhE K1RpbWVzLVJvbWFuL0ZvbnRCQm94Wy0xNjggLTI4MSAxMDAwIDkyNF0vRmxh Z3MgMzQKL0FzY2VudCA5MjQKL0NhcEhlaWdodCA2NzYKL0Rlc2NlbnQgLTI4 MQovSXRhbGljQW5nbGUgMAovU3RlbVYgMTExCi9NaXNzaW5nV2lkdGggMjUw Ci9YSGVpZ2h0IDQ2MQovQ2hhclNldCgvc3BhY2UvcGFyZW5sZWZ0L3BhcmVu cmlnaHQvcGx1cy9jb21tYS9oeXBoZW4vcGVyaW9kL3plcm8vb25lL3R3by90 aHJlZS9mb3VyL2ZpdmUvc2l4L3NldmVuL2VpZ2h0L25pbmUvY29sb24vYXQv QS9CL0MvRC9FL0YvSC9JL0ovSy9ML00vTy9QL1EvUi9TL1QvVS9hL2IvYy9k L2UvZi9nL2gvaS9qL2svbC9tL24vby9wL3Evci9zL3QvdS92L3cveC95L3op Ci9Gb250RmlsZTMgMTAgMCBSPj4KZW5kb2JqCjEwIDAgb2JqCjw8L1N1YnR5 cGUvVHlwZTFDL0ZpbHRlci9GbGF0ZURlY29kZS9MZW5ndGggMTkgMCBSPj5z dHJlYW0KeJydeXlYU9fW94mQnFOrWIkpRGyStlpb56FWsM7iTBWQGREZwhiS QAKEkIQECNNmDFMIkDDIjMyDoNY6oNg6z9oBbWtvp9uW3r7fPtzN+73fDrz3 9j7PvX99z8MfIfvstfb5rbV+67d2GITtHILBYLAP7Trg7ee60is6Tihb7SmJ CxFbv15JOzHoJXPoN2zQVnT676q/r2O+QbjV2synefOIuiW8FfZTrQunli+g X742a6iA0cIYYUy96+3p+97Klav2SKQpCdGRUXLBehcXF0FoiuB/VwSuQll0 pFjwDv6QJBRJpHFCsXyN4JhQKJBHCQUR0SKhYM9Rd/+DR/YL3t1/xFuwXygW JoSIBO6JoaLoMIFbdJhQLBO+J4iQJAhEs/8IwiTi8Gh5tEQsWyPYJROECGRS YVg03iRUhAml1oVVAqkwIS5aJsOfBdEyQWRCiFguDBfIJYJocZgoMdzqHn8f IRHLBdIECV6PwyvYlLtEJpeFJURL5QLs0d113+wZ5VEhcqtfWTReFkgi8JPh krBE69v8c00eEi2WCeRChdzqJ1QoCI+WSUUhKdgvNiVNiJ45QqIsWhz5p/dV ggRhZEhCuEgom7FrReXP9xP8y1uHSKWilJm9kpmn/uk/Wi4TiiLWHImOC02U CWZiKjgicRG4CTyFkYmikIR/X/kzXv9/ESQIYqfbrhRx2Ee7lZLwPVLhUdf4 CPe9CZEe+2RRnvLoYwcSY7wOJsV6H0oWhRxWxIW+/97OTR+sFGxe5bzaZc2W des3bHyXIFYTbxEBxFHClVhDvE0EEu7EXmIp4UHsI9YRywhPYj+xnniHOEYc IDYQywkv4iCxkfAmDhHvE+8RPsRhYhOxgvAl3IgPiJWEH/ERsZtYRfgTR4g9 BIeYS7xJvE44EPMIBuFIzCe4hB2xmFhAOBELCXuCTSwiXiF24UwmbPGWUcYW Rucc2zlSG7aNj02D7Zu2lUwuU8H8gyVmfUfuIc9S+6imV1Rz589NfdX+1Zx5 9vMS592fL5//o12g3Z0FbyzwXXDmtaOvXVhILqxd+MxexZ7LLlvkuqhu0Rec Nzm/vy56/XMHwsHGYa2D2GHSkemY4FjgeIn7NjeOe3nx4cXFsNCOhsCs6KN/ M9t/PenAltJ34W6OUqiWq5RR4V4pzoA6wNqtKeni03EkuFdV9ayYMqMuBflQ b1SDvdS0O8kuBnvVqt1ZFNtdQXeRZ+Fupt1UBLBAvwH6TQtjysfMOVGoK496 iAroG47JLBQ+nXYYvZOWGpjLVcAAMxmYbGypLKwqNvJH4CtMaGINrKyUGuKB mgsSlBk+eZQClpLTGljOgSRsgRRqYdrROUg8QY9bGLSE9uEgJxWrozo9NTtP n5PJX4FaEQlL9FXZRlDBBbUmw1ABZUKharIpu0ZbqoQslOdYmGGKa/GvjgQq LlDJ3ZRx6vhEhRroQXqBrlhhUBYpASVPTpbFdCQM3jp7GS69yIdr6FXGjory xkIu9g9M9OJLjM8moOuETTX9FqeksDy/AlAtZaoAPsongadW7ZdNqWCUifQr UlWCbgqeJs+eqa/sBNS5RtFBPhKSwE2j9Lc+FG4iA4rTKsE5CqaQPwRf3hUk Uh7ez3tGZhYEpERrvVKxR3fdfZrZygBPbOi34DpO4N5ImSeg1h3+HJK/X7r3 RbdRG1HOL0ksldSm1AFuU7254fLe0c0BfsmRoXz/YMl+sJVCrz9bA23O9Fe3 d/OaG83NbVcpu7/PB+YpWzPj52G6dtJm6gP6LmfanwTOaZqtWRj5TjO5u1hd Be5RtDtOAKPxYQlOgEASxU8XoFS6gIkCFKwvs4xpYDc1vYIES9UKZz3el2Ym txcpG8BvFL2chEXTo8wqlt3UWpxrPfTzHoXZ/qdJmDW5BZIO7BU1Uys4bJOC nMiq1IJd1pwCuzQa5xyKHaOAX5NPwb26ntPdHZYz4CwYTukXtcY1SysPVV8w 1pU31lFs0al6U8/AYmi7YQzt4rmgbzk/g5GsbhXefkXWHOa2GHgnhYfJY1NC Mw4CKkBd1sGHA7YY6DpDmYk3zGKLIDkeuGGbT/BqPoYZtNGL2xj0B9CJk1Kk 89enapRctTI63Q9QiMeC9fR7TBP6lcxBc5m1rLIHJtM1QEEeC5mn32Oq6Pkk zg1dC72hhQFewqqXNrCQPslBr69biXjI6ft3oD20/+m/IA86rP0RLeJrJJyJ qyvRG4h5Yv/O8IjG/iS+5ELqXQz3j6PX7/Gsti6AanoutvYCJk/YwJ91HLj8 ArOaTMnLBTmAisyoGOHTn5FZBYHKKN1hJTeVLM0vAaWA6inThfLROlLYIa0I w2cn1yAmegstevQ+nHO51zJ6mn+YhAxbkJmZptOmJCfoxIBy9vocLoDMcw8f f3LuuA8fB0w9TP/ezwCQspkicKWlwyRmEkuflJqWDNKBqlBVElDtX3ocuABn 0Udeez8SrgFoDljTtuljz2uH/iL8BUAm+GVk/AsqpXaX68E4Z8B1A8csJ3u9 zoleAmhDwYPfw1fgsisXksL6eR2xJon5EGUt7rgaZW0KjNN10HbX7O9OwMQX DuysMvoNTlm+AZQB6rRBd5w/HUUCX63WXU+x31HCLJJ9opbcma+pAJ9ScD3Z FdOoG8ShmfsXaAuXwdf2frviWHCCdzD/Cakv8E6K0ngpuXDFTk7b5e6Rh+d2 IArZBOzcc9yzu41nLbY7UNwByW4GvQo6cwJ3nBQfBp7gZKv4SkpvZmfuOAXH WRk3dK0JHXHdQXW+wB8cTxFGHw+VugIXnCf3N0Hyx0dXITHIQ0/hKk7F09M9 42AcNEqM71OYHtXd9P/txrCy6O2QZQP/e8qZg46hJWgpClrdse6iB3/M7dvw b/Dp/eESKIBBv4ife13ju93Z3L0Sx9x2BIykdMdeDGhfDdCrwF2zL9k3PUGZ IJdHRZ1I8QFB4ERtVLvvSMxvGGMw2fLz4CjVN3Su/hKw+v6zBr+ZpFmQ4cDO /xtkcMCeNM0OzOIiBewhIQX+6Lx++cbVnm/Bz+Bl3FOvS/vvIEY3egNnUgOL vURBPsssS8fVyhZNdnLaK0oHXtxIjT7Le+o9sAkgAqC5oZu3HPcTe6S9D6iw NFxzM2GtUtdAz3HavsUe53PqhANbBafpbZyg5riycGx54UpkizGwv7nhh7FL TWeH+QhzwbEUlSs+1zkl1FaRXpUZpWCQon/UkrvRVxwgy9boMhIVMWkiIALS 8mRLbGvSDXATh/2Th8+vt8cc4NnBF5irl3fAkBbGV7iCrlsbxgJIscbM7b11 VTpxLa86vlJXA6gGc21Dd3DT0Y98pEHx/PjgjMi8DymX2C/RLRbYrU3bY2Vq o4l0KdRWgEcUvAXtcLF/Asywvh+et+I5MQlTrXiK4Ak4zGnKrclrAk/BxfKh xmsd3ePgJhhU90pbw4bfb1mNX7aGxa7BjTSjXAf2UNN7SbA5Q7cjc5Zztxaq qsADDC6dZNtsMDRDp+vHl65w90dksjq/QM6bbds9cGgAlvUqZnu3nP4BbuNI WP6IOClDbwNKygoAIcWZhqzC3ApQQ0EA15BfApPG6Ew1QJWclRmardKqNUpJ xnF8nDUwgdVSUtLMg04sXLX1qpr11ClWDnybOd3CYuf/a8v/R8N3VdA95Cy+ 2yyMexM2Bnoxp6qwLL8SUKfKUoP5qJAE/pmZh9Mwcpkmcn+h0gSuUDCR7Bg6 31xSlqWr4FWrK/RGQDWaalo6kyzRWCu7BvD/gvnMJzFS467gzrT6K3CjBTqc gjF9xZgJ78LUJ80tNnQM7cjp204mYx5KNiSWOkrLpcXxgFqza886t2bhzRi+ IiVFqc+KEWuSQTJIqkwbSHb1lcaACMrnafCPj542DV7g9bbXdIBecP344K5C xClx9ClIKAPVwNLZ1VxalmfIKwTmXGNeCegB/ab25s4WUxc4D1qzGtQdFMqc XszpzPxcfw5Q33VqIj12hCPmRo/W4QaDcfA8vxh+z7na0jXS3qSMq+bVRlUG gXDqmDgq6GjIjZ94uA6nFv3JAFMeaCMHsQo/HN7yx21jd81I/8Obvd/NMGQQ WgKXYnKYpYfg99pcrh7m9YV0JN+TjXo53t4K55+EG4EJs2JFPlXCQi+nxjhD YETRJboY2LESICZ4V7wsLCgi1FfuBQJBcE1kh89o7K8AvgJ+av3rwMjA8Gjd GCaGmUwe/5d8iqJlMJWDE4VM+zBF9SFOkWXWFCm2psiS/5AiT/5MEZzMzlor m8wk8/YiLCDuU1AMl5FfgBpV5RbsjaurT7PQayyaentwHbrcdWCvgrfhLU42 XMLUsnJz0zPz8rIBVw90Bdpiil1TmphYFO+01dNvz7GOsOeB/LGYruQaGRBx g2MlftEiY72Cl3QqvUFzhUpF64wsQ3lRAc5CtlsZqMptwB2iRt/SnF3r9OSz sbvnZZ1Hu/mI+ExSrToFmrn97S0jV9qiNzTwZnWd00WrrjuCuWI9/IkzONJW 0weo86ci9/JRAAn26zQe+llh512uLwV3KBhI/hFycYt7iOxoEA9eIbMKo9MS Mo6lcqUZzFSyKL8QFANqqDTjJH86FteOPt0jA+8XmUiPEl0l3j+ji+jfrQzy fBJqrAzyZGq5mSMsTC4Xn0ahsMTxlrGtsrO5o8k8CEbBkKYnvlU44NxkZZFW FvuMgnyUUa4FrrPA69J2WhVYGwa+WGO0Au9PXi6A6WicCTex2H7TgfQLTktF aT9cOHZytX9SSKSCFxQfpV+ZS9nBdt0Fel6HPYBZ6CNMz89hKaZL9p13MF+W 5xXqeZnp+vRMndDrZGBqNsV+rs3UZ2YtBjn5uYW5FPvOJ+gia3Pbyavnehsv dPI0FUkJSl0i4Eaomj/lwz6S/fw3FnaBIX7V1GWB8yftr09+gL38AfdYRdJa L6YKq5fKAmN+GeA2zhCIigR7dNp9VurVmciDxRnl1heCJJv2g0kcqVwukdTL W1ob6ltb5A1i3GQ+UXcnWmg0EG2xb30B1b85sJfQx3B7TWJly7VaJdABrO1L KXZnVUREaazTSRCsjBFHiVJCgC/4cPgYtHW/L7wQekpYpilOAjLqgH/Q7g1u l6CdP0/OakNRzHoW282cX1llcsKcmm3OxJZUvf0Z/U6Q/PbZ71963UCO9/hv jYUMg4vUpcHe62cGFZFdvPbYqvhqd1wsM/0XnhuGPWbGN5Pw8qQNFELEgT4k uF9R8cSAxXW7gnySWaH9R1PQpW2z1lG7mdxZgsN5j4Kh8DLy/U8rVon+71aQ z4wmHfO00Ftq1FX21yfgvgkrLLLznCxWkBYLSPgT2kCyO5Xk9ZxKHfCnUBoJ JFlqtVaaKEqLA1R47OlBPrsGPkPvqMmzOVUqEEShYnLrWb8HPX3Vzc28oSGm M8l2K8o9b+mu+LiKi6OMxKenbGeCDD+YifL30JsO4mQVusuFGvcULFSL800F pvxywG0oU+LaWECCHTkJ2mS9Tp2hBiqgMyjrUhrUZal4EktRJsb2yy/COVcf w7f47N/8pnb8p+Djpkj/dZABdRhWwxTBmc4hwQdarbMVpjNm0qVIbcRdlS69 SEIWqE9pQHOoRpYebmdOj7P0aDuzkdUIGQ1maIv5ELe12cEI2yKt+minJm1z DjZ0HhsqTLW2ZyveD6qME0UY7/MK8vns/ILBVlvorXg2jbdKDT6eTXNMOqAA ulxdtu5dVOm4DJbpq3KqQBmeTisMXYV4OvVTkb25Bt3IUjgP9TsW6AszCzNK UkoySkAJMBhNPXARfOjYervE0FZIzbKUs4nxCHfcKRsOLheMItVZrgnkTydh TZyTs8facTUmMjk/rAAriS4udCOhAB0tzC7IKcgpzOGW6gr1IItKT9en8RQq KDaxggtF5pCP0Wro49jafO/m9ZHWUq6pqKoAC6NSEzqpZvXlVWQDNchIV6vT 8MRi50g7k9kFwjRxhkcqNy7DOvXjaWHKzHg5Cf972IaGdCUHTFZWf1mO4VEp yKf66kSwHMtHzKGZwtQ4aeBh0U7gCgIapaMpHZk9Vtn+YCaBP7cmcDeWmHqc wLspdINEu75wgdsh41HX7T5e42DZCHhi7SHqXvp/zAy4BYeonT7PAQ8qKr6w zqf9CvLL2b3TYeSJtKR4tb70VCZP15VajQWCNFEuDRmIG4PE+H24kG8992y8 buN4qcjTmeWZQAlUeplWuh4lOG6GcuUIAMDIBfUVhlEDjlawiuzJrtRavA0K g6IkZR0KdHwXVmWX55aDci6oqyw9XYSfOqAiG3JLdfV+0An96FgZV6ouwZ2g 1FjZhCeSTMfvkLYqokhfArgGUFxW3Y+Hwl8cawcLiy3YBTnDk4xnEzZmHOES VkV+ZT4e6jrK1XjuiSeBt1rjloVjrDWRbkVp5eASRf+Ao+GrFOr2KWduC654 jtGOjfbmr+C7Xziwz9BHbnKU6rQsFa6mzKphPsQMciGrVdWY2H3S7A6o1bsC 3OSW5IbGOkt9UV5xnoGfW5ZXCgxUc0d9/7lmsTfvCInWfKTMOClMTFZHgxiK 7ec2HDw+PFB3YZzHNvoW1yYNOJ3GL9/Vi5fQWizORVnpSWkytSQ9CVBRktND /CISXB3ohvOHZ+cQa3m9nLSh1085cCrzykERaMutzMBaLVEe5+WH9qJNjjCE /GdUe6w5VKn5R5/Tzt5QdJnJbda7kwcUPEo21f3XW493XnQFMi5QqzMleVlA C9KsV0hjZtI3X12+7mekhoWO0BO6375RXXq7mGtGJis/Agu90cSA7a/DIBNL BJKrRcNoEfRz/ArOMze3VXfkc01og4psyy3X4yJIS0tMEGXp01I0yWqLo2Qw 5TQwgiqjoS0fxz1MVUa2gWpN+/Ef0FJHxEIucaG6zJgcrgouM5GStGJjicFk tPCfwQXfIUGRvkAP9FyQkpkdgzvwlBnX9hsXGHjodbP+2ZjptzjGwlLr5VJT WWogH5VYo686loejn2AifQpUFaCPgl0kKCgtNxhONQzVdAFquD4Ky5U4PD1p tZ7W7ik3kZ75qhpwnYLFZNtQZ+0IoK7USp35KJIEnhkZHpn4IYmJjC+IK9DU gj4uDCXh3PBr2z384j28eSlXopu9QTCQqJ3dqEdkZoGfIlrrpbLKczykv/op 4+YETP7GBq6e2sbJKghRSTOOa7hiq/4pyS8CBkD1l6WH8KerSdHzmM+h4ySc A9+GC3b8sPyIR2RAMs/Htm+0+8LT89uQHbIJPOzie6Lh1Iwas8Dln+Hhwh52 Pt3+BPcQWoQ16TLaI5iFbk17MA/D0+hDkv3bE/jhJAsd+IyDMlkwE/7ExPQ9 tdTCgNtf2PSivZyHZ9uaekbufd01o3vhooifN06Gno2v9+1IS3X83K05qmVn 44GKHWA5eFvjLD0icYuL3ONB2f19tboPzu2MHYIFFujRbg8m4Px79Px+B/bE lOsDzrYoP80BrMG2Izv4Gtz9ffcPV67yz5271fwAT+7r4etoAXpvuduKg/Ea U7shvzi/hH8DGpj7WDph4oCqIgpwT4Ko1OhoHx/RDkxV7OVufQHnhkebz5zm sW8Hl/SIzzr98hzaQA50QW/DRegjFIw+QGuRL/KC76N3Ydjd88bhn/ns5Wgb 7cJpyTXJcsNzA5Xu8ccVkeJEKaAi5J0XLtaeP93Lb+nosjTmzrSmHvrKMwZ8 etcGHqQ/50A/VjUoKig09H7rmG5QJemykgAXCVmIgXR5uSAP5HKzinJKO/oz NQM8SEZ88hF4l1p5ZOe2ZKWhPo4X2yAvx8UtVqclCEdljybuN398nj9ypu4K eAzuK0b8zwcPezaguXUznlto2xEG+B4WvbShPad24nlH56HY7YeI7e6IBdBi 8H7zxiH/3uCLsusYv9de/IgZcdHuLxAZnJQpdOM3w5WYNwm4volCoaiP8+L8 HrQQzQk/snej9w9w7celLRYTv666vazLOs2M49Q5fpMBJTdtYAHs4yDJQShB 0ptQir9CEig9iKVoIT5WhoV+8JBBe0+t4EyzWO5oeU3G0O1a4yXAhXEsFDD9 PfMGC0ZNvYVTqlBvSf2b/a99DuyTX6E3Obc6z4yDa9T5qL4TUTJ5NI9dF9ug 6C4x5OeX8NgnCwwA5DuVlugzYiUR3hH8GQSm7FrsweO9D6HkjtdjLPPhEE1y fvmkaxDcpr7cc205Ymw5sO6QWfzHWh77W0S4JIr2L17xbD9kwnlPbv7+JPoc mv+Sl7KU88C3MRIcow6dOHFoh9/Y15/VD46N8thPzj3hbBiLHAGXqbEzA5/e Hgre5yk54SnkZeUAkJ1j1TTQ51M8dObZ9MFyjhvM+9RKfVA3RNviufovNlcT OdDWeMl8+9zfHo/D1wB0oqDLWvgK4qBXN7yDnJDd/S2QuDRYPfAJLxRtRjYC tCqeglfhFAckZ6dr0yWyCE0koHYEPoKvXqm+VtfEr7GcqmgF1NdnNqGtfAyh Fhflwl85cQnxIvGphNb2plOt7fFNcXipXYtzPR/aMOASOvjfH0DI9t83YUyb puybGOAnaPzJqvmcOV4gRBkR5XFMiEiAXgGI0YleGT3WE/yxbAx8DIZqurvG LvVCCsD5FAxZB99EDryceM7EGcREQiQMFmzaFPx/YDiMOgNtXvBnpdYnVfcs jMkXNnDYek93YB1TTVYWVxQVFdQaaoqw8qorU8XyrbfmMRny1HRVujonxMrO rSZya35qOThL0T/PtJjHdMNDBhx/YUNL4BscTXGEpyb9BC63DBbcCl2//b7v MXjK/XXLk7e9fBPDInix0apYpWtdlmP/X7s77gBq4orn5m1BazZv4KN9yIOZ Ri+2WlV3099ds4etz4T3Hdh34Gr4FecqaMhtTL+e2OcJ9lGbPQ/tkiYbGiN4 cbVKg/VnDZU6KWoo8e6TZ00Do/zRgaYxcAt8nDoqbk+uV1QGm/FYd2f0VM/F xV8cvrzWP1gZG8mLk6riE4+ash0HHpzp/BRQN84Eu0WqIhMS+HFxEtV+GfW/ dAswRmcxRrsC4+Ijglzfj1o+EwJ2z3vfLBsIbEq8EFtZ7fjhp5Iu6SPZLe1j 8Cv4w/hl8/WW623d98cou2QL7WOBXhZW29znr7aVzpv3vHbefIL4f6IdczwK ZW5kc3RyZWFtCmVuZG9iagoxOSAwIG9iago2ODMzCmVuZG9iagoxMiAwIG9i ago8PC9TdWJ0eXBlL1R5cGUxL0Jhc2VGb250L0pBSFVYRCtUaW1lcy1Sb21h bi9UeXBlL0ZvbnQvTmFtZS9SMTIvRm9udERlc2NyaXB0b3IgMTEgMCBSL0Zp cnN0Q2hhciAxL0xhc3RDaGFyIDI1NS9XaWR0aHNbIDMzMyAzMzMgMzMzIDI3 OCA1NTYgNTU2IDE2NyAzMzMgNjExIDI3OCA1NjQgMzMzIDMzMyA2MTEgNDQ0 CjI1MCAyNTAgMjUwIDI1MCAyNTAgMjUwIDI1MCAyNTAgMjUwIDI1MCAyNTAg MjUwIDI1MCAyNTAgMjUwIDI1MAoyNTAgMzMzIDQwOCA1MDAgNTAwIDgzMyA3 NzggMTgwIDMzMyAzMzMgNTAwIDU2NCAyNTAgMzMzIDI1MCAyNzgKNTAwIDUw MCA1MDAgNTAwIDUwMCA1MDAgNTAwIDUwMCA1MDAgNTAwIDI3OCAyNzggNTY0 IDU2NCA1NjQgNDQ0CjkyMSA3MjIgNjY3IDY2NyA3MjIgNjExIDU1NiA3MjIg NzIyIDMzMyAzODkgNzIyIDYxMSA4ODkgNzIyIDcyMgo1NTYgNzIyIDY2NyA1 NTYgNjExIDcyMiA3MjIgOTQ0IDcyMiA3MjIgNjExIDMzMyAyNzggMzMzIDQ2 OSA1MDAKMzMzIDQ0NCA1MDAgNDQ0IDUwMCA0NDQgMzMzIDUwMCA1MDAgMjc4 IDI3OCA1MDAgMjc4IDc3OCA1MDAgNTAwCjUwMCA1MDAgMzMzIDM4OSAyNzgg NTAwIDUwMCA3MjIgNTAwIDUwMCA0NDQgNDgwIDIwMCA0ODAgNTQxIDI1MAo1 MDAgMjUwIDMzMyA1MDAgNDQ0IDEwMDAgNTAwIDUwMCAzMzMgMTAwMCA1NTYg MzMzIDg4OSAyNTAgMjUwIDI1MAoyNTAgMzMzIDMzMyA0NDQgNDQ0IDM1MCA1 MDAgMTAwMCAzMzMgOTgwIDM4OSAzMzMgNzIyIDI1MCAyNTAgNzIyCjI1MCAz MzMgNTAwIDUwMCA1MDAgNTAwIDIwMCA1MDAgMzMzIDc2MCAyNzYgNTAwIDU2 NCAyNTAgNzYwIDMzMwo0MDAgNTY0IDMwMCAzMDAgMzMzIDUwMCA0NTMgMjUw IDMzMyAzMDAgMzEwIDUwMCA3NTAgNzUwIDc1MCA0NDQKNzIyIDcyMiA3MjIg NzIyIDcyMiA3MjIgODg5IDY2NyA2MTEgNjExIDYxMSA2MTEgMzMzIDMzMyAz MzMgMzMzCjcyMiA3MjIgNzIyIDcyMiA3MjIgNzIyIDcyMiA1NjQgNzIyIDcy MiA3MjIgNzIyIDcyMiA3MjIgNTU2IDUwMAo0NDQgNDQ0IDQ0NCA0NDQgNDQ0 IDQ0NCA2NjcgNDQ0IDQ0NCA0NDQgNDQ0IDQ0NCAyNzggMjc4IDI3OCAyNzgK NTAwIDUwMCA1MDAgNTAwIDUwMCA1MDAgNTAwIDU2NCA1MDAgNTAwIDUwMCA1 MDAgNTAwIDUwMCA1MDAgNTAwXQo+PgplbmRvYmoKMTUgMCBvYmoKPDwvU3Vi dHlwZS9UeXBlMS9CYXNlRm9udC9SVEZUWEQrVGltZXMtQm9sZC9UeXBlL0Zv bnQvTmFtZS9SMTUvRm9udERlc2NyaXB0b3IgMTQgMCBSL0ZpcnN0Q2hhciAx L0xhc3RDaGFyIDI1NS9XaWR0aHNbIDMzMyAzMzMgMzMzIDI3OCA1NTYgNTU2 IDE2NyAzMzMgNjY3IDI3OCA1NzAgMzMzIDMzMyA2NjcgNDQ0CjI1MCAyNTAg MjUwIDI1MCAyNTAgMjUwIDI1MCAyNTAgMjUwIDI1MCAyNTAgMjUwIDI1MCAy NTAgMjUwIDI1MAoyNTAgMzMzIDU1NSA1MDAgNTAwIDEwMDAgODMzIDI3OCAz MzMgMzMzIDUwMCA1NzAgMjUwIDMzMyAyNTAgMjc4CjUwMCA1MDAgNTAwIDUw MCA1MDAgNTAwIDUwMCA1MDAgNTAwIDUwMCAzMzMgMzMzIDU3MCA1NzAgNTcw IDUwMAo5MzAgNzIyIDY2NyA3MjIgNzIyIDY2NyA2MTEgNzc4IDc3OCAzODkg NTAwIDc3OCA2NjcgOTQ0IDcyMiA3NzgKNjExIDc3OCA3MjIgNTU2IDY2NyA3 MjIgNzIyIDEwMDAgNzIyIDcyMiA2NjcgMzMzIDI3OCAzMzMgNTgxIDUwMAoz MzMgNTAwIDU1NiA0NDQgNTU2IDQ0NCAzMzMgNTAwIDU1NiAyNzggMzMzIDU1 NiAyNzggODMzIDU1NiA1MDAKNTU2IDU1NiA0NDQgMzg5IDMzMyA1NTYgNTAw IDcyMiA1MDAgNTAwIDQ0NCAzOTQgMjIwIDM5NCA1MjAgMjUwCjUwMCAyNTAg MzMzIDUwMCA1MDAgMTAwMCA1MDAgNTAwIDMzMyAxMDAwIDU1NiAzMzMgMTAw MCAyNTAgMjUwIDI1MAoyNTAgMzMzIDMzMyA1MDAgNTAwIDM1MCA1MDAgMTAw MCAzMzMgMTAwMCAzODkgMzMzIDcyMiAyNTAgMjUwIDcyMgoyNTAgMzMzIDUw MCA1MDAgNTAwIDUwMCAyMjAgNTAwIDMzMyA3NDcgMzAwIDUwMCA1NzAgMjUw IDc0NyAzMzMKNDAwIDU3MCAzMDAgMzAwIDMzMyA1NTYgNTQwIDI1MCAzMzMg MzAwIDMzMCA1MDAgNzUwIDc1MCA3NTAgNTAwCjcyMiA3MjIgNzIyIDcyMiA3 MjIgNzIyIDEwMDAgNzIyIDY2NyA2NjcgNjY3IDY2NyAzODkgMzg5IDM4OSAz ODkKNzIyIDcyMiA3NzggNzc4IDc3OCA3NzggNzc4IDU3MCA3NzggNzIyIDcy MiA3MjIgNzIyIDcyMiA2MTEgNTU2CjUwMCA1MDAgNTAwIDUwMCA1MDAgNTAw IDcyMiA0NDQgNDQ0IDQ0NCA0NDQgNDQ0IDI3OCAyNzggMjc4IDI3OAo1MDAg NTU2IDUwMCA1MDAgNTAwIDUwMCA1MDAgNTcwIDUwMCA1NTYgNTU2IDU1NiA1 NTYgNTAwIDU1NiA1MDBdCj4+CmVuZG9iagoyIDAgb2JqCjw8L1Byb2R1Y2Vy KEdOVSBHaG9zdHNjcmlwdCA3LjA1KQovVGl0bGUoTWljcm9zb2Z0IFdvcmQg LSB0eXBlc2Vzc2lvbi5kb2MpCi9DcmVhdG9yKFBTY3JpcHQ1LmRsbCBWZXJz aW9uIDUuMikKL0NyZWF0aW9uRGF0ZSgxMC8zMS8yMDAzIDEyOjQwOjI3KQov QXV0aG9yKGhhbmkpPj5lbmRvYmoKeHJlZgowIDIwCjAwMDAwMDAwMDAgNjU1 MzUgZiAKMDAwMDAwNDY0OSAwMDAwMCBuIAowMDAwMDE5ODI2IDAwMDAwIG4g CjAwMDAwMDQ1ODEgMDAwMDAgbiAKMDAwMDAwNTAwNiAwMDAwMCBuIAowMDAw MDA0ODU4IDAwMDAwIG4gCjAwMDAwMDQ2OTcgMDAwMDAgbiAKMDAwMDAwNDQy MSAwMDAwMCBuIAowMDAwMDAwMDE1IDAwMDAwIG4gCjAwMDAwMDQ0MDEgMDAw MDAgbiAKMDAwMDAxMDU1OCAwMDAwMCBuIAowMDAwMDEwMTIzIDAwMDAwIG4g CjAwMDAwMTc0OTggMDAwMDAgbiAKMDAwMDAwNTUxOCAwMDAwMCBuIAowMDAw MDA1MTUzIDAwMDAwIG4gCjAwMDAwMTg2NjAgMDAwMDAgbiAKMDAwMDAwNDc4 NSAwMDAwMCBuIAowMDAwMDA0ODE1IDAwMDAwIG4gCjAwMDAwMTAxMDIgMDAw MDAgbiAKMDAwMDAxNzQ3NyAwMDAwMCBuIAp0cmFpbGVyCjw8IC9TaXplIDIw IC9Sb290IDEgMCBSIC9JbmZvIDIgMCBSCj4+CnN0YXJ0eHJlZgoyMDAwMQol JUVPRgo= --Murder_of_Crows_345_000-- From owner-reliable_computing [at] interval [dot] louisiana.edu Tue Dec 2 22:50:26 2003 Received: (from daemon@localhost) by interval.louisiana.edu (8.11.3/8.11.3/ull-interval-math-majordomo-1.3) id hB34oPq16938 for reliable_computing-outgoing; Tue, 2 Dec 2003 22:50:25 -0600 (CST) Received: from cs.utep.edu (mail.cs.utep.edu [129.108.5.3]) by interval.louisiana.edu (8.11.3/8.11.3/ull-interval-math-majordomo-1.3) with ESMTP id hB34oJw16934 for ; Tue, 2 Dec 2003 22:50:20 -0600 (CST) Received: from aragorn (aragorn [129.108.5.35]) by cs.utep.edu (8.11.7/8.11.3) with SMTP id hB34oEL22607; Tue, 2 Dec 2003 21:50:14 -0700 (MST) Message-Id: <200312030450.hB34oEL22607 [at] cs [dot] utep.edu> Date: Tue, 2 Dec 2003 21:50:14 -0700 (MST) From: Vladik Kreinovich Reply-To: Vladik Kreinovich Subject: ISSAC 2004, Call for papers2 To: reliable_computing [at] interval [dot] louisiana.edu, interval [at] cs [dot] utep.edu MIME-Version: 1.0 Content-Type: TEXT/plain; charset=us-ascii Content-MD5: 1uhxw2izLgkds0wjkS+U6g== X-Mailer: dtmail 1.3.0 @(#)CDE Version 1.4 SunOS 5.8 sun4u sparc Sender: owner-reliable_computing [at] interval [dot] louisiana.edu Precedence: bulk forwarding; as always, the organizers would love to see papers relating interval computations with symbolic computations ------------- Begin Forwarded Message ------------- From: "Ibolya Szilagyi" ********************************************************************* CALL FOR PAPERS Second Announcement ISSAC-2004 International Symposium on Symbolic and Algebraic Computation University of Cantabria, Santander, Spain http://www.risc.uni-linz.ac.at/issac2004 ********************************************************************* ISSAC is the yearly premier international symposium in Symbolic and Algebraic Computation. It provides an opportunity to learn of new developments and to present original research results in all areas of symbolic mathematical computation. Planned activities include invited presentations by *Pablo Parrilo, Francisco Santos and Jan Verschelde*, submitted research papers, poster sessions, tutorial courses, vendor exhibits, and software demonstrations. ISSAC 2004 will be held at the University of Cantabria in Santander, Spain, from July 4 to 7, 2004. Paper Submission ---------------- Authors should submit full papers before midnight Eastern Standard Time on 7 January 2004. Papers must describe original, previously unpublished work that has not been simultaneously submitted for publication elsewhere. Students presenting papers of which they are an author will have their registration fee waived. Submission should be done electronically via the web form: https://issac2004.risc.uni-linz.ac.at Important Dates --------------- *Deadline for Submissions: January 7, 2004 *Notification of Acceptance: March 3, 2004 *Camera-ready copy received: April 16, 2004 Conference Topics ----------------- Topics of the meeting include, but are not limited to: -Algorithmic mathematics: Algebraic, symbolic and symbolic-numeric algorithms. Simplification, function manipulation, equations, summation, integration, ODE/PDE, linear algebra, number theory, group and geometric computing. -Computer Science: Theoretical and practical problems in symbolic computation. Systems, problem solving environments, user interfaces, software, libraries, parallel/distributed computing and programming languages for symbolic computation, concrete analysis, benchmarking, theoretical and practical complexity of computer algebra algorithms, automatic differentiation, code generation, mathematical data structures and exchange protocols. -Applications: Problem treatments using algebraic, symbolic or symbolic-numeric computation in an essential or a novel way. Engineering, economics and finance, physical and biological sciences, computer science, logic, mathematics, statistics, education. Instructions to Authors ----------------------- Original research results and insightful analyses of current concerns are solicited for submission. Submissions must not substantially duplicate work published or submitted for publication elsewhere. Late submissions will be rejected. Papers will be reviewed by a program committee and additional referees. Proceedings will be distributed at the symposium. Papers must be in English and should not exceed 8 pages in the standard format for ACM proceedings, or 20 to 22 pages of text in LaTeX 12pt article style. Each paper should have an abstract as well as an introductory section that: -Describes the problem. -Motivates the study of the problem. -States the main results and compares them to other work (including theoretical or empirical performance). -Summarizes the original aspects of the contribution. Formatting requirements will be based on the ACM Proceedings Templates. For instance in LaTeX2e you should use the acm_proc_article-sp.cls document class to format your document. For information concerning ACM Proceedings Templates see www.acm.org/sigs/pubs/proceed/template.html. Authors of accepted papers are expected to present their work at the symposium, or make arrangements to have it presented, and will be required to sign the ACM copyright agreement www.acm.org/pubs/copyright_form.html). Some papers may be accepted for poster-session presentation; these will not appear in the proceedings. Abstracts of all posters will be distributed at the symposium. For poster submissions see www.risc.uni-linz.ac.at/issac2004/cfposter.html. Best student author award ------------------------- This award will be given to the best student author. An author who is a Full-time student at the time of submission is eligible for the award. This status should be indicated when the paper is submitted. For details concerning the selection procedure for this award as well as the procedure for the ISSAC Distinguished Paper Award, see www.acm.org/sigsam/ISSAC_Awards.html. Organizing Committee -------------------- General Chair: Josef Schicho SIGSAM Chair: Emil Volcheck Program Committee Chair: Michael Singer Program Committee : Moulay Barkatou Arjeh M. Cohen Shuhong Gao Karin Gatermann Patricia Gianni Serkan Hosten Richard Liska Bernard Mourrain Masayuki Noro Marie-Francoise Roy Bruno Salvy Rafael Sendra Gilles Villard Dongming Wang Stephen Watt Poster Committee: Howard Cheng Claude-Pierre Jeannerod Ilias Kotsireas Ha Le Wen-shin Lee Ziming Li (chair) Tutorials: Thomas Sturm Proceedings Editor: Jaime Gutierrez Local Arrangements: Luis M. Pardo Treasurer: Carsten Schneider Publicity Committee: Tobias Beck Janka Pilnikova Ibolya Szilagyi Web/Online Registration: Gabor Bodnar ===================================================================== Please send all the comments to the appropriate contact: issac2004 [at] risc [dot] uni-linz.ac.at ------------- End Forwarded Message ------------- From owner-reliable_computing [at] interval [dot] louisiana.edu Tue Dec 2 23:22:41 2003 Received: (from daemon@localhost) by interval.louisiana.edu (8.11.3/8.11.3/ull-interval-math-majordomo-1.3) id hB35MfB17112 for reliable_computing-outgoing; Tue, 2 Dec 2003 23:22:41 -0600 (CST) Received: from cs.utep.edu (mail.cs.utep.edu [129.108.5.3]) by interval.louisiana.edu (8.11.3/8.11.3/ull-interval-math-majordomo-1.3) with ESMTP id hB35Maw17108 for ; Tue, 2 Dec 2003 23:22:36 -0600 (CST) Received: from aragorn (aragorn [129.108.5.35]) by cs.utep.edu (8.11.7/8.11.3) with SMTP id hB35MUR22766; Tue, 2 Dec 2003 22:22:30 -0700 (MST) Message-Id: <200312030522.hB35MUR22766 [at] cs [dot] utep.edu> Date: Tue, 2 Dec 2003 22:22:30 -0700 (MST) From: Vladik Kreinovich Reply-To: Vladik Kreinovich Subject: application of interval techniques to diasgnosis and fault detection To: reliable_computing [at] interval [dot] louisiana.edu, interval [at] cs [dot] utep.edu Cc: Jose.Ragot [at] ensem [dot] inpl-nancy.fr MIME-Version: 1.0 Content-Type: TEXT/plain; charset=ISO-8859-1 Content-MD5: QwUgaCYgDwizRW11b8G2yg== X-Mailer: dtmail 1.3.0 @(#)CDE Version 1.4 SunOS 5.8 sun4u sparc Content-Transfer-Encoding: 8bit X-MIME-Autoconverted: from QUOTED-PRINTABLE to 8bit by interval.louisiana.edu id hB35Mbw17109 Sender: owner-reliable_computing [at] interval [dot] louisiana.edu Precedence: bulk Dear Friends, This is FYI. I have just learned about an exciting activity performed by Professor Ragot from Nancy, France. ******************************************************************** From: José Ragot ... we are doing some work about diagnosis and fault detection using interval computation. Enclosed you will find some publications in this field. best regards Prof. José RAGOT -- José RAGOT.................................................................. CRAN - 2, Avenue de la Forêt de Haye.................................. Institut National Polytechnique de Lorraine, ENSG............................ 54 516 Vandoeuvre les Nancy, France.................................. ............................................................................. Tel. (INPL) : (33) 03 83.59.59.59 Tel.(direct) : (33) 03 83 59 56 82 Secrétariat : (33) 03 83 59 56 84 Mobile : 06 07 34 51 37 Fax : (33) 03 83.59.56.44 .................................. Email : jragot [at] ensem [dot] inpl-nancy.fr .................................. jragot [at] ensg [dot] inpl-nancy.fr .................................. http://www.ensem.inpl-nancy.fr/~jragot/ ****************************************************************** ADROT P. PLOIX S., RAGOT J. Caractérisation des incertitudes dans un modèle linéaire statique. Journal européen des systèmes automatisés, 36 (6), p. 799-824, 2002. MAQUIN D., ADROT O., RAGOT J. Estimation d'état de systèmes incertains. Application à la détection de défauts de capteurs. Troisième Conférence Internationale sur l'Automatisation Industrielle, Montréal, Canada, 7-9 juin 1999. RAGOT J., MAQUIN D., ADROT O. LMI approach for data reconciliation. 38th Conference of Metallurgists, Symposium Optimization and Control in Minerals, Metals and Materials Processing, Quebec, Canada, August 22-26, 1999. ADROT O., MAQUIN D., RAGOT J. Fault detection with model parameter structured uncertainties. European Control Conference, ECC'99, Karlsruhe, Germany, August 31, September 3, 1999. PLOIX S., ADROT O., RAGOT J. Une nouvelle approche pour la caractérisation des incertitudes : l'approche bornante. Chimiométrie'99, "De la mesure à l'information", Paris, France, 2-3 décembre 1999. PLOIX S., ADROT O., RAGOT J. Parameter uncertainty computation in static linear models. 38th IEEE Conference on Decision and Control, CDC'99, Phoenix, Arizona, USA, Decem ber 7-10, 1999. ADROT O., MAQUIN D., RAGOT J. Bounding approach to the fault detection of a class of uncertain dynamic systems. IFAC Symposium on Fault detection, Safety and Supervision of Technical Processes, SAFEPROCESS'2000, Budapest, Hungary, June 14-16, 2000. PLOIX S., ADROT O., RAGOT J. Bounding approach to the diagnosis of a class of uncertain static systems. IFAC Symposium on Fault detection, Safety and Supervision of Technical Processes, SAFEPROCESS'2000, Budapest, Hungary, June 14-16, 2000. Adrot O., Maquin D., Ragot J. Diagnosis of an uncertain static system. 39th IEEE Conference on Decision and Control, Darling Harbour, Australia, December 12-15, 2000. JANATI IDRISSI H., ADROT O., RAGOT J. Fault detection of uncertain models using polytope projection. 15th IFAC World congrss, Barcelone, 2002. JANATI IDRISSI H., ADROT O., RAGOT J. Détection de multi-défauts pour un modèle à paramètres incertains. Conférence Internationale Francophone d'Automatique, CIFA'2002, Nantes, France, 8-10 juillet 2002. JANATI IDRISSI H., KRATZ F., RAGOT J. Fault detection and isolation for uncertain system. IEEE Conference on Decision and Control, CDC'2002, Las Vegas, Nevada, USA, December 10 - 13, 2002. ADROT O., PLOIX S., RAGOT J. Détection de défauts appliquée à des modèles statiques affectés par des incertitudes structurées. 17ème Colloque GRETSI'99 sur le traitement du Signal et des Images, Vannes, France, 13-17 septembre 1999. JANAI IDRISSI H., RAGOT J. La prise en compte d'incertitudes bornées. Application à la validation de mesures. Colloque C2I, 31 janvier-1 février 2001, Paris. JANATI IDRISSI H., ADROT O., MAQUIN D., RAGOT J. Analyse d'un système de mesure à ;paramètres incertains. GRETSI, 2001 JANATI IDRISSI J., ADROT O., MAQUIN D., RAGOT J. Analyse d'un système de mesure à paramètres incertains. 18ème colloque sur le traitement du signal et des images, GRETSI, Toulpouse, 10-13 septembre 2001. ADROT O., JANATI IDRISSI H., MAQUIN D. Fault detection based on interval analysis. 15th IFAC World Congress on Automatic Control, Barcelona, Spain, July 21-26, 2002. JANATI IDRISSI H., ADROT O., RAGOT J., KRATZ F. Fault detection of uncertain models using polytope projection. 15th IFAC World Congress on Automatic Control, Barcelona, Spain, July 21-26, 2002. JANATI IDRISSI H., ADROT O., RAGOT J. Détection de multi-défauts pour un modèle à paramètres incertains. Conférence Internationale Francophone d'Automatique, CIFA'2002, Nantes, France, 8-10 juillet 2002. JANATI IDRISSI H., KRATZ F., RAGOT J. Fault detection and isolation for uncertain systems. IEEE Conference on Decision and Control, CDC'2002, Las Vegas, Nevada, USA, December 10 - 13, 2002. CHERRIER E., BOUTAYEB M., RAGOT J. Estimation des bornes de l'état d'un système incertain. Approche par intervalles. Journées Doctorales d'Automatique, Valenciennes, France, 25-27 juin 2003. From owner-reliable_computing [at] interval [dot] louisiana.edu Wed Dec 3 08:05:36 2003 Received: (from daemon@localhost) by interval.louisiana.edu (8.11.3/8.11.3/ull-interval-math-majordomo-1.3) id hB3E5ac17939 for reliable_computing-outgoing; Wed, 3 Dec 2003 08:05:36 -0600 (CST) Received: from fe3.cox-internet.com (fe3-cox.cox-internet.com [66.76.2.40]) by interval.louisiana.edu (8.11.3/8.11.3/ull-interval-math-majordomo-1.3) with ESMTP id hB3E5Sw17935 for ; Wed, 3 Dec 2003 08:05:29 -0600 (CST) Received: from Inspiron-8200 ([68.226.139.111]) by fe3.cox-internet.com (InterMail vK.4.04.00.03 201-232-140-20030416 license f018ea6efd6984189790b5f401fab223) with SMTP id <20031203140522.QDGN636.fe3@Inspiron-8200>; Wed, 3 Dec 2003 08:05:22 -0600 Message-Id: <2.2.32.20031203140443.009f68a4 [at] pop [dot] louisiana.edu> X-Sender: rbk5287 [at] pop [dot] louisiana.edu X-Mailer: Windows Eudora Pro Version 2.2 (32) Mime-Version: 1.0 Content-Type: text/plain; charset="us-ascii" Date: Wed, 03 Dec 2003 08:04:43 -0600 To: Marc van Woerkom From: "R. Baker Kearfott" Subject: Re: Another New Global Optimization Complexity Paper ... Cc: reliable_computing [at] interval [dot] louisiana.edu Sender: owner-reliable_computing [at] interval [dot] louisiana.edu Precedence: bulk Marc, In my view, "interval arithmetic" is a basic tool in global optimization, that can be used in many specific ways. It is not just for high precision, but it also will not be successful if universally applied in a naive way in all contexts. Global optimization algorithms will continue to improve as the contexts in which existing techniques are applied are made more appropriate. Interval arithmetic, not just for high precision, also provides speed benefits (validated or not) in various contexts. Best regards, Baker At 05:27 AM 12/3/2003 +0100, Marc van Woerkom wrote: >Hello! > >When should one stick to interval arithmetic global optimization? > >Is it just to work with high precision, thus other methods >might find no good solution, because of numerical errors? > >Or is it fast as well or provides some other benefits in comparison >to other methods which would also come up with valid results? > >Regards, >Marc > > --------------------------------------------------------------- R. Baker Kearfott, rbk [at] louisiana [dot] edu (337) 482-5346 (fax) (337) 482-5270 (work) (337) 993-1827 (home) URL: http://interval.louisiana.edu/kearfott.html Department of Mathematics, University of Louisiana at Lafayette Box 4-1010, Lafayette, LA 70504-1010, USA --------------------------------------------------------------- From owner-reliable_computing [at] interval [dot] louisiana.edu Wed Dec 3 22:41:36 2003 Received: (from daemon@localhost) by interval.louisiana.edu (8.11.3/8.11.3/ull-interval-math-majordomo-1.3) id hB44faK18695 for reliable_computing-outgoing; Wed, 3 Dec 2003 22:41:36 -0600 (CST) Received: from cs.utep.edu (mail.cs.utep.edu [129.108.5.3]) by interval.louisiana.edu (8.11.3/8.11.3/ull-interval-math-majordomo-1.3) with ESMTP id hB44fPw18691 for ; Wed, 3 Dec 2003 22:41:26 -0600 (CST) Received: from aragorn (aragorn [129.108.5.35]) by cs.utep.edu (8.11.7/8.11.3) with SMTP id hB44f8101608; Wed, 3 Dec 2003 21:41:09 -0700 (MST) Message-Id: <200312040441.hB44f8101608 [at] cs [dot] utep.edu> Date: Wed, 3 Dec 2003 21:41:09 -0700 (MST) From: Vladik Kreinovich Reply-To: Vladik Kreinovich Subject: Sabbatical in Spain? forwarding To: reliable_computing [at] interval [dot] louisiana.edu, interval [at] cs [dot] utep.edu MIME-Version: 1.0 Content-Type: TEXT/plain; charset=us-ascii Content-MD5: w50Nl9GfvQlC+FTkRzj1Bg== X-Mailer: dtmail 1.3.0 @(#)CDE Version 1.4 SunOS 5.8 sun4u sparc Sender: owner-reliable_computing [at] interval [dot] louisiana.edu Precedence: bulk Opportunity to closely colleaborate with our friends from Spain and Catalonia. ------------- Begin Forwarded Message ------------- Mime-Version: 1.0 X-Sender: bernardo [at] post [dot] uv.es Dear colleagues, The Spanish Government has issued a call for visits that may be of interest for spending a sabbatical here. Two modalities: - PhD obtained before 01/01/98. The stay should last between 3 and 12 months. - PhD obtained after 01/01/98. The stay should last between 9 and 18 months. Check http://www.univ.mecd.es Two deadlines: * Before November 5th, if starting date between April 1st and October 1st 2004. * March 31st, if starting date between October 1st and March 1st 2005. All the best, Jose -- __________________________________________________ Prof. Dr. Jose M. Bernardo Dept. Estadistica, Univ. Valencia Fac Matematicas, 46100-Burjassot, Valencia, Spain Tel and Fax +34 96 364 3560 (direct) Tel +34 96 354 4362, Fax +34 96 354 4735 (office) , www.uv.es/~bernardo __________________________________________________ ------------- End Forwarded Message ------------- From owner-reliable_computing [at] interval [dot] louisiana.edu Fri Dec 5 17:06:57 2003 Received: (from daemon@localhost) by interval.louisiana.edu (8.11.3/8.11.3/ull-interval-math-majordomo-1.3) id hB5N6uG21336 for reliable_computing-outgoing; Fri, 5 Dec 2003 17:06:56 -0600 (CST) Received: from nwkea-mail-1.sun.com (nwkea-mail-1.sun.com [192.18.42.13]) by interval.louisiana.edu (8.11.3/8.11.3/ull-interval-math-majordomo-1.3) with ESMTP id hB5N6nw21332 for ; Fri, 5 Dec 2003 17:06:49 -0600 (CST) Received: from phys-mpk-1 ([129.146.11.81]) by nwkea-mail-1.sun.com (8.12.10/8.12.9) with ESMTP id hB5N6UUP014596 for ; Fri, 5 Dec 2003 15:06:42 -0800 (PST) Received: from susila.sfbay.sun.com (vpn-129-150-18-154.SFBay.Sun.COM [129.150.18.154]) by mpk-mail1.sfbay.sun.com (iPlanet Messaging Server 5.2 HotFix 1.16 (built May 14 2003)) with ESMTP id <0HPG00HGH2UUN4@mpk-mail1.sfbay.sun.com> for reliable_computing [at] interval [dot] louisiana.edu; Fri, 05 Dec 2003 15:06:30 -0800 (PST) Date: Fri, 05 Dec 2003 15:06:30 -0800 From: Gregory Tarsy Subject: Request for info about fma in interval arithmetic To: reliable_computing [at] interval [dot] louisiana.edu Reply-to: Gregory.Tarsy [at] Sun [dot] COM Message-id: <3FD10F76.1B4C35D0 [at] susila [dot] sfbay.sun.com> Organization: Sun Microsystems, Inc. MIME-version: 1.0 X-Mailer: Mozilla 4.79C-CCK-MCD [en] (X11; U; SunOS 5.9 sun4u) Content-type: text/plain; charset=us-ascii Content-transfer-encoding: 7bit X-Accept-Language: en Sender: owner-reliable_computing [at] interval [dot] louisiana.edu Precedence: bulk I am looking for references for the use of fused multiply-add in interval arithmetic. By "fused" multiply-add, I refer to a floating point operation that computes a * b + c with just one rounding of the final result (i.e., the product a * b is not rounded before c is added). I know that this operation can be useful in simulating multiple precision arithmetic and evaluating ordinary arithmetic ex- pressions accurately, and both of these ideas have been combined with interval arithmetic previously, but I'm looking for any papers that specifically discuss interval computations using fused multiply- add. Any pointers would be greatly appreciated. ........................................ Gregory V. Tarsy Manager, Floating Point and Numerical Computing Processors and Network Products Sun Microsystems Inc. 16 Network Circle, M/S MPK16-303 Menlo Park, CA 94025 V:(650)786-8908 F:(650)786-9551 gregory.tarsy [at] sun [dot] com ........................................ From owner-reliable_computing [at] interval [dot] louisiana.edu Sat Dec 6 01:03:10 2003 Received: (from daemon@localhost) by interval.louisiana.edu (8.11.3/8.11.3/ull-interval-math-majordomo-1.3) id hB673AS21842 for reliable_computing-outgoing; Sat, 6 Dec 2003 01:03:10 -0600 (CST) Received: from smtp2.rz.tu-harburg.de (smtp2.rz.tu-harburg.de [134.28.205.13]) by interval.louisiana.edu (8.11.3/8.11.3/ull-interval-math-majordomo-1.3) with ESMTP id hB6732w21838 for ; Sat, 6 Dec 2003 01:03:03 -0600 (CST) Received: from flunder01.rz.tu-harburg.de (flunder01.rz.tu-harburg.de [134.28.202.138]) by smtp2.rz.tu-harburg.de (8.12.7/8.12.7) with ESMTP id hB672sJQ013634 (version=TLSv1/SSLv3 cipher=EDH-RSA-DES-CBC3-SHA bits=168 verify=NO); Sat, 6 Dec 2003 08:02:54 +0100 Received: from omega.ti3.tu-harburg.de (omega.ti3.tu-harburg.de [134.28.20.55]) by flunder01.rz.tu-harburg.de (8.12.6/8.12.6) with ESMTP id hB672rKK005837; Sat, 6 Dec 2003 08:02:53 +0100 Received: by omega.ti3.tu-harburg.de (Postfix, from userid 65534) id 68B7D2472A; Sat, 6 Dec 2003 08:02:53 +0100 (CET) Received: from omega.ti3.tu-harburg.de (localhost [127.0.0.1]) by omega.ti3.tu-harburg.de (Postfix) with ESMTP id D9FA42472C; Sat, 6 Dec 2003 08:02:52 +0100 (CET) Message-ID: <3891395.1070694172845.SLOX.WebMail.wwwrun [at] omega [dot] ti3.tu-harburg.de> From: "Siegfried M. Rump" To: Gregory.Tarsy [at] Sun [dot] COM Subject: Request for info about fma in interval arithmetic Cc: reliable_computing [at] interval [dot] louisiana.edu Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Transfer-Encoding: 7bit X-Priority: 3 (normal) X-Mailer: SuSE Linux Openexchange Server 4 - WebMail (Build 2.3780) X-Operating-System: Linux 2.4.21-138-smp i386 (JVM 1.3.1_04) Organization: TU Hamburg-Harburg References: <3FD10F76.1B4C35D0 [at] susila [dot] sfbay.sun.com> Date: Sat, 6 Dec 2003 08:02:52 +0100 (CET) X-Spam-Status: No, hits=-6.3 required=5.0 tests=AWL,BAYES_01,QUOTED_EMAIL_TEXT,REFERENCES version=2.55 X-Spam-Level: X-Spam-Checker-Version: SpamAssassin 2.55 (1.174.2.19-2003-05-19-exp) X-Scanned-By: TUHH Rechenzentrum content checker Sender: owner-reliable_computing [at] interval [dot] louisiana.edu Precedence: bulk Is that a request for the (n+1)-st patent? Am Sa 06.12.2003 00:06, Gregory Tarsy schrieb: > I am looking for references for the use of fused multiply-add in > interval arithmetic. By "fused" multiply-add, I refer to a floating > point operation that computes a * b + c with just one rounding of > the final result (i.e., the product a * b is not rounded before c > is added). I know that this operation can be useful in simulating > multiple precision arithmetic and evaluating ordinary arithmetic ex- > pressions accurately, and both of these ideas have been combined > with interval arithmetic previously, but I'm looking for any papers > that specifically discuss interval computations using fused multiply- > add. Any pointers would be greatly appreciated. > > > ........................................ > Gregory V. Tarsy > Manager, Floating Point and Numerical Computing > Processors and Network Products > Sun Microsystems Inc. > 16 Network Circle, M/S MPK16-303 > Menlo Park, CA 94025 > V:(650)786-8908 F:(650)786-9551 > gregory.tarsy [at] sun [dot] com > ........................................ > ================================================= Prof. Dr. Siegfried M. Rump Inst. f. Computer Science III Technical University Hamburg-Harburg Schwarzenbergstr. 95 21071 Hamburg Germany From owner-reliable_computing [at] interval [dot] louisiana.edu Sat Dec 6 14:13:11 2003 Received: (from daemon@localhost) by interval.louisiana.edu (8.11.3/8.11.3/ull-interval-math-majordomo-1.3) id hB6KDA222773 for reliable_computing-outgoing; Sat, 6 Dec 2003 14:13:10 -0600 (CST) Received: from cs.utep.edu (mail.cs.utep.edu [129.108.5.3]) by interval.louisiana.edu (8.11.3/8.11.3/ull-interval-math-majordomo-1.3) with ESMTP id hB6KD4w22769 for ; Sat, 6 Dec 2003 14:13:04 -0600 (CST) Received: from aragorn (aragorn [129.108.5.35]) by cs.utep.edu (8.11.7/8.11.3) with SMTP id hB6KCfH25763; Sat, 6 Dec 2003 13:12:41 -0700 (MST) Message-Id: <200312062012.hB6KCfH25763 [at] cs [dot] utep.edu> Date: Sat, 6 Dec 2003 13:12:41 -0700 (MST) From: Vladik Kreinovich Reply-To: Vladik Kreinovich Subject: new book of potential interest to interval community To: reliable_computing [at] interval [dot] louisiana.edu, interval [at] cs [dot] utep.edu MIME-Version: 1.0 Content-Type: TEXT/plain; charset=ISO-8859-1 Content-MD5: Qs3+CcF39wGd7EF7H7eYLg== X-Mailer: dtmail 1.3.0 @(#)CDE Version 1.4 SunOS 5.8 sun4u sparc Content-Transfer-Encoding: 8bit X-MIME-Autoconverted: from QUOTED-PRINTABLE to 8bit by interval.louisiana.edu id hB6KD5w22770 Sender: owner-reliable_computing [at] interval [dot] louisiana.edu Precedence: bulk Scientific Computing on Itanium-based Systems by Marius Cornea, Ping Tak Peter Tang, and John Harrison Intel Press, 2002, ISBN 0971288771 http://www.intel.com/intelpress/sum_scientific.htm Written for professionals who need to write, port, or maintain reliable, accurate, and efficient numerical software for the Itanium® architecture, Scientific Computing on Itanium®-based Systems shows you how to construct key numerical infrastructure and application programs. This book describes the crucial techniques required for stability and reliability in developing numerical kernels and applications. Through numerous tested examples, the authors explain how you can get the most from the 64-bit architecture. This combination of explanation with example helps you to make difficult computations more easily and to increase the performance of your numerical software. This book provides examples to solve problems encountered in scientific and engineering computations, such as: * Polynomial evaluation * Complex arithmetic * Quad-precision arithmetic * Software pipelining, to include register rotation and modulo-scheduled loop support * SIMD instructions * Interval arithmetic * Fast-Fourier Transformation (FFT) algorithms * Numerical linear algebra and basic linear algebra subprograms(BLAS) * Vector Math Library (VML) * Cryptography About the Authors Marius Cornea, John Harrison, and Peter Tang are Intel Technology experts, bringing you up-to-date information on the Intel® Itanium® 2 Processor. They are specialists in algorithm design, exception handling, and formal validation of numerical software. In this book, they share with you the results of their combined research for 32- and 64-bit architecture. From owner-reliable_computing [at] interval [dot] louisiana.edu Sat Dec 6 14:41:30 2003 Received: (from daemon@localhost) by interval.louisiana.edu (8.11.3/8.11.3/ull-interval-math-majordomo-1.3) id hB6KfU422904 for reliable_computing-outgoing; Sat, 6 Dec 2003 14:41:30 -0600 (CST) Received: from cs.utep.edu (mail.cs.utep.edu [129.108.5.3]) by interval.louisiana.edu (8.11.3/8.11.3/ull-interval-math-majordomo-1.3) with ESMTP id hB6KfPw22900 for ; Sat, 6 Dec 2003 14:41:25 -0600 (CST) Received: from aragorn (aragorn [129.108.5.35]) by cs.utep.edu (8.11.7/8.11.3) with SMTP id hB6KfGO25939; Sat, 6 Dec 2003 13:41:16 -0700 (MST) Message-Id: <200312062041.hB6KfGO25939 [at] cs [dot] utep.edu> Date: Sat, 6 Dec 2003 13:41:15 -0700 (MST) From: Vladik Kreinovich Reply-To: Vladik Kreinovich Subject: Re: Request for info about fma in interval arithmetic To: Gregory.Tarsy [at] Sun [dot] COM, rump@tu-harburg.de Cc: reliable_computing [at] interval [dot] louisiana.edu MIME-Version: 1.0 Content-Type: TEXT/plain; charset=us-ascii Content-MD5: MIqD0MTWbD8kVHlwScdvoA== X-Mailer: dtmail 1.3.0 @(#)CDE Version 1.4 SunOS 5.8 sun4u sparc Sender: owner-reliable_computing [at] interval [dot] louisiana.edu Precedence: bulk Actually, fused multiply-add operation (fma for short) a+b*c is a new hardware-supported operation that makes scientific computations faster. It was first implemented in Itanium, The main idea is that in many cases, e.g., in computing the scalar (dot) product of two vectors a1*b1+...+an*bn, the standard way is: we start with s:=0 and then, at every step, replace s with s+ak*bk. Usually, this requires 2 operations: addition and multiplication, with two fetches from memory etc. Replacing them with a single hadrware-supported operation maskes it almost twice faster. The introduction of this new operation brings in a lot of interesting challenges. First, traditionally, when researchers try to speed up algorithms, they try to miminize the overall number of additions and multiplications (sometimes, with a larger weight attached to multiplication). In this minimization, we do not take into consideration in what order these operations are perfomed. With the new fused operation, it is desirable to re-analyze these results to see whether + and * can be combined - thus further speeding up computations on processors that allows fma. Researchers have done a lot of owrk on that, but many problems still remain open. For the interval community, D. Scott described this feature and the related challenges in his plenary talk "Sixty-four bit architecture and the Itanium processor" at the SCAN'2000 conference in Karlsruhe (a short abstract appeared in the conference Abstracts, p. 20). An interesting challenge to interval community is to modify our algorithms accordingly so as to make the largest use of this new feature. There are also related theoretical problems: our basic straightforward interval computation mean replacing operations with real numbers with operations with intervals. Of course, we can easily describe the interval form of a*b*c because it is SUE, but it raises interesting questions about the general algebraic properties of interval arithmetic in the presence of this new ternary non-traditional operation. For those who are interested in theoretical aspects, you may want to take a look at a short paper that I wrote after SCAN'2000 providing even more fundamental motivations for fma: Vladik Kreinovich, "Itanium's New Basic Operation of Fused Multiply-Add: Theoretical Explanation and Theoretical Challenge", ACM SIGACT News, 2001, Vol. 32, No. 1 (118), pp. 115-117. (downlodable as http://www.cs.utep.edu/vladik/2000/tr00-42.pdf or as http://www.cs.utep.edu/vladik/2000/tr00-42.ps.gz) There has been active attempts to show how interval techniques can be combined with the fused operation. Some results are presented in the book Scientific Computing on Itanium-based Systems by Marius Cornea, Ping Tak Peter Tang, and John Harrison Intel Press, 2002, ISBN 0971288771 http://www.intel.com/intelpress/sum_scientific.htm (I have just sent an ad about this book to the interval mailing list. I have ordered this book from amazon.com, but I have not seen it yet.) Vladik > From: "Siegfried M. Rump" > To: Gregory.Tarsy [at] Sun [dot] COM > Subject: Request for info about fma in interval arithmetic > Cc: reliable_computing [at] interval [dot] louisiana.edu > Mime-Version: 1.0 > Content-Transfer-Encoding: 7bit > X-Priority: 3 (normal) > X-Operating-System: Linux 2.4.21-138-smp i386 (JVM 1.3.1_04) > Date: Sat, 6 Dec 2003 08:02:52 +0100 (CET) > X-Spam-Status: No, hits=-6.3 required=5.0 tests=AWL,BAYES_01,QUOTED_EMAIL_TEXT,REFERENCES version=2.55 > X-Spam-Level: > X-Spam-Checker-Version: SpamAssassin 2.55 (1.174.2.19-2003-05-19-exp) > X-Scanned-By: TUHH Rechenzentrum content checker > > Is that a request for the (n+1)-st patent? > > > Am Sa 06.12.2003 00:06, Gregory Tarsy schrieb: > > > I am looking for references for the use of fused multiply-add in > > interval arithmetic. By "fused" multiply-add, I refer to a floating > > point operation that computes a * b + c with just one rounding of > > the final result (i.e., the product a * b is not rounded before c > > is added). I know that this operation can be useful in simulating > > multiple precision arithmetic and evaluating ordinary arithmetic ex- > > pressions accurately, and both of these ideas have been combined > > with interval arithmetic previously, but I'm looking for any papers > > that specifically discuss interval computations using fused multiply- > > add. Any pointers would be greatly appreciated. > > > > > > ........................................ > > Gregory V. Tarsy > > Manager, Floating Point and Numerical Computing > > Processors and Network Products > > Sun Microsystems Inc. > > 16 Network Circle, M/S MPK16-303 > > Menlo Park, CA 94025 > > V:(650)786-8908 F:(650)786-9551 > > gregory.tarsy [at] sun [dot] com > > ........................................ > > > > > ================================================= > > Prof. Dr. Siegfried M. Rump > Inst. f. Computer Science III > Technical University Hamburg-Harburg > Schwarzenbergstr. 95 > 21071 Hamburg > Germany > From owner-reliable_computing [at] interval [dot] louisiana.edu Sat Dec 6 15:20:53 2003 Received: (from daemon@localhost) by interval.louisiana.edu (8.11.3/8.11.3/ull-interval-math-majordomo-1.3) id hB6LKqr23055 for reliable_computing-outgoing; Sat, 6 Dec 2003 15:20:52 -0600 (CST) Received: from fe4.cox-internet.com (fe4-cox.cox-internet.com [66.76.2.49]) by interval.louisiana.edu (8.11.3/8.11.3/ull-interval-math-majordomo-1.3) with ESMTP id hB6LKlw23051 for ; Sat, 6 Dec 2003 15:20:47 -0600 (CST) Received: from Inspiron-8200 ([68.226.139.111]) by fe4.cox-internet.com (InterMail vK.4.04.00.03 201-232-140-20030416 license f018ea6efd6984189790b5f401fab223) with SMTP id <20031206212040.HDXJ23448.fe4@Inspiron-8200>; Sat, 6 Dec 2003 15:20:40 -0600 Message-Id: <2.2.32.20031206212003.009e1614 [at] pop [dot] louisiana.edu> X-Sender: rbk5287 [at] pop [dot] louisiana.edu X-Mailer: Windows Eudora Pro Version 2.2 (32) Mime-Version: 1.0 Content-Type: text/plain; charset="us-ascii" Date: Sat, 06 Dec 2003 15:20:03 -0600 To: Vladik Kreinovich , Gregory.Tarsy [at] Sun [dot] COM From: "R. Baker Kearfott" Subject: Re: Request for info about fma in interval arithmetic Cc: reliable_computing [at] interval [dot] louisiana.edu Sender: owner-reliable_computing [at] interval [dot] louisiana.edu Precedence: bulk Vladik et al, I may be misunderstanding "fused multiply", but the concept seems to have been around for quite some time (and not be new). In particular, supercomputer architectures of the 1980's used "pipelining." For example, on a certain machine (I forget whether the Cray I or the Cyber 205, or perhaps the IBM 3090), a multiplication required 5 machine cycles, whereas an addition required 3. When performing a "SAXPY" operation (after the LINPACK routine SAXPY, standing for Single A X Plus Y), a sequence of entire "saxpy" operations could be put into the pipeline, so that a result would be started every machine cycle and a result would finish every machine cycle, for performance of "ax + y" every machine cycle, a speedup of 8 for long vectors. Of course, the implementation of this concept may be different in present-day circuitry. Finally, it would be exciting were a computer manufacturer to implement interval arithmetic at this level of efficiency. I suspect there are a number of important details in such an implementation that aren't immediately obvious. Best regards, Baker At 01:41 PM 12/6/2003 -0700, Vladik Kreinovich wrote: >Actually, fused multiply-add operation (fma for short) a+b*c is a new >hardware-supported operation that makes scientific computations faster. It was >first implemented in Itanium, >The main idea is that in many cases, e.g., in computing the scalar (dot) >product of two vectors a1*b1+...+an*bn, the standard way is: we start with s:=0 >and then, at every step, replace s with s+ak*bk. Usually, this requires 2 >operations: addition and multiplication, with two fetches from memory etc. >Replacing them with a single hadrware-supported operation maskes it almost >twice faster. > >The introduction of this new operation brings in a lot of interesting >challenges. First, traditionally, when researchers try to speed up algorithms, >they try to miminize the overall number of additions and multiplications >(sometimes, with a larger weight attached to multiplication). In this >minimization, we do not take into consideration in what order these operations >are perfomed. With the new fused operation, it is desirable to re-analyze these >results to see whether + and * can be combined - thus further speeding up >computations on processors that allows fma. > >Researchers have done a lot of owrk on that, but many problems still remain >open. > >For the interval community, D. Scott described this feature and the >related challenges in his plenary talk "Sixty-four bit architecture >and the Itanium processor" at the SCAN'2000 conference in Karlsruhe >(a short abstract appeared in the conference Abstracts, p. 20). > >An interesting challenge to interval community is to modify our algorithms >accordingly so as to make the largest use of this new feature. There are also >related theoretical problems: our basic straightforward interval computation >mean replacing operations with real numbers with operations with intervals. Of >course, we can easily describe the interval form of a*b*c because it is SUE, >but it raises interesting questions about the general algebraic properties of >interval arithmetic in the presence of this new ternary non-traditional >operation. > >For those who are interested in theoretical aspects, you may want to take a >look at a short paper that I wrote after SCAN'2000 providing even more >fundamental motivations for fma: > > Vladik Kreinovich, > "Itanium's New Basic Operation of Fused Multiply-Add: > Theoretical Explanation and Theoretical Challenge", > ACM SIGACT News, 2001, Vol. 32, No. 1 (118), pp. 115-117. > >(downlodable as http://www.cs.utep.edu/vladik/2000/tr00-42.pdf or as >http://www.cs.utep.edu/vladik/2000/tr00-42.ps.gz) > >There has been active attempts to show how interval techniques can be combined >with the fused operation. Some results are presented in the book > >Scientific Computing on Itanium-based Systems >by Marius Cornea, Ping Tak Peter Tang, and John Harrison >Intel Press, 2002, ISBN 0971288771 >http://www.intel.com/intelpress/sum_scientific.htm > >(I have just sent an ad about this book to the interval mailing list. I have >ordered this book from amazon.com, but I have not seen it yet.) > >Vladik > >> From: "Siegfried M. Rump" >> To: Gregory.Tarsy [at] Sun [dot] COM >> Subject: Request for info about fma in interval arithmetic >> Cc: reliable_computing [at] interval [dot] louisiana.edu >> Mime-Version: 1.0 >> Content-Transfer-Encoding: 7bit >> X-Priority: 3 (normal) >> X-Operating-System: Linux 2.4.21-138-smp i386 (JVM 1.3.1_04) >> Date: Sat, 6 Dec 2003 08:02:52 +0100 (CET) >> X-Spam-Status: No, hits=-6.3 required=5.0 >tests=AWL,BAYES_01,QUOTED_EMAIL_TEXT,REFERENCES version=2.55 >> X-Spam-Level: >> X-Spam-Checker-Version: SpamAssassin 2.55 (1.174.2.19-2003-05-19-exp) >> X-Scanned-By: TUHH Rechenzentrum content checker >> >> Is that a request for the (n+1)-st patent? >> >> >> Am Sa 06.12.2003 00:06, Gregory Tarsy schrieb: >> >> > I am looking for references for the use of fused multiply-add in >> > interval arithmetic. By "fused" multiply-add, I refer to a floating >> > point operation that computes a * b + c with just one rounding of >> > the final result (i.e., the product a * b is not rounded before c >> > is added). I know that this operation can be useful in simulating >> > multiple precision arithmetic and evaluating ordinary arithmetic ex- >> > pressions accurately, and both of these ideas have been combined >> > with interval arithmetic previously, but I'm looking for any papers >> > that specifically discuss interval computations using fused multiply- >> > add. Any pointers would be greatly appreciated. >> > >> > >> > ........................................ >> > Gregory V. Tarsy >> > Manager, Floating Point and Numerical Computing >> > Processors and Network Products >> > Sun Microsystems Inc. >> > 16 Network Circle, M/S MPK16-303 >> > Menlo Park, CA 94025 >> > V:(650)786-8908 F:(650)786-9551 >> > gregory.tarsy [at] sun [dot] com >> > ........................................ >> > >> >> >> ================================================= >> >> Prof. Dr. Siegfried M. Rump >> Inst. f. Computer Science III >> Technical University Hamburg-Harburg >> Schwarzenbergstr. 95 >> 21071 Hamburg >> Germany >> > > --------------------------------------------------------------- R. Baker Kearfott, rbk [at] louisiana [dot] edu (337) 482-5346 (fax) (337) 482-5270 (work) (337) 993-1827 (home) URL: http://interval.louisiana.edu/kearfott.html Department of Mathematics, University of Louisiana at Lafayette Box 4-1010, Lafayette, LA 70504-1010, USA --------------------------------------------------------------- From owner-reliable_computing [at] interval [dot] louisiana.edu Sat Dec 6 16:48:06 2003 Received: (from daemon@localhost) by interval.louisiana.edu (8.11.3/8.11.3/ull-interval-math-majordomo-1.3) id hB6Mm5e23241 for reliable_computing-outgoing; Sat, 6 Dec 2003 16:48:05 -0600 (CST) Received: from cs.utep.edu (mail.cs.utep.edu [129.108.5.3]) by interval.louisiana.edu (8.11.3/8.11.3/ull-interval-math-majordomo-1.3) with ESMTP id hB6Mlvw23233 for ; Sat, 6 Dec 2003 16:47:57 -0600 (CST) Received: from aragorn (aragorn [129.108.5.35]) by cs.utep.edu (8.11.7/8.11.3) with SMTP id hB6Mloa26617; Sat, 6 Dec 2003 15:47:50 -0700 (MST) Message-Id: <200312062247.hB6Mloa26617 [at] cs [dot] utep.edu> Date: Sat, 6 Dec 2003 15:47:49 -0700 (MST) From: Vladik Kreinovich Reply-To: Vladik Kreinovich Subject: Re: FMA operation To: beebe [at] math [dot] utah.edu Cc: reliable_computing [at] interval [dot] louisiana.edu, Gregory.Tarsy [at] Sun [dot] COM MIME-Version: 1.0 Content-Type: TEXT/plain; charset=us-ascii Content-MD5: jYaXdwvoZkNrbQzmjypb9Q== X-Mailer: dtmail 1.3.0 @(#)CDE Version 1.4 SunOS 5.8 sun4u sparc Sender: owner-reliable_computing [at] interval [dot] louisiana.edu Precedence: bulk Dear Nelson, Thanks a lot for your correction and useful links! My apologies for the mistake. Vladik > Date: Sat, 6 Dec 2003 15:30:31 -0700 (MST) > From: "Nelson H. F. Beebe" > To: Vladik Kreinovich > > You wrote today: > > >> Actually, fused multiply-add operation (fma for short) a+b*c is a new > >> hardware-supported operation that makes scientific computations faster. It > >> was first implemented in Itanium, > > This is not correct. The IBM RS/6000 systems, shipped in February > 1990, were the first, and the PowerPC descendants used by IBM and > Apple include these instructions. Since then, IBM S/390 G5 (1999), > Intel IA-64 (2000), SGI MIPS IV (R5000, R8000, R10000, and R12000) > (1992 and later), and SPARC64-GP (2000) have provided the operation. > HP PA-RISC (1996) provides it, but with double rounding. > > The 1999 C Standard provides a standard library routine, fma(x,y,z), > for this operation, with a single rounding. > > P.S. The latest version of the my bibliography on the IA-64 > processor is at http://www.math.utah.edu/pub/tex/bib/index-table-i.html#intel-ia-64.html > A discussion of the FMA instruction from a performance viewpoint can > be found in http://www.math.utah.edu/pub/benchmarks/usirep.pdf > ------------------------------------------------------------------------------- > - Nelson H. F. Beebe Tel: +1 801 581 5254 - > - Center for Scientific Computing FAX: +1 801 581 4148 - > - University of Utah Internet e-mail: beebe [at] math [dot] utah.edu - > - Department of Mathematics, 110 LCB beebe [at] acm [dot] org beebe [at] computer [dot] org - > - 155 S 1400 E RM 233 - > - Salt Lake City, UT 84112-0090, USA URL: http://www.math.utah.edu/~beebe - > ------------------------------------------------------------------------------- From owner-reliable_computing [at] interval [dot] louisiana.edu Sat Dec 6 17:09:23 2003 Received: (from daemon@localhost) by interval.louisiana.edu (8.11.3/8.11.3/ull-interval-math-majordomo-1.3) id hB6N9MD23394 for reliable_computing-outgoing; Sat, 6 Dec 2003 17:09:22 -0600 (CST) Received: from cs.utep.edu (mail.cs.utep.edu [129.108.5.3]) by interval.louisiana.edu (8.11.3/8.11.3/ull-interval-math-majordomo-1.3) with ESMTP id hB6N9Hw23390 for ; Sat, 6 Dec 2003 17:09:17 -0600 (CST) Received: from aragorn (aragorn [129.108.5.35]) by cs.utep.edu (8.11.7/8.11.3) with SMTP id hB6N9AV26766; Sat, 6 Dec 2003 16:09:10 -0700 (MST) Message-Id: <200312062309.hB6N9AV26766 [at] cs [dot] utep.edu> Date: Sat, 6 Dec 2003 16:09:09 -0700 (MST) From: Vladik Kreinovich Reply-To: Vladik Kreinovich Subject: Re: Request for info about fma in interval arithmetic To: Gregory.Tarsy [at] Sun [dot] COM, rbk [at] louisiana [dot] edu Cc: reliable_computing [at] interval [dot] louisiana.edu MIME-Version: 1.0 Content-Type: TEXT/plain; charset=us-ascii Content-MD5: XoYTyRFPOEjd9p/AvRYjiA== X-Mailer: dtmail 1.3.0 @(#)CDE Version 1.4 SunOS 5.8 sun4u sparc Sender: owner-reliable_computing [at] interval [dot] louisiana.edu Precedence: bulk Dear Baker, Thanks for the email. I already made one wrong statement about fam being known only recently, so I may be wrong again but here is my understanding of how fma is different from pipelining. If I understand it correctly, while in the 1980s, pipelining was a new technique that was only used on supercomputers, at present, pipelining similar to what you described is implemented in most PCs: smart compilers pre-fetch data that are necessary for future operations while the CPU is performing the current operation. This pipelining is already built into the speed of different operations. The difference between this and fam is that fam provides an additional speed up. Let me give an example. Suppose that we want to compute a*(1+b). >From the normal viewpoint, we have 2 operations: first, we add 1 and b, and the we multiply. These operations can be (and will be) pipelined, and they will be pipelined with previous and future operations, but overall, we need the time of 2 (pipelined) arithmetic operations to compute this expression. If we have fam implemented, we can do it in 1 step by transforming this expression into a+a*x. This transformation is not something normal compilers would do because in the absence of fam operation in the list of hardware supported operations, there is no advantage in this trasnformation; however, once fam is hardware supported, this transformation makes computations faster. This is, of course, a simplified example; if I remember correctly, Scott had more realistic examples as well. I agree with you that I may be underestimating the complexity of implementing intervals at this level. What I would like to emphasize is that irrespective of hardware support for interval fam, we can use fam operation of floating point numbers to speed up algorithms underlying computations. I am looking forward to reading the Intel book to see what they have. Vladik > X-Sender: rbk5287 [at] pop [dot] louisiana.edu > Mime-Version: 1.0 > Date: Sat, 06 Dec 2003 15:20:03 -0600 > To: Vladik Kreinovich , Gregory.Tarsy [at] Sun [dot] COM > From: "R. Baker Kearfott" > Subject: Re: Request for info about fma in interval arithmetic > Cc: reliable_computing [at] interval [dot] louisiana.edu > > Vladik et al, > > I may be misunderstanding "fused multiply", but the concept seems to > have been around for quite some time (and not be new). > In particular, supercomputer > architectures of the 1980's used "pipelining." For example, on > a certain machine (I forget whether the Cray I or the Cyber 205, or > perhaps the IBM 3090), a multiplication required 5 machine cycles, > whereas an addition required 3. When performing a "SAXPY" operation > (after the LINPACK routine SAXPY, standing for Single A X Plus Y), > a sequence of entire "saxpy" operations could be put into the pipeline, > so that a result would be started every machine cycle and a result would > finish every machine cycle, for performance of "ax + y" every machine > cycle, a speedup of 8 for long vectors. > > Of course, the implementation of this concept may be different in > present-day circuitry. > > Finally, it would be exciting were a computer manufacturer to implement > interval arithmetic at this level of efficiency. I suspect there are > a number of important details in such an implementation that aren't > immediately obvious. > > Best regards, > > Baker > > At 01:41 PM 12/6/2003 -0700, Vladik Kreinovich wrote: > >Actually, fused multiply-add operation (fma for short) a+b*c is a new > >hardware-supported operation that makes scientific computations faster. It was > >first implemented in Itanium, > >The main idea is that in many cases, e.g., in computing the scalar (dot) > >product of two vectors a1*b1+...+an*bn, the standard way is: we start with > s:=0 > >and then, at every step, replace s with s+ak*bk. Usually, this requires 2 > >operations: addition and multiplication, with two fetches from memory etc. > >Replacing them with a single hadrware-supported operation maskes it almost > >twice faster. > > > >The introduction of this new operation brings in a lot of interesting > >challenges. First, traditionally, when researchers try to speed up > algorithms, > >they try to miminize the overall number of additions and multiplications > >(sometimes, with a larger weight attached to multiplication). In this > >minimization, we do not take into consideration in what order these operations > >are perfomed. With the new fused operation, it is desirable to re-analyze > these > >results to see whether + and * can be combined - thus further speeding up > >computations on processors that allows fma. > > > >Researchers have done a lot of owrk on that, but many problems still remain > >open. > > > >For the interval community, D. Scott described this feature and the > >related challenges in his plenary talk "Sixty-four bit architecture > >and the Itanium processor" at the SCAN'2000 conference in Karlsruhe > >(a short abstract appeared in the conference Abstracts, p. 20). > > > >An interesting challenge to interval community is to modify our algorithms > >accordingly so as to make the largest use of this new feature. There are also > >related theoretical problems: our basic straightforward interval computation > >mean replacing operations with real numbers with operations with intervals. Of > >course, we can easily describe the interval form of a*b*c because it is SUE, > >but it raises interesting questions about the general algebraic properties of > >interval arithmetic in the presence of this new ternary non-traditional > >operation. > > > >For those who are interested in theoretical aspects, you may want to take a > >look at a short paper that I wrote after SCAN'2000 providing even more > >fundamental motivations for fma: > > > > Vladik Kreinovich, > > "Itanium's New Basic Operation of Fused Multiply-Add: > > Theoretical Explanation and Theoretical Challenge", > > ACM SIGACT News, 2001, Vol. 32, No. 1 (118), pp. 115-117. > > > >(downlodable as http://www.cs.utep.edu/vladik/2000/tr00-42.pdf or as > >http://www.cs.utep.edu/vladik/2000/tr00-42.ps.gz) > > > >There has been active attempts to show how interval techniques can be combined > >with the fused operation. Some results are presented in the book > > > >Scientific Computing on Itanium-based Systems > >by Marius Cornea, Ping Tak Peter Tang, and John Harrison > >Intel Press, 2002, ISBN 0971288771 > >http://www.intel.com/intelpress/sum_scientific.htm > > > >(I have just sent an ad about this book to the interval mailing list. I have > >ordered this book from amazon.com, but I have not seen it yet.) > > > >Vladik > > > >> From: "Siegfried M. Rump" > >> To: Gregory.Tarsy [at] Sun [dot] COM > >> Subject: Request for info about fma in interval arithmetic > >> Cc: reliable_computing [at] interval [dot] louisiana.edu > >> Mime-Version: 1.0 > >> Content-Transfer-Encoding: 7bit > >> X-Priority: 3 (normal) > >> X-Operating-System: Linux 2.4.21-138-smp i386 (JVM 1.3.1_04) > >> Date: Sat, 6 Dec 2003 08:02:52 +0100 (CET) > >> X-Spam-Status: No, hits=-6.3 required=5.0 > >tests=AWL,BAYES_01,QUOTED_EMAIL_TEXT,REFERENCES version=2.55 > >> X-Spam-Level: > >> X-Spam-Checker-Version: SpamAssassin 2.55 (1.174.2.19-2003-05-19-exp) > >> X-Scanned-By: TUHH Rechenzentrum content checker > >> > >> Is that a request for the (n+1)-st patent? > >> > >> > >> Am Sa 06.12.2003 00:06, Gregory Tarsy schrieb: > >> > >> > I am looking for references for the use of fused multiply-add in > >> > interval arithmetic. By "fused" multiply-add, I refer to a floating > >> > point operation that computes a * b + c with just one rounding of > >> > the final result (i.e., the product a * b is not rounded before c > >> > is added). I know that this operation can be useful in simulating > >> > multiple precision arithmetic and evaluating ordinary arithmetic ex- > >> > pressions accurately, and both of these ideas have been combined > >> > with interval arithmetic previously, but I'm looking for any papers > >> > that specifically discuss interval computations using fused multiply- > >> > add. Any pointers would be greatly appreciated. > >> > > >> > > >> > ........................................ > >> > Gregory V. Tarsy > >> > Manager, Floating Point and Numerical Computing > >> > Processors and Network Products > >> > Sun Microsystems Inc. > >> > 16 Network Circle, M/S MPK16-303 > >> > Menlo Park, CA 94025 > >> > V:(650)786-8908 F:(650)786-9551 > >> > gregory.tarsy [at] sun [dot] com > >> > ........................................ > >> > > >> > >> > >> ================================================= > >> > >> Prof. Dr. Siegfried M. Rump > >> Inst. f. Computer Science III > >> Technical University Hamburg-Harburg > >> Schwarzenbergstr. 95 > >> 21071 Hamburg > >> Germany > >> > > > > > > --------------------------------------------------------------- > R. Baker Kearfott, rbk [at] louisiana [dot] edu (337) 482-5346 (fax) > (337) 482-5270 (work) (337) 993-1827 (home) > URL: http://interval.louisiana.edu/kearfott.html > Department of Mathematics, University of Louisiana at Lafayette > Box 4-1010, Lafayette, LA 70504-1010, USA > --------------------------------------------------------------- From owner-reliable_computing [at] interval [dot] louisiana.edu Sat Dec 6 21:01:13 2003 Received: (from daemon@localhost) by interval.louisiana.edu (8.11.3/8.11.3/ull-interval-math-majordomo-1.3) id hB731Dd23728 for reliable_computing-outgoing; Sat, 6 Dec 2003 21:01:13 -0600 (CST) Received: from sys06.mail.msu.edu (sys06.mail.msu.edu [35.9.75.106]) by interval.louisiana.edu (8.11.3/8.11.3/ull-interval-math-majordomo-1.3) with ESMTP id hB7316w23724 for ; Sat, 6 Dec 2003 21:01:06 -0600 (CST) Received: from 12-245-210-11.client.attbi.com ([12.245.210.11] helo=TP570MSUMB) by sys06.mail.msu.edu with asmtp (Exim 4.10 #3) (TLSv1:RC4-MD5:128) (authenticated as berz) id 1ASpAJ-000AgJ-00 for reliable_computing [at] interval [dot] louisiana.edu; Sat, 06 Dec 2003 22:00:59 -0500 From: "Martin Berz" To: "Reliable Computing" Subject: Fourth International Conference on Automatic Differentiation Date: Sat, 6 Dec 2003 22:00:46 -0500 Message-ID: MIME-Version: 1.0 Content-Type: text/plain; charset="us-ascii" Content-Transfer-Encoding: 7bit X-Priority: 3 (Normal) X-MSMail-Priority: Normal X-Mailer: Microsoft Outlook IMO, Build 9.0.6604 (9.0.2911.0) X-MimeOLE: Produced By Microsoft MimeOLE V6.00.2800.1165 Importance: Normal Sender: owner-reliable_computing [at] interval [dot] louisiana.edu Precedence: bulk Dear Colleagues, this meeting may be, or should be, of interest to this community. I hope we will see many of you there and have a substantial interval representation. Sincerely, Martin Berz *********************************************************************** FIRST ANNOUNCEMENT AND CALL FOR PAPERS -- AD2004 Fourth International Conference on Automatic Differentiation (AD2004) July 19-23, 2004 The University of Chicago Gleacher Center Chicago, IL, USA http://www.autodiff.org/ad04/ The Fourth International Conference on Automatic Differentiation (AD2004) will be held July 20-23, 2004 in downtown Chicago. It will be preceded by a short course on automatic differentiation on July 19. Submissions are requested in all areas of automatic differentiation theory, tools, and applications. Reports on novel applications of automatic differentiation and on the interface between automatic differentiation and other areas of computer science and mathematics are especially welcome. Extended abstracts of approximately two pages are requested by March 29, 2004. See http://www.autodiff.org/ad04 for more details. From owner-reliable_computing [at] interval [dot] louisiana.edu Sat Dec 6 22:56:25 2003 Received: (from daemon@localhost) by interval.louisiana.edu (8.11.3/8.11.3/ull-interval-math-majordomo-1.3) id hB74uP623972 for reliable_computing-outgoing; Sat, 6 Dec 2003 22:56:25 -0600 (CST) Received: from relay.pair.com (relay.pair.com [209.68.1.20]) by interval.louisiana.edu (8.11.3/8.11.3/ull-interval-math-majordomo-1.3) with SMTP id hB74uGw23968 for ; Sat, 6 Dec 2003 22:56:17 -0600 (CST) Received: (qmail 86397 invoked from network); 7 Dec 2003 04:56:12 -0000 Received: from cpe00045a0ac3bf-cm014340013788.cpe.net.cable.rogers.com (HELO ?192.168.1.2?) (65.49.108.156) by relay.pair.com with SMTP; 7 Dec 2003 04:56:12 -0000 X-pair-Authenticated: 65.49.108.156 Mime-Version: 1.0 X-Sender: jat711k [at] www [dot] peda.com Message-Id: In-Reply-To: <200312062309.hB6N9AV26766 [at] cs [dot] utep.edu> References: <200312062309.hB6N9AV26766 [at] cs [dot] utep.edu> Date: Sat, 6 Dec 2003 23:55:52 -0500 To: reliable_computing [at] interval [dot] louisiana.edu From: Jeff Tupper Subject: Re: Request for info about fma in interval arithmetic Content-Type: text/plain; charset="iso-8859-1" ; format="flowed" Content-Transfer-Encoding: 8bit X-MIME-Autoconverted: from quoted-printable to 8bit by interval.louisiana.edu id hB74uIw23969 Sender: owner-reliable_computing [at] interval [dot] louisiana.edu Precedence: bulk Fused multiply-add instructions are available on PowerPC processors as well as (from what I see on the net) RS/6000 processors. Compilers do try to arrange instructions (and assign registers etc.) to optimize performance. Modern processors execute multiple instructions at one time (in fact, they issue [begin] multiple instructions each cycle). Part of the processor's job (note that, with some architectures, this job can be seen as being shared with the compiler) is to assign incoming (decoded) instructions to units (in this case, floating-point units --- note that the floating-point units in a processor might be specialized in various ways and so be asymmetric ). In theory, at least, fused multiply-add instructions could be assigned multiple units for their work (and / or multiple time slots in one unit). But with (one of) the current PowerPC processors --- the 970 (a.k.a. the G5 --- see http://developer.apple.com/hardware/ve/g5.html) each of the floating-point units (there are two in the processor) --- each floating-point unit can complete one fused multiply-add per cycle (however, there is a latency of 6 cycles; 6 floating-point operations may be "in progress" at one time, per floating-point unit). Jeff >Dear Baker, > >Thanks for the email. > >I already made one wrong statement about fam being known only recently, >so I may be wrong again but here is my understanding of how fma is different >from pipelining. > >If I understand it correctly, while in the 1980s, pipelining was a new >technique that was only used on supercomputers, at present, pipelining similar >to what you described is implemented in most PCs: smart compilers pre-fetch >data that are necessary for future operations while the CPU is performing the >current operation. > >This pipelining is already built into the speed of different operations. The >difference between this and fam is that fam provides an additional speed up. >Let me give an example. Suppose that we want to compute a*(1+b). > >>From the normal viewpoint, we have 2 operations: first, we add 1 >>and b, and the >we multiply. These operations can be (and will be) pipelined, and they will be >pipelined with previous and future operations, but overall, we need >the time of >2 (pipelined) arithmetic operations to compute this expression. > >If we have fam implemented, we can do it in 1 step by transforming this >expression into a+a*x. This transformation is not something normal compilers >would do because in the absence of fam operation in the list of hardware >supported operations, there is no advantage in this trasnformation; however, >once fam is hardware supported, this transformation makes computations faster. > >This is, of course, a simplified example; if I remember correctly, Scott had >more realistic examples as well. > >I agree with you that I may be underestimating the complexity of implementing >intervals at this level. > >What I would like to emphasize is that irrespective of hardware support for >interval fam, we can use fam operation of floating point numbers to speed up >algorithms underlying computations. > >I am looking forward to reading the Intel book to see what they have. > >Vladik > > >> X-Sender: rbk5287 [at] pop [dot] louisiana.edu >> Mime-Version: 1.0 >> Date: Sat, 06 Dec 2003 15:20:03 -0600 >> To: Vladik Kreinovich , Gregory.Tarsy [at] Sun [dot] COM >> From: "R. Baker Kearfott" >> Subject: Re: Request for info about fma in interval arithmetic >> Cc: reliable_computing [at] interval [dot] louisiana.edu >> >> Vladik et al, >> >> I may be misunderstanding "fused multiply", but the concept seems to >> have been around for quite some time (and not be new).  >> In particular, supercomputer >> architectures of the 1980's used "pipelining." For example, on >> a certain machine (I forget whether the Cray I or the Cyber 205, or > > perhaps the IBM 3090), a multiplication required 5 machine cycles, > > whereas an addition required 3. When performing a "SAXPY" operation > > (after the LINPACK routine SAXPY, standing for Single A X Plus Y), >> a sequence of entire "saxpy" operations could be put into the pipeline, >> so that a result would be started every machine cycle and a result would >> finish every machine cycle, for performance of "ax + y" every machine >> cycle, a speedup of 8 for long vectors.  >> >> Of course, the implementation of this concept may be different in >> present-day circuitry. >> >> Finally, it would be exciting were a computer manufacturer to implement >> interval arithmetic at this level of efficiency. I suspect there are >> a number of important details in such an implementation that aren't >> immediately obvious. >> >> Best regards, >> >> Baker >> >> At 01:41 PM 12/6/2003 -0700, Vladik Kreinovich wrote: >> >Actually, fused multiply-add operation (fma for short) a+b*c is a new >> >hardware-supported operation that makes scientific computations faster. It >was >> >first implemented in Itanium, >> >The main idea is that in many cases, e.g., in computing the scalar (dot) >> >product of two vectors a1*b1+...+an*bn, the standard way is: we start with >> s:=0 >> >and then, at every step, replace s with s+ak*bk. Usually, this requires 2 >> >operations: addition and multiplication, with two fetches from memory etc. >> >Replacing them with a single hadrware-supported operation maskes it almost >> >twice faster. >> > >> >The introduction of this new operation brings in a lot of interesting >> >challenges. First, traditionally, when researchers try to speed up >> algorithms,  >> >they try to miminize the overall number of additions and multiplications >> >(sometimes, with a larger weight attached to multiplication). In this >> >minimization, we do not take into consideration in what order these >operations >> >are perfomed. With the new fused operation, it is desirable to re-analyze >> these >> >results to see whether + and * can be combined - thus further speeding up >> >computations on processors that allows fma. >> > >> >Researchers have done a lot of owrk on that, but many problems still remain >> >open. >> > >> >For the interval community, D. Scott described this feature and the >> >related challenges in his plenary talk "Sixty-four bit architecture >> >and the Itanium processor" at the SCAN'2000 conference in Karlsruhe >> >(a short abstract appeared in the conference Abstracts, p. 20). >> > >> >An interesting challenge to interval community is to modify our algorithms >> >accordingly so as to make the largest use of this new feature. There are >also >> >related theoretical problems: our basic straightforward interval >>computation >> >mean replacing operations with real numbers with operations with intervals. >Of >> >course, we can easily describe the interval form of a*b*c because >>it is SUE, >> >but it raises interesting questions about the general algebraic properties >of >> >interval arithmetic in the presence of this new ternary non-traditional >> >operation. >> > >> >For those who are interested in theoretical aspects, you may want to take a >> >look at a short paper that I wrote after SCAN'2000 providing even more >> >fundamental motivations for fma: >> > >> > Vladik Kreinovich, >> > "Itanium's New Basic Operation of Fused Multiply-Add: >> > Theoretical Explanation and Theoretical Challenge", >> > ACM SIGACT News, 2001, Vol. 32, No. 1 (118), pp. 115-117. >> >  >> >(downlodable as http://www.cs.utep.edu/vladik/2000/tr00-42.pdf or as >> >http://www.cs.utep.edu/vladik/2000/tr00-42.ps.gz) >> > >> >There has been active attempts to show how interval techniques can be >combined >> >with the fused operation. Some results are presented in the book >> > >> >Scientific Computing on Itanium-based Systems >> >by Marius Cornea, Ping Tak Peter Tang, and John Harrison >> >Intel Press, 2002, ISBN 0971288771 >> >http://www.intel.com/intelpress/sum_scientific.htm >> > >> >(I have just sent an ad about this book to the interval mailing >>list. I have >> >ordered this book from amazon.com, but I have not seen it yet.) >> > >> >Vladik >> >  >> >> From: "Siegfried M. Rump" >> >> To: Gregory.Tarsy [at] Sun [dot] COM >> >> Subject: Request for info about fma in interval arithmetic > > >> Cc: reliable_computing [at] interval [dot] louisiana.edu >> >> Mime-Version: 1.0 >> >> Content-Transfer-Encoding: 7bit >> >> X-Priority: 3 (normal) >> >> X-Operating-System: Linux 2.4.21-138-smp i386 (JVM 1.3.1_04) >> >> Date: Sat, 6 Dec 2003 08:02:52 +0100 (CET) >> >> X-Spam-Status: No, hits=-6.3 required=5.0 >> >tests=AWL,BAYES_01,QUOTED_EMAIL_TEXT,REFERENCES version=2.55 >> >> X-Spam-Level: >> >> X-Spam-Checker-Version: SpamAssassin 2.55 (1.174.2.19-2003-05-19-exp) >> >> X-Scanned-By: TUHH Rechenzentrum content checker >> >> >> >> Is that a request for the (n+1)-st patent? >> >> >> >> >> >> Am Sa 06.12.2003 00:06, Gregory Tarsy schrieb: >> >> >> >> > I am looking for references for the use of fused multiply-add in >> >> > interval arithmetic. By "fused" multiply-add, I refer to a floating >> >> > point operation that computes a * b + c with just one rounding of >> >> > the final result (i.e., the product a * b is not rounded before c >> >> > is added). I know that this operation can be useful in simulating >> >> > multiple precision arithmetic and evaluating ordinary arithmetic ex- >> >> > pressions accurately, and both of these ideas have been combined >> >> > with interval arithmetic previously, but I'm looking for any papers >> >> > that specifically discuss interval computations using fused multiply- >> >> > add. Any pointers would be greatly appreciated. >> >> > >> >> > >> >> > ........................................ >> >> > Gregory V. Tarsy >> >> > Manager, Floating Point and Numerical Computing >> >> > Processors and Network Products >> >> > Sun Microsystems Inc. >> >> > 16 Network Circle, M/S MPK16-303 >> >> > Menlo Park, CA 94025 >> >> > V:(650)786-8908 F:(650)786-9551 >> >> > gregory.tarsy [at] sun [dot] com >> >> > ........................................ >> >> > >> >> >> >> >> >> ================================================= >> >> >> >> Prof. Dr. Siegfried M. Rump >> >> Inst. f. Computer Science III >> >> Technical University Hamburg-Harburg >> >> Schwarzenbergstr. 95 >> >> 21071 Hamburg >> >> Germany >> >> >> > >> > >> >> --------------------------------------------------------------- >> R. Baker Kearfott, rbk [at] louisiana [dot] edu (337) 482-5346 (fax) >> (337) 482-5270 (work) (337) 993-1827 (home) >> URL: http://interval.louisiana.edu/kearfott.html >> Department of Mathematics, University of Louisiana at Lafayette >> Box 4-1010, Lafayette, LA 70504-1010, USA >> --------------------------------------------------------------- -- From owner-reliable_computing [at] interval [dot] louisiana.edu Sun Dec 7 08:48:34 2003 Received: (from daemon@localhost) by interval.louisiana.edu (8.11.3/8.11.3/ull-interval-math-majordomo-1.3) id hB7EmXD25074 for reliable_computing-outgoing; Sun, 7 Dec 2003 08:48:33 -0600 (CST) Received: from fe3.cox-internet.com (fe3-cox.cox-internet.com [66.76.2.40]) by interval.louisiana.edu (8.11.3/8.11.3/ull-interval-math-majordomo-1.3) with ESMTP id hB7EmRw25070 for ; Sun, 7 Dec 2003 08:48:28 -0600 (CST) Received: from Inspiron-8200 ([68.226.139.111]) by fe3.cox-internet.com (InterMail vK.4.04.00.03 201-232-140-20030416 license f018ea6efd6984189790b5f401fab223) with SMTP id <20031207144819.OPGK636.fe3@Inspiron-8200> for ; Sun, 7 Dec 2003 08:48:19 -0600 Message-Id: <2.2.32.20031207144735.00a07920 [at] pop [dot] louisiana.edu> X-Sender: rbk5287 [at] pop [dot] louisiana.edu X-Mailer: Windows Eudora Pro Version 2.2 (32) Mime-Version: 1.0 Content-Type: text/plain; charset="us-ascii" Date: Sun, 07 Dec 2003 08:47:35 -0600 To: reliable_computing [at] interval [dot] louisiana.edu From: "R. Baker Kearfott" Subject: new GlobSol installation script Sender: owner-reliable_computing [at] interval [dot] louisiana.edu Precedence: bulk Colleagues, George Corliss has provided a corrected GlobSol installation script for Sun machines, with the Sun compiler. Before, certain scripts had to be changed manually to obtain proper paths for the installation directory. Now, the installation should be automatic. I have placed the script, "unpack_sun_sun" in the "Download GlobSol" area. Best regards, Baker --------------------------------------------------------------- R. Baker Kearfott, rbk [at] louisiana [dot] edu (337) 482-5346 (fax) (337) 482-5270 (work) (337) 993-1827 (home) URL: http://interval.louisiana.edu/kearfott.html Department of Mathematics, University of Louisiana at Lafayette Box 4-1010, Lafayette, LA 70504-1010, USA --------------------------------------------------------------- From owner-reliable_computing [at] interval [dot] louisiana.edu Sun Dec 7 20:41:19 2003 Received: (from daemon@localhost) by interval.louisiana.edu (8.11.3/8.11.3/ull-interval-math-majordomo-1.3) id hB82fI725660 for reliable_computing-outgoing; Sun, 7 Dec 2003 20:41:18 -0600 (CST) Received: from nwkea-mail-2.sun.com (nwkea-mail-2.sun.com [192.18.42.14]) by interval.louisiana.edu (8.11.3/8.11.3/ull-interval-math-majordomo-1.3) with ESMTP id hB82fCw25656 for ; Sun, 7 Dec 2003 20:41:12 -0600 (CST) Received: from phys-soms-1 ([129.158.80.181]) by nwkea-mail-2.sun.com (8.12.10/8.12.9) with ESMTP id hB82erxA024146 for ; Sun, 7 Dec 2003 18:40:54 -0800 (PST) Received: from conversion-daemon.soms-mail1.Aus.Sun.COM by soms-mail1.Aus.Sun.COM (iPlanet Messaging Server 5.2 HotFix 1.16 (built May 14 2003)) id <0HPK00I011JD0E@soms-mail1.Aus.Sun.COM> for reliable_computing [at] interval [dot] louisiana.edu; Mon, 08 Dec 2003 13:40:53 +1100 (EST) Received: from [129.158.87.155] (retreat.Aus.Sun.COM [129.158.87.155]) by soms-mail1.Aus.Sun.COM (iPlanet Messaging Server 5.2 HotFix 1.16 (built May 14 2003)) with ESMTP id <0HPK008WK2431T@soms-mail1.Aus.Sun.COM> for reliable_computing [at] interval [dot] louisiana.edu; Mon, 08 Dec 2003 13:40:51 +1100 (EST) Date: Mon, 08 Dec 2003 13:37:04 +1100 From: Richard Smith - Systems Engineer - Melbourne Subject: Re: Request for info about fma in interval arithmetic In-reply-to: <3FD10F76.1B4C35D0 [at] susila [dot] sfbay.sun.com> To: reliable_computing [at] interval [dot] louisiana.edu Message-id: <1070851023.7669.2.camel@retreat> Organization: MIME-version: 1.0 X-Mailer: Ximian Evolution 1.2.2 Content-type: text/plain Content-transfer-encoding: 7BIT References: <3FD10F76.1B4C35D0 [at] susila [dot] sfbay.sun.com> Sender: owner-reliable_computing [at] interval [dot] louisiana.edu Precedence: bulk Naively, doesn't the use of FMA in interval arithmetic have the potential for computing narrower bounds, if internally only one interval rounding is performed? Maybe this translates into significant benefit for some algorithms. I guess in general there is scope for replacing the composition of a large number of primitive operations with a highly-optimized routine, hence the existence of Level 3 BLAS routines like DGEMM (for matrix-matrix multiply). Vendors generally supply implementations tuned for their architectures. As for whether FMA is a performance win, this requires an understanding of modern processor architecture, in particular what the most common bottlenecks are. Lack of FMA is rarely close to the top of the list. Pipeline stalls due to memory latency are much more common, having insufficient instruction-level parallelism (ILP) is another. The net effect is usually a relatively low number of instructions per cycle (IPC) executed. Numbers like 0.2 to 0.5 are quite common on a a 4-way superscalar processor. Since almost all modern processors have hardware performance counters built in to them, I encourage people to use them to measure their codes on a particular pipeline, e.g. IPC and flops/clock. There will also be counters to explain some of the pipeline stalls e.g. L1 cache misses. When rounding modes have to be changed frequently, some architectures implement this as a write to a floating point status register which can stall a pipeline. Unfortunately accessing the performance counters on Linux will require compiling your own kernel with support for the counters added. Different rounding may also be in issue when trying to exploit Single Instruction Multiple Data (SIMD) instructions to perform multiple floating point instructions in parallel. SIMD instructions usually share rounding mode. The difference between instruction latency and throughput also needs to be considered, since this impacts code with low ILP. A dependent instruction can't be computed until the results it depends on are available, which may be 3 or 4 cycles. If there aren't enough independent instructions available to be executed, the pipeline once again will stall. Software IA implementations usually require multiple conditional instructions to test signs and special cases (e.g. empty intervals). Conditional instructions if incorrectly predicted also lead to pipeline stalls. [Misprediction carries a multiple-cycle penalty.] Of course, use of FMA or any other expression substitution requires a sound theory for when such substitutions can be made without violating containment, hence the various papers from Bill Walster on CSET-equivalence. This seems to me to be an argument for having native INTERVAL datatypes, rather than use of OO classes and methods. It may prove too difficult to do much expression substitution in a CSET-equivalent way without INTERVAL datatypes. -- ============================================================================ ,-_|\ Richard Smith - SE Melbourne / \ Sun Microsystems Australia Phone : +61 3 9869 6200 richard.smith [at] Sun [dot] COM Direct : +61 3 9869 6224 \_,-._/ 476 St Kilda Road Fax : +61 3 9869 6290 v Melbourne Vic 3004 Australia =========================================================================== From owner-reliable_computing [at] interval [dot] louisiana.edu Mon Dec 8 08:53:43 2003 Received: (from daemon@localhost) by interval.louisiana.edu (8.11.3/8.11.3/ull-interval-math-majordomo-1.3) id hB8Erhn26703 for reliable_computing-outgoing; Mon, 8 Dec 2003 08:53:43 -0600 (CST) Received: (from rbk5287@localhost) by interval.louisiana.edu (8.11.3/8.11.3/ull-interval-math-majordomo-1.3) id hB8ErdS26698 for reliable_computing [at] interval [dot] louisiana.edu; Mon, 8 Dec 2003 08:53:39 -0600 (CST) Received: from uniton.integrable-solutions.net (gdr.net1.nerim.net [62.212.99.186]) by interval.louisiana.edu (8.11.3/8.11.3/ull-interval-math-majordomo-1.3) with ESMTP id hB6Mlvw23234 for ; Sat, 6 Dec 2003 16:47:57 -0600 (CST) Received: from uniton.integrable-solutions.net (localhost [127.0.0.1]) by uniton.integrable-solutions.net (8.12.3/8.12.3/SuSE Linux 0.6) with ESMTP id hB6MiM9b015024; Sat, 6 Dec 2003 23:44:22 +0100 Received: (from gdr@localhost) by uniton.integrable-solutions.net (8.12.3/8.12.3/Submit) id hB6MiLMg015023; Sat, 6 Dec 2003 23:44:21 +0100 X-Authentication-Warning: uniton.integrable-solutions.net: gdr set sender to gdr@integrable-solutions.net using -f To: Vladik Kreinovich Cc: Gregory.Tarsy [at] sun [dot] com, rump@tu-harburg.de, reliable_computing [at] interval [dot] louisiana.edu Subject: Re: Request for info about fma in interval arithmetic References: <200312062041.hB6KfGO25939 [at] cs [dot] utep.edu> From: Gabriel Dos Reis In-Reply-To: <200312062041.hB6KfGO25939 [at] cs [dot] utep.edu> Organization: Integrable Solutions Date: 06 Dec 2003 23:44:21 +0100 Message-ID: Lines: 10 MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Sender: owner-reliable_computing [at] interval [dot] louisiana.edu Precedence: bulk Vladik Kreinovich writes: | Actually, fused multiply-add operation (fma for short) a+b*c is a new | hardware-supported operation that makes scientific computations | faster. It was first implemented in Itanium, really? I thought the idea has been there for a long time and it was first "popularized" by Power/PowerPC-based architectures... -- Gaby From owner-reliable_computing [at] interval [dot] louisiana.edu Tue Dec 9 09:21:52 2003 Received: (from daemon@localhost) by interval.louisiana.edu (8.11.3/8.11.3/ull-interval-math-majordomo-1.3) id hB9FLpk28122 for reliable_computing-outgoing; Tue, 9 Dec 2003 09:21:51 -0600 (CST) Received: from cs.utep.edu (mail.cs.utep.edu [129.108.5.3]) by interval.louisiana.edu (8.11.3/8.11.3/ull-interval-math-majordomo-1.3) with ESMTP id hB9FLlw28118 for ; Tue, 9 Dec 2003 09:21:47 -0600 (CST) Received: from aragorn (aragorn [129.108.5.35]) by cs.utep.edu (8.11.7/8.11.3) with SMTP id hB9FLfh18335; Tue, 9 Dec 2003 08:21:41 -0700 (MST) Message-Id: <200312091521.hB9FLfh18335 [at] cs [dot] utep.edu> Date: Tue, 9 Dec 2003 08:21:41 -0700 (MST) From: Vladik Kreinovich Reply-To: Vladik Kreinovich Subject: Deadline Extended to Jan 15 To: reliable_computing [at] interval [dot] louisiana.edu, interval [at] cs [dot] utep.edu MIME-Version: 1.0 Content-Type: TEXT/plain; charset=us-ascii Content-MD5: mDP5uiEwq0kVfXudHOCnpg== X-Mailer: dtmail 1.3.0 @(#)CDE Version 1.4 SunOS 5.8 sun4u sparc Sender: owner-reliable_computing [at] interval [dot] louisiana.edu Precedence: bulk ------------- Begin Forwarded Message ------------- ------------------------------------------------------------------ Call for Papers *********** The deadline has been moved to January 15, 2004 ************** The Fourth International Conference on Rough Sets and Current Trends in Computing (RSCTC'2004) The Linnaeus Centre for Bioinformatics Uppsala University and the Swedish University of Agricultural Sciences Uppsala, Sweden URL: http://www.lcb.uu.se/rsctc.php June 1-5, 2004 Rough set theory, proposed by Z. Pawlak in 1982, is a model of approximate reasoning. The main idea is based on the indiscernibility relation that describes indistinguishable objects. Concepts are represented by lower and upper approximations. This time, the conference will include a special track onbioinformatics. Rough set theory is applicable in bioinformatics as can be seen from many publications showing its potential in biology, medical areas, nursing,etc. Other applications of rough set theory include a broad spectrum of areas, starting from financing and ending with music. The bi-annual international conferences on Rough Sets and Current Trends in Computing (RSCTC) gather researchers and practitioners from academia and industry to present state-of-the-art in rough set theory and its applications. Conference History: RSCTC is an outgrowth of a series of annual International Workshops devoted to the subject of rough sets held alternatively in Canada, China, Japan, Poland and the USA (RSKD, RSSC, RSFDGrC, RSGrC series). The first RSCTC conference was held in Warsaw, Poland in 1998. The second RSCTC conference was held in Banff, Canada, and the third in Malvern, Pennsylvania, U.S.A. Aim and Scope The aim of RSCTC'2004 is to provide researchers and practitioners of Information Technology an opportunity to highlight new research directions, new applications, and a growing number of relations between rough sets and areas such as knowledge discovery and data mining, intelligent information systems, multi-agent systems and bioinformatics. Relevant topics include, but are not limited to: Rough set theory and applications Applications of soft computing Approximate and uncertain reasoning Bioinformatics Complexity aspects of soft computing Computational intelligence Computing with words Decision support systems Evolutionary computing Fuzzy set theory and applications Granular computing Hybrid and integrated intelligent systems Intelligent decision support systems Intelligent information systems Knowledge discovery and data mining Logical aspects of soft computing Machine learning Multi-agent systems Multi-criteria decision analysis Neural networks Non-classical logic Pattern recognition and image processing Petri nets and concurrency Spatial reasoning Statistical inference Web intelligence Paper Submission Both research and application papers are solicited. All submitted papers will be reviewed on the basis of technical quality, relevance, significance, and clarity. Electronic submission is encouraged and preferred. Please send a Microsoft Word, postscript (PS) or PDF version of your paper, and an ASCII version of the cover page by January 15, 2004 to the web page: http://rsctc2004.med.shimane-u.ac.jp If submitting via hard copy, please submit four (4) hard-copies of your paper by January 15, 2004 to: Prof. Shusaku Tsumoto RSCTC2004 Department of Medical Informatics, Shimane University, School of Medicine (Formerly, Shimane Medical Univeristy) 89-1 Enya-cho, Izumo 693-8501 Japan Your draft paper should be less than ten (10) double-spaced, single-column pages including all figures, tables, and references. The ASCII version of a cover page must include author(s) full address, e-mail, paper title and a 200 word abstract, and up to 5 keywords. When preparing your final paper, please read the Information for Authors and use the style files for Proceedings and Other Multi-Author Volumes supplied by Springer-Verlag regarding Lecture Notes in Computer Science (LNCS). Springer-Verlag author instructions are available at: http://www.springer.de/comp/lncs/authors.html. Publication All accepted papers will be published in the conference proceedings by Springer-Verlag in the series of Lecture Notes in Computer Science/ Lecture Notes in Artificial Intelligence (LNCS/LNAI) http://www.springer.de/comp/lncs/index.html. LNCS is also published, in parallel to the printed books, in full-text electronic version. A selected number of accepted papers will be expanded and revised for inclusion in journals "Advances in Rough Sets" by Springer-Verlag, "Knowledge and Information Systems: An International Journal" by Springer-Verlag, or "Fundamenta Informaticae" by IOS Press. Important Dates Full paper submission by: January 15, 2004 Acceptance notices: February 9, 2004 Camera-ready papers due by: March 1 2004 Conference: June 1-5, 2004. Organization Honorary Chairs Professor Zdzislaw Pawlak Professor Lotfi Zadeh Organizing Committee Jan Komorowski (Sweden) Shusaku Tsumoto (Japan) Program Committee (PC) PC Chairs Shusaku Tsumoto (Japan) Jan Komorowski (Sweden) Jerzy Grzymala-Busse (U.S.A.) Roman Slowinski (Poland) Program Committee James Alpigini, Penn State University, U.S.A. Peter Apostoli, The Univ. of Toronto, Canada Malcolm Beynon, Cardiff University, UK Hans Dieter Burkhard, Humboldt University Berlin, Germany Cory Butz , University of Regina, Canada Nick Cercone, Dalhousie University, Canada Chien-Chung Chan, University of Akron, U.S.A. Andrzej Czyzewski, Gdansk University of Technology, Poland Jitender S. Deogun, University of Nebraska, U.S.A. Dieder Dubois, IRIT, Universite Paul Sabatier, France Ivo Duentsch, Brock University, Canada Salvatore Greco, University of Catania, Italy Jerzy Grzymala-Busse, University of Kansas. U.S.A. Xiaohua (Tony) Hu, Drexel University, U.S.A. Masahiro Inuiguchi, Osaka University, Japan Jouni Jarvinen, Turku Centre for Computer Science, Finland Janusz Kacprzyk, Polish Academy of Science, Poland Daijin Kim, Dept. of Computer Eng., POSTECH, Korea Jan Komorowski, Uppsala University, Sweden Jacek Koronacki, Polish Academy of Science, Poland Bozena Kostek, Gdansk University of Technology, Poland Vladik Kreinovich, University of Texas at El Paso, U.S.A. Marzena Kryszkiewicz, Warsaw University of Technology, Poland Churn-Jung Liau, Academia Sinica, Taiwan T.Y. Lin, San Jose State University, U.S.A. Chunnian Liu, Beijing University of Technology, China Qing Liu, Nan Chang University, China Pawan Lingras, St. Mary University, Canada Benedetto Matarazzo, University of Catania, Italy Lawrence J. Mazlack, University of Cincinnati, U.S.A. Ernestina Mensalvas, Universidad Politecnica de Madrid, Spain Nakata Michinori, Josai International University, Japan Sadaki Miyamoto, Tsukuba University, Japan Mikhail Moshkov, Turku Centre for Computer Science, Russia Tetsuya Murai, Hokkaido University, Japan Sestuo Ohsuga, Waseda University, Japan Ewa Orlowska, National Institute of Telecommunications, Poland Sankar Pal, Indian Statistical Institute, India Witold Pedrycz, University of Alberta, Canada James Peters, University of Manitoba, Canada Lech Polkowski, Polish Japanese Institute of Technology, Poland Sheela Ramanna, University of Winnipeg, Canada Vijay V. Raghavan, University of Louisiana at Lafayette, U.S.A. Zbigniew.W. Ras, University of North Calorina, Charlotte, U.S.A. Andrzej Skowron, Warsaw University, Poland Roman Slowinski, Poznan U