MIT OpenCourseWare http://ocw.mit.edu 6.189 Multicore Programming Primer, January (IAP) 2007 Please use the following citation format: Rodric Rabbah, 6.189 Multicore Programming Primer, January (IAP) 2007. (Massachusetts Institute of Technology: MIT OpenCourseWare). http://ocw.mit.edu (accessed MM DD, YYYY). License: Creative Commons Attribution-Noncommercial-Share Alike. Note: Please use the actual date you accessed this material in your citation. For more information about citing these materials or our Terms of Use, visit: http://ocw.mit.edu/terms6.189 IAP 2007 Lecture 17 The Raw Experience 1 6.189 IAP 2007 MIT Dr. Rodric Rabbah © Copyrights by IBM Corp. and by other(s) 2007Raw Chips October 02 2 6.189 IAP 2007 MIT Dr. Rodric Rabbah © Copyrights by IBM Corp. and by other(s) 2007Raw Microprocessor ● Tiled microprocessor with point-to-point pipelined scalar operand network ● Each tiles is 4 mm x 4mm  MIPS-style compute processor – Single-issue 8-stage pipe – 32b FPU – 32K D Cache, I Cache ● 4 on-chip mesh networks  Two for operands  One for cache misses, I/O  One for message passing 6.189 IAP 2007 MIT Dr. Rodric Rabbah © Copyrights by IBM Corp. and by other(s) 20073Raw Microprocessor ● 16 tiles (16 issue) ● 180 nm ASIC (IBM SA-27E) ● ~100 million transistors ● 1 million gates ● 3-4 years of development ● 1.5 years of testing ● 200K lines of test code ● Core Frequency:  425 MHz @ 1.8 V  500 MHz @ 2.2 V ● Frequency competitive with IBM-implemented PowerPCs in same process ● 18W average power 6.189 IAP 2007 MIT Dr. Rodric Rabbah © Copyrights by IBM Corp. and by other(s) 20074One Cycle in the Life of a Tiled Processor ● Application uses as many tiles as needed to exploit its parallelism Image by MIT OCW.Image by MIT OCW. 5 6.189 IAP 2007 MIT Dr. Rodric Rabbah © Copyrights by IBM Corp. and by other(s) 2007Raw Motherboard 6.189 IAP 2007 MIT Dr. Rodric Rabbah © Copyrights by IBM Corp. and by other(s) 20076Raw in Action 6.189 IAP 2007 MIT 7Dr. Rodric Rabbah © Copyrights by IBM Corp. and by other(s) 2007MPEG-2 Encoder Performance 16 Speedup8 4 1 1 4 8 # of Tiles 16 10 30 60 Frames/s 350 x 240 Images 720 x 480 Images 1 4 8 16 4 8 10 30 60 SpeedupFrames/s # of Tiles  Square – Linear speedup and-optimize parallel implseline macro8  Diamond – H d, slice parallel implementation  Circle – Slice ementation  Triangle – Ba block parallel implementation 6.189 IAP 2007 MIT Dr. Rodric Rabbah © Copyrights by IBM Corp. and by other(s) 2007MPEG-2 Encoder Performance Encoding Rate (frames/s) # Tiles 352 x 240 640 x 480 720 x 480 1 4.30 1.14 1.00 2 8.48 2.24 1.97 4 16.18 4.45 3.84 8 30.82 8.69 7.52 16 58.65 16.74 14.57 32 103* 30* 64 158* 51.90 * Estimated data rates 6.189 IAP 2007 MIT Dr. Rodric Rabbah © Copyrights by IBM Corp. and by other(s) 20079Programmable Graphics Pipeline Input Vertex Vertex Sync Triangle Setup Pixel Pixel P simplified graphics pipeline V 6.189 IAP 2007 MIT Dr. Rodric Rabbah © Copyrights by IBM Corp. and by other(s) 200710Phong Shading ● Per-pixel phong-shaded polyhedron ● 162 vertices, 1 light Output, rendered using Raw simulator 6.189 IAP 2007 MIT Dr. Rodric Rabbah © Copyrights by IBM Corp. and by other(s) 200711Phong Shading (64-tiles) Fixed pipeline Reconfigurable pipeline ● 33% faster ● 150% better utilization Dr. Rodric Rabbah © Copyrights by IBM Corp. and by other(s) 200712 6.189 IAP 2007 MITShadow Volumes ● 4 textured triangles ● 1 point light ● Rendered in 3 passes Output, rendered using Raw simulator 6.189 IAP 2007 MIT Dr. Rodric Rabbah © Copyrights by IBM Corp. and by other(s) 200713Shadow Volumes (64-tiles) Fixed pipeline Pass 1 Pass 2 Pass 3 Pass 1 Pass 2 Pass 3 Reconfigurable pipeline ● 40% faster cycles 14 6.189 IAP 2007 MIT Dr. Rodric Rabbah © Copyrights by IBM Corp. and by other(s) 20071020 Element Microphone Array 6.189 IAP 2007 MIT Dr. Rodric Rabbah © Copyrights by IBM Corp. and by other(s) 200715Case Study: Beamformer MFLOPS 1,600 1,400 1,200 1,000 800 600 400 200 0 19 1,430 Optimized StreamIt 420 MHz 16 tile Raw 640 Unoptimized StreamIt 420 MHz 64 tile Raw 240 C program 1 GHz Pentium III C program 420 MHz single tile Raw 6.189 IAP 2007 MIT Dr. Rodric Rabbah © Copyrights by IBM Corp. and by other(s) 200716The Raw Experience ● Insights into the design Raw architecture ● Raw parallelizing compiler ● StreamIt language and Compiler 6.189 IAP 2007 MIT Dr. Rodric Rabbah © Copyrights by IBM Corp. and by other(s) 200717Scalability Problems in Wide Issue Processors Control Wide Fetch (16 inst) Unified Load/Store Queue PC RF ALUALUALUALU ALUALUALUALU ALUALUALUALU ALUALUALUALUBypass Net 6.189 IAP 2007 MIT Dr. Rodric Rabbah © Copyrights by IBM Corp. and by other(s) 200718Area and Frequency Scalability Problems ~N3 ~N2 N ALUs ALUALUALUALU ALUALUALUALU ALUALUALUALU ALUALUALUALUBypass Net RF 6.189 IAP 2007 MIT Dr. Rodric Rabbah © Copyrights by IBM Corp. and by other(s) 200719Operand Routing is Global ALUALUALUALU ALUALUALUALU ALUALUALUALU ALUALUALUALUBypass Net RF >> + 20 6.189 IAP 2007 MIT Dr. Rodric Rabbah © Copyrights by IBM Corp. and by other(s) 2007Idea: Make Operand Routing Local ALUALUALUALU ALUALUALUALU ALUALUALUALU ALUALUALUALUBypass Net RF 21 6.189 IAP 2007 MIT Dr. Rodric Rabbah © Copyrights by IBM Corp. and by other(s) 2007Idea: Exploit Locality ALUALUALUALU ALUALUALUALU ALUALUALUALU ALUALUALUALU RF Bypass Net 22 6.189 IAP 2007 MIT Dr. Rodric Rabbah © Copyrights by IBM Corp. and by other(s) 2007Replace Crossbar with Point-To-Point Network ALUALUALUALU ALUALUALUALU ALUALUALUALU ALUALUALUALU RF 23 6.189 IAP 2007 MIT Dr. Rodric Rabbah © Copyrights by IBM Corp. and by other(s) 2007Replace Crossbar with Point-To-Point Network ALUALUALUALU ALUALUALUALU ALUALUALUALU ALUALUALUALU RF >> + Dr. Rodric Rabbah © Copyrights by IBM Corp. and by other(s) 200724 6.189 IAP 2007 MITOperand Transport Latency Crossbar Point-to-Point Network Non-local Placement ~ N ~ N½ Locality-driven Placement ~ N ~1 6.189 IAP 2007 MIT Dr. Rodric Rabbah © Copyrights by IBM Corp. and by other(s) 200725Distribute the Register File RF ALUALUALUALU RF RF RF RF ALUALUALUALU RF RF RF RF ALUALUALUALU ALUALUALUALU RFRF RFRF RFRF RFRF 6.189 IAP 2007 MIT Dr. Rodric Rabbah © Copyrights by IBM Corp. and by other(s) 200726CMore Scalability Problems LBSALAEALUALUALUALU ALUALUALUALU ALUALUALUALU ALUALUALUALU RFRF