DOCSLIB.ORG

Threaded Code Variations and Optimizations

Total Page：16

File Type：pdf, Size：1020Kb

Download full-text PDF Read full-text
Threaded Code Variations and Optimizations Threaded Co de Variations and Optimizations M. Anton Ertl TU Wien Abstract header of x code field dovar code orth has been traditionally implemented as in- F body direct threaded co de, where the co de for non- es is the co de- eld address of the word. To primitiv header of foo docol code um b ene t from combining sequences get the maxim code field primitives into sup erinstructions, the co de pro- of code field e should b e a primitivefol- duced for a non-primitiv body @ code wed by a parameter (e.g., lit addr for variables). lo code field pap er takes a lo ok at the steps from a tra- This ;s code ditional threaded-co de implementation to sup erin- structions, and at the size and sp eed e ects of the Figure 1: Traditional indirect threaded co de various steps. The use of sup erinstructions gives sp eedups of up to a factor of 2 on large b enchmarks on pro cessors with branch target bu ers, but re- pap er describ es the various steps from an indirect quires more space for the primitives and the opti- threaded implementation to an implementation us- mization tables, and also a little more space for the ing sup erinstructions (Section 2), and evaluates the threaded co de. e ect of these steps on run-time and co de size (Sec- tion 4); these steps also a ect cache consistency is- sues on the 386 architecture, whichcanhave a large 1 Intro duction in uence on p erformance (Section 3). Traditionally,Forth has b een implemented using an interpreter for indirect threaded co de. However, over time programs have tended to dep end less on 2 Threaded co de variations sp eci c features of this implementation technique, and an increasing number of Forth systems have Wewillusethefollowing co de as a running example: used other implementation techniques, in particu- lar native co de compilation. One of the goals of the Gforth pro ject is to variable x provide comp etetive p erformance, another goal is p ortability to a wide range of machines. To meet : foo x @ ; the p ortability goal, we decided to stay with a threaded-co de engine compiled with GCC [Ert93]; to regain ground lost on the eciency front, wede- cided to combine sequences of primitives into su- 2.1 Traditional indirect threaded p erinstructions. This technique has b een prop osed co de bySchutz [Sch92] and implemented by Wil Baden In indirect threaded co de (Fig. 1) the co de of a colon in this4th [Bad95] and by Marcel Hendrix in a ver- sion of eforth. It is related to the concepts of sup er- de nition consists of a sequence of the co de eld ad- combinators [Hug82] and sup erop erators [Pro95]. dresses (CFAs) of the words contained in the colon de nition. SuchaCFA points to a co de eld that Non-primitives in traditional indirect threaded contains the address of the machine co de that p er- co de cannot be combined into sup erinstructions, forms the function of the word. but it is p ossible to compile them into using prim- itives that can be combined, and then into using The reason for the indirection through the co de direct threading instead of indirect threading. This eld is to supp ort non-primitives, likethevariable x in our example: The dovar routine can compute Corresp ondence Address: Institut fur Computer- the body address by adding the co de eld size to sprachen, Technische Universitat Wien, Argentinierstrae 8, A-1040 Wien, Austria; [email protected] en.a c.a t the CFA. header of x header of x jmp dovar dovar code jmp dovar dovar code body body header of foo header of foo docol code jmp docol docol code jmp docol lit lit code body @ code @ code ;s code @ ;s ;s code Figure 2: Direct threaded co de, traditional variant Figure 3: Primitive-centric direct threaded co de 2.2 Traditional-style direct threaded co de class co de colon de nition call body For primitives the indirection is only needed be- constant lit value cause we do not knowinadvance whether the next value lit body @ word is a primitive or not. So, in order to avoid variable lit body the overhead of the indirection for primitives, some user variable useraddr offset Forth implementors have implemented a variantof defered lit body @ execute this scheme using direct threaded co de (see Fig. 2). eld lit offset + For primitives, the threaded co de for a word p oints do es-de ned lit body call does-code directly to the machine co de, and the execution to- ;co de-de ned lit cfa execute ken p oints there, to o. For non-primitives, the threaded co de cannot Figure 4: Compiling non-primitives for the point directly to the machine-co de routine (the primitive-centric scheme (with the least number of doer ), b ecause the do er would not know how to additional primitives) nd the body of the word. So the threaded co de points to a (variant of the) co de eld, and that eld contains a machine-co de jump to the do er. This change replaces a load in every word by a sumes a one-cell execution token (represented bya jump in every non-primitive. Because only 20%{ CFA), not a primitive with an argument. 25% of the dynamically executed words are non- Figure 4 shows how the various classes of non- primitives, direct threading is faster on most pro ces- primitives can b e compiled. Instead of using a se- sors, but not always on some p opular pro cessors quence of several primitives for some classes, new (see Section 3). primitives could be intro duced that p erform the The main disadvantage of direct threading in whole action in one primitive. However, if we com- an implementation like Gforth is that it requires bine frequent sequences of primitives into sup erin- architecture-sp eci c co de for creating the contents structions, this will happ en automatically; explic- of the co de elds (Gforth currently supp orts direct itly intro ducing these primitives may actually de- threading on 7 architectures). grade the ecacy of the sup erinstruction optimiza- tion, b ecause the optimizer would not know without 2.3 Primitive-centric threaded co de additional e ort that, e.g., our value-primitive is the same as the lit @ combination it found elsewhere. An alternative metho d to provide the b o dy address is to simply layitdown into the threaded co de as This scheme takes more space for the co de of non- immediate parameter. Then the threaded co de for primitives; this is the original reason for preferring a non-primitivedoes not p oint to the \co de eld" the traditional style when memory is scarce. There of the nonprimitive, but to a primitive that gets probably is little di erence from traditional-style the inline parameter and p erforms the function of direct threading in run-time p erformance on most the non-primitive. For the variable x in our exam- pro cessors: Only non-primitives are a ected; in the ple this primitiveis lit (the run-time primitivefor traditional-style scheme there is a jump and an ad- literal). dition to compute the b o dy address, whereas in the primitive-centric scheme there is a load with often This scheme is shown in Figure 3. The non- fully-exp osed latency. For some p opular pro cessors primitives still have a co de eld, whichisnotused the p erformance di erence can b e large, though (see in ordinary threaded co de execution. But it is used Section 3). for execute (and do defer), b ecause execute con- header of x : myconst create , does> @ ; code field dovar code 5 myconst five body docol code indirect header of foo call code field code field (does>) lit lit code @ code field ;s @ @ code ;s header of five code field dodoes code code field ;s code 5 Figure 5: Hybrid direct/indirect threaded co de double-indirect call Hybrid direct/indirect threaded 2.4 (does>) co de dodoes code @ e-centric scheme the Forth engine With the primitiv ;s (inner interpreter, primitives, and do ers) is divided to two parts that are relatively indep endent of in header of five h other: eac code field 5 Ordinary threaded co de. Figure 6: A does>-de ned word with indirect and Execute (and do defer), co de elds and execu- double-indirect threaded co de. tion tokens. Another interesting consequence of the division Only execute (and do defer) deals with execution 1 between ordinary threaded co de and execute etc. tokens and co de elds ,sowe can mo dify these com- in primitive-centric co de is that do ers (e.g., do col) p onents in co ordinated ways. The mo di cation we can only be invoked by execute; therefore only are interested in is to use indirect threaded co de execute has to rememb er the CFA for use by the elds; of course this requires an additional indirec- do ers. Unfortunately,theobvious way of expressing tion in the execute co de, it requires adding co de this in GNU C leads to the conservative assump- elds to primitives (see Fig. 5), and the execution tion that this value is alive (needs to b e preserved) tokens are the addresses of the co de elds. How- across all primitives, and this results in sub optimal ever, the ordinary threaded co de p oints directly to register allo cation.
Load more

Recommended publications
  • An Architectural Trail to Threaded-Code Systems
    Some programmers question certain claims made for threaded-code systems. The author defends these claims by tracing such a system from its origin in simple concepts. An Architectural Trail to Threaded-Code Systems Peter M. Kogge, IBM Federal Systems Division Interest in software systems based on threaded-code * the concepts and syntax of a high-order language, or concepts has grown remarkably in recent years. Ad- HOL; and vocates of such systems regularly make claims for them * a set of commands for a conversational monitor. that many classical programmers regard as bordering on The ISA (such as it is) is for an abstract machine that is the impossible. Typically, proponents claim that it is very efficiently implemented by short code segments in possible to construct, in 5K to 10K bytes of code, a soft- almost any current machine architecture. The HOL con- ware package that cepts include hierarchical construction of programs, * is conversational like APL, Lisp, or Basic; block structures, GOTOless programming, and user- * includes a compile facility with many high-order definable data structures. The commands for the conver- language, structured-programming constructs; sational monitor permit direct interaction with objects * exhibits performance very close to that of machine- defined by the programmer in terms he defines, not sim- coded program equivalents; ply in the terms of the underlying machine (i.e., core * is written largely in itself and consequently is largely dumps). Further, the command set is directly executable portable; in a conversational mode and includes virtually all com- * places no barriers among combinations of system, mands available in the system; when so executed, they compiler, or application code; function just as they would if found in an executed pro- * can include an integrated, user-controlled virtual gram.
    [Show full text]
  • Basic Threads Programming: Standards and Strategy
    Basic Threads Programming: Standards and Strategy Mike Dahlin [email protected] February 13, 2007 1 Motivation Some people rebel against coding standards. I don’t understand the logic. For concurrent programming in particular, there are a few good solutions that have stood the test of time (and many unhappy people who have departed from these solutions.) For concurrent programming, debugging won’t work. You must rely on (a) writing correct code and (b) writing code that you and others can read and understand. Following the rules below will help you write correct, readable code. Rules 2 through 6 below are required coding standards for CS372. Answers to homework and exam problems and project code that do not follow these standards are by definition incorrect. Section 5 discusses additional restrictions for project code (or exam pseudo-code) in Java. Again, answers that deviate from these required coding standars are, by definition, incorrect for this class. If you believe you have a better way to write concurrent programs, that’s great! Bring it to us (before you use it on an assignment!) We will examine your approach the same way we hope that a civil engineering manager would examine a proposal by a bright young civil engineer who has a proposal for a better config- uration of rebar for reinforcing a new bridge: we will hope you have found a better way, but the burden of proof for explaining the superiority of your approach and proving that there are no hidden pitfalls is on you. 2 Coding standards for concurrent programs: 1.
    [Show full text]
  • Effective Inline-Threaded Interpretation of Java Bytecode
    Effective Inline-Threaded Interpretation of Java Bytecode Using Preparation Sequences Etienne Gagnon1 and Laurie Hendren2 1 Sable Research Group Universit´eduQu´ebec `a Montr´eal, [email protected] 2 McGill University Montreal, Canada, [email protected] Abstract. Inline-threaded interpretation is a recent technique that im- proves performance by eliminating dispatch overhead within basic blocks for interpreters written in C [11]. The dynamic class loading, lazy class initialization, and multi-threading features of Java reduce the effective- ness of a straight-forward implementation of this technique within Java interpreters. In this paper, we introduce preparation sequences, a new technique that solves the particular challenge of effectively inline-threa- ding Java. We have implemented our technique in the SableVM Java virtual machine, and our experimental results show that using our tech- nique, inline-threaded interpretation of Java, on a set of benchmarks, achieves a speedup ranging from 1.20 to 2.41 over switch-based inter- pretation, and a speedup ranging from 1.15 to 2.14 over direct-threaded interpretation. 1 Introduction One of the main advantages of interpreters written in high-level languages is their simplicity and portability, when compared to static and dynamic compiler-based systems. One of their main drawbacks is poor performance, due to a high cost for dispatching interpreted instructions. In [11], Piumarta and Riccardi introduced a technique called inlined-threading which reduces this overhead by dynamically inlining instruction sequences within basic blocks, leaving a single instruction dispatch at the end of each sequence. To our knowledge, inlined-threading has not been applied to Java interpreters before.
    [Show full text]
  • Threaded Code Is Not in Fact Implemented in the Computer's Hardware
    desired generality in others. In summary, when writing hard code the user needs relatively many instructions, each of which executes relatively rapidly. An interpreter, by contrast, is a vehicle by which the user can choose his own instruction set to correspond Programming R. Morris to his specific problem. Obviously such freedom allows Techniques Editor a much shorter program for that: problem to be written. The penalty is that the instruction set of the interpreter Threaded Code is not in fact implemented in the computer's hardware. Instead the interpreter must itself be a computer pro- James R. Bell gram which simulates the action of the interpretive in- Digital Equipment Corporation struction set in terms of the actual instruction set. This can be a time-consuming proposition. Thus interpretive code tends to be shorter but slower than hard code. The concept of "threaded code" is presented as an It is instructive to look at the relation between the alternative to machine language code. Hardware and host hardware and the alternatives discussed. In the software realizations of it are given. In software it is case of hard code an instruction directs the flow of realized as interpretive code not needing an interpreter. processing by its actual execution from the IR, or in- Extensions and optimizations are mentioned. struction register, of the machine. In the case of an in- Key Words and Phrases: interpreter, machine code, terpreter, an "instruction" is in fact merely a datum time tradeoff, space tradeoff, compiled code, subroutine from the interpreting program. Thus it directs the flow calls, threaded code of processing from an accumulator or the equivalent.
    [Show full text]
  • Csc 453 Interpreters & Interpretation
    CSc 453 Interpreters & Interpretation Saumya Debray The University of Arizona Tucson Interpreters An interpreter is a program that executes another program. An interpreter implements a virtual machine , which may be different from the underlying hardware platform. Examples: Java Virtual Machine; VMs for high-level languages such as Scheme, Prolog, Icon, Smalltalk, Perl, Tcl. The virtual machine is often at a higher level of semantic abstraction than the native hardware. This can help portability, but affects performance. CSc 453: Interpreters & Interpretation 2 1 Interpretation vs. Compilation CSc 453: Interpreters & Interpretation 3 Interpreter Operation ip = start of program; while ( ¬ done ) { op = current operation at ip ; execute code for op on current operands; advance ip to next operation; } CSc 453: Interpreters & Interpretation 4 2 Interpreter Design Issues Encoding the operation I.e., getting from the opcode to the code for that operation (“dispatch”): byte code (e.g., JVM) indirect threaded code direct threaded code. Representing operands register machines: operations are performed on a fixed finite set of global locations (“registers”) (e.g.: SPIM); stack machines: operations are performed on the top of a stack of operands (e.g.: JVM). CSc 453: Interpreters & Interpretation 5 Byte Code Operations encoded as small integers (~1 byte). The interpreter uses the opcode to index into a table of code addresses: while ( TRUE ) { byte op = ∗∗∗ip; switch ( op ) { case ADD: … perform addition ; break; case SUB: … perform subtraction ; break; … } } Advantages: simple, portable. Disadvantages: inefficient. CSc 453: Interpreters & Interpretation 6 3 Direct Threaded Code Indexing through a jump table (as with byte code) is expensive. Idea : Use the address of the code for an operation as the opcode for that operation.
    [Show full text]
  • Static Analysis of Lockless Microcontroller C Programs
    Static Analysis of Lockless Microcontroller C Programs Eva Beckschulze Sebastian Biallas Stefan Kowalewski Embedded Software Laboratory RWTH Aachen University, Germany {lastname}@embedded.rwth-aachen.de Concurrently accessing shared data without locking is usually a subject to race conditions resulting in inconsistent or corrupted data. However, there are programs operating correctly without locking by exploiting the atomicity of certain operations on a specific hardware. In this paper, we describe how to precisely analyze lockless microcontroller C programs with interrupts by taking the hardware architecture into account. We evaluate this technique in an octagon-based value range analysis using access-based localization to increase efficiency. 1 Introduction Static analysis based on abstract interpretation [7] is a formal method that found its way into practice by several commercial code analysis tools. Proving the absence of run-time errors in microcontroller programs is of particular importance as microcontrollers are often deployed in safety-critical systems. However, C code analyzers usually do not cope with C extensions and hardware-specific control prevalent in microcontroller programs. This control is not only necessary for data input/output but also needed to implement interrupt service routines (ISRs), which allows some form of concurrency and can be used for asynchronous hardware communication and periodic tasks. Since the control functions of the hardware are often exposed through normal memory accesses, a sound analysis of microcontroller programs has to reflect these registers in its memory model. On the Atmega16 [2], for example, it is possible to enable/disable interrupts by a write to the SREG register which is located at the memory address 0x5F.
    [Show full text]
  • A Case Study of Multi-Threading in the Embedded Space
    A Case Study of Multi-Threading in the Embedded Space Greg Hoover Forrest Brewer Timothy Sherwood University of California, Santa University of California, Santa University of California, Santa Barbara Barbara Barbara Engineering I Engineering I Engineering I Santa Barbara, California Santa Barbara, California Santa Barbara, California 93106 93106 93106 [email protected] [email protected] [email protected] ABSTRACT vironment. Such systems open the possibility of early de- The continuing miniaturization of technology coupled with tection of structural failure, tracking microclimates in for- wireless networks has made it feasible to physically embed est canopies, and observing migratory patterns of any num- sensor network systems into the environment. Sensor net ber of species [16, 11, 29]. In an ideal world, these tiny processors are tasked with the job of handling a disparate digital systems would be operating a wireless network, han- set of interrupt driven activity, from networks to timers to dling sensor readings, controlling electro-statically-activated the sensors themselves. In this paper, we demonstrate the devices, processing software updates, and performing dis- advantages of a tiny multi-threaded microcontroller design tributed computations. To handle all of these functions at which targets embedded applications that need to respond the required throughput, we argue for the use of dynamic to events at high speed. While multi-threading is typically multi-threading at all levels of the microcontroller design. used to improve resource utilization, in the embedded space At first this concept may seem counterintuitive as most it can provide zero-cycle context switching and interrupt ser- multi-threaded architectures were developed to better-exploit vice threads (IST), enabling complex programmable control an abundance of resources [25], something that these sys- in latency constrained environments.
    [Show full text]
  • The Flow Programming Language: an Implicitly-Parallelizing, Programmer-Safe Language
    ““WriteWrite Once,Once, ParallelizeParallelize Anywhere”Anywhere” TheThe FlowFlow ProgrammingProgramming Language:Language: AnAn implicitly-parallelizing,implicitly-parallelizing, programmer-safeprogrammer-safe languagelanguage Luke Hutchison Moore'sMoore's LawLaw ● TheThe numbernumber ofof transistorstransistors thatthat cancan bebe placedplaced inexpensivelyinexpensively onon anan integratedintegrated circuitcircuit hashas doubleddoubled approximatelyapproximately everyevery twotwo years.years. Moore'sMoore's LawLaw ● ComputingComputing isis work;work; transistorstransistors dodo work;work; moremore transistorstransistors workingworking inin parallelparallel shouldshould yieldyield fasterfaster computerscomputers ● =>=> Moore'sMoore's LawLaw hashas (so(so far)far) alsoalso appliedapplied toto CPUCPU speedspeed ● BUTBUT nownow we'rewe're hittinghitting multiplemultiple walls:walls: ● TransistorTransistor sizesize – tunneling;tunneling; lithographylithography featurefeature sizesize vs.vs. wavelengthwavelength – nono moremore 2D2D densitydensity increasesincreases ● ThermalThermal envelopeenvelope – functionfunction ofof frequencyfrequency andand featurefeature sizesize =>=> wewe havehave hithit aa clockspeedclockspeed wallwall ● DataData widthwidth – 128-bit128-bit CPUs?CPUs? NeedNeed parallelparallel controlcontrol flowflow insteadinstead ButBut it'sit's worseworse thanthan thatthat It'sIt's notnot justjust thethe endend ofof Moore'sMoore's Law...Law... ...it's...it's thethe beginningbeginning ofof anan eraera ofof buggierbuggier software.software.
    [Show full text]
  • Virtual Machine Showdown: Stack Versus Registers
    Virtual Machine Showdown: Stack Versus Registers Yunhe Shi, David Gregg, Andrew Beatty M. Anton Ertl Department of Computer Science Institut f¨urComputersprachen University of Dublin, Trinity College TU Wien Dublin 2, Ireland Argentinierstraße 8 {yshi, David.Gregg, A-1040 Wien, Austria Andrew.Beatty}@cs.tcd.ie [email protected] ABSTRACT be interpreted or compiled. The most popular VMs, such as Virtual machines (VMs) are commonly used to distribute the Java VM, use a virtual stack architecture, rather than programs in an architecture-neutral format, which can eas- the register architecture that dominates in real processors. ily be interpreted or compiled. A long-running question in A long-running question in the design of VMs is whether the design of VMs is whether stack architecture or register stack architecture or register architecture can be implemented architecture can be implemented more efficiently with an more efficiently with an interpreter. On the one hand stack interpreter. We extend existing work on comparing virtual architectures allow smaller VM code so less code must be stack and virtual register architectures in two ways. Firstly, fetched per VM instruction executed. On the other hand, our translation from stack to register code is much more stack machines require more VM instructions for a given sophisticated. The result is that we eliminate an average computation, each of which requires an expensive (usually of more than 47% of executed VM instructions, with the unpredictable) indirect branch for VM instruction dispatch. register machine bytecode size only 25% larger than that of Several authors have discussed the issue [12, 15, 11, 16] and the corresponding stack bytecode.
    [Show full text]
  • Multiple Cores + SIMD + Threads) (And Their Performance Characteristics
    Lecture 2: Modern Multi-Core Architecture: (multiple cores + SIMD + threads) (and their performance characteristics) CMU 15-418: Parallel Computer Architecture and Programming (Spring 2012) Announcements ▪ WWW - http://www.cs.cmu.edu/afs/cs/academic/class/15418-s12/www/ ▪ We have another TA! - Mike Mu (Yes, our TAs are Michael and Mike) (CMU 15-418, Spring 2012) Review 1. Why has single threaded performance topped out in recent years? 2. What prevented us from obtaining maximum speedup from the parallel programs we wrote last time? (CMU 15-418, Spring 2012) Today ▪ Today we will talk computer architecture ▪ Four key concepts about how modern computers work - Two concern execution - Two concern access to memory ▪ Knowing some architecture will help you - Understand and optimize the performance of your programs - Gain intuition about what workloads might bene!t from fast parallel machines (CMU 15-418, Spring 2012) Part 1: parallel execution (CMU 15-418, Spring 2012) Example program Compute sin(x) using tailor expansion: sin(x) = x - x3/3! + x5/5! + x7/7! + ... void sinx(int N, int terms, float* x, float* result) { for (int i=0; i<N; i++) { float value = x[i]; float numer = x[i] * x[i] * x[i]; int denom = 6; // 3! int sign = -1; for (int j=1; j<=terms; j++) { value += sign * numer / denom numer *= x[i] * x[i]; denom *= (j+3) * (j+4); sign *= -1; } result[i] = value; } } (CMU 15-418, Spring 2012) Compile program void sinx(int N, int terms, float* x, float* result) x[i] { for (int i=0; i<N; i++) { float value = x[i]; float numer = x[i] * x[i] * x[i]; int denom = 6; // 3! ld r0, addr[r1] int sign = -1; mul r1, r0, r0 mul r1, r1, r0 for (int j=1; j<=terms; j++) ..
    [Show full text]
  • Generating Multithreaded Code from Parallel Haskell for Symmetric Multiprocessors Alejandro Caro
    Generating Multithreaded Code from Parallel Haskell for Symmetric Multiprocessors by Alejandro Caro S.B., Computer Science and Engineering, MIT 1990 S.M., Electrical Engineering and Computer Science, MIT 1993 Engineer in Computer Science, MIT 1998 Submitted to the Department of Electrical Engineering and Computer Science in partial fulfillment of the requirements for the degree of Doctor of Philosophy at the MASSACHUSETTS INSTITUTE OF TECHNOLOGY February 1999 @ 1999 Massachusetts Institute of Technology. All rights reserved. A u th or ....................... ....... ......... ................ Department 4f Electrical Engineering and Computer Science January 27, 1999 Certified by .:... .....- :...... 5.< ....... ............. V.. Arvind.I..v( Johnson Professor of Computer Science Thesis Supervisor 2' Accepted by .. .... .... .. .. .... Arthur C. Smith Chairman, Departmeital Committee on Graduate Students Generating Multithreaded Code from Parallel Haskell for Symmetric Multiprocessors by Alejandro Caro Submitted to the Department of Electrical Engineering and Computer Science on January 27, 1999, in partial fulfillment of the requirements for the degree of Doctor of Philosophy Abstract This dissertation presents pHc, a new compiler for Parallel Haskell (pH) with complete support for the entire language. pH blends the powerful features of a high-level, higher- order language with implicitly parallel, non-strict semantics. The result is a language that is easy to use but tough to compile. The principal contribution of this work is a new code- generation algorithm that works directly on A*, a terse intermediate representation based on the A-calculus. All the power of the original language is preserved in A*, and in addition, the new representation makes it easy to express the important concept of threads, groups of expressions that can execute sequentially, at a high-level.
    [Show full text]
  • The Vector-Thread Architecture
    Appears in, The 31st Annual International Symposium on Computer Architecture (ISCA-31), Munich, Germany, June 2004 The Vector-Thread Architecture Ronny Krashinsky, Christopher Batten, Mark Hampton, Steve Gerding, Brian Pharris, Jared Casper, and Krste Asanovic´ MIT Computer Science and Artificial Intelligence Laboratory, 32 Vassar Street, Cambridge, MA 02139 ronny,cbatten,krste @csail.mit.edu Abstract parallel code, each VP directs its own control flow by fetching its The vector-thread (VT) architectural paradigm unifies the vector own AIBs. Implementations of the VT architecture can also exploit and multithreaded compute models. The VT abstraction provides instruction-level parallelism within AIBs. the programmer with a control processor and a vector of virtual In this way, the VT architecture supports a modeless intermin- processors (VPs). The control processor can use vector-fetch com- gling of all forms of application parallelism. This flexibility pro- mands to broadcast instructions to all the VPs or each VP can use vides new ways to parallelize codes that are difficult to vectorize or thread-fetches to direct its own control flow. A seamless intermix- that incur excessive synchronization costs when threaded. Instruc- ing of the vector and threaded control mechanisms allows a VT ar- tion locality is improved by allowing common code to be factored chitecture to flexibly and compactly encode application parallelism out and executed only once on the control processor, and by execut- and locality, and a VT machine exploits these to improve perfor- ing the same AIB multiple times on each VP in turn. Data locality mance and efficiency. We present SCALE, an instantiation of the is improved as most operand communication is isolated to within VT architecture designed for low-power and high-performance em- an individual VP.
    [Show full text]