The demand for flexibility in media processing motivates the use of programmable processors. Stream processing bridges the gap between inflexible special-purpose solutions and current programmable architectures that cannot meet the computational demands of media-processing applications. The central idea behind stream processing is to organize an application into streams and kernels to expose the inherent locality and concurrency in media-processing applications. The performance of the Imagine stream processor on these media application is given.

/pdf/programmable-stream-processors-z6l619i6g3.pdf

Programmable stream processors

Chip multiprocessors (CMPs) substantially increase capacity pressure on the on-chip memory hierarchy while requiring fast access. Neither private nor shared caches can provide both large capacity and fast access in CMPs. We observe that compared to symmetric multiprocessors (SMPs), CMPs change the latency-capacity tradeoff in two significant ways. We propose three novel ideas to exploit the changes: (1) Though placing copies close to requestors allows fast access for read-only sharing, the copies also reduce the already-limited on-chip capacity in CMPs. We propose controlled replication to reduce capacity pressure by not making extra copies in some cases, and obtaining the data from an existing on-chip copy. This option is not suitable for SMPs because obtaining data from another processor is expensive and capacity is not limited to on-chip storage. (2) Unlike SMPs, CMPs allow fast on-chip communication between processors for read-write sharing. Instead of incurring slow access to read-write shared data through coherence misses as do SMPs, we propose in-situ communication to provide fast access without making copies or incurring coherence misses. (3) Accessing neighborsý caches is not as expensive in CMPs as it is in SMPs. We propose capacity stealing in which private data that exceeds a coreýs capacity is placed in a neighboring cache with less capacity demand. To incorporate our ideas, we use a hybrid of private, per-processor tag arrays and a shared data array. Because the shared data array is slow, we employ non-uniform access and distance associativity from previous proposals to hold frequently-accessed data in regions close to the requestor. We extend the previously-proposed Non-uniform access with Replacement And Placement usIng Distance associativity (NuRAPID) to CMPs, and call our cache CMP-NuRAPID. Our results show that for a 4-core CMP with 8 MB cache, CMP-NuRAPID improves performance by 13% over a shared cache and 8% over private caches for three commercial multithreaded workloads.

/pdf/optimizing-replication-communication-and-capacity-allocation-4zso50j1uf.pdf

Optimizing Replication, Communication, and Capacity Allocation in CMPs

This paper is an in-depth review on silicon implementations of threshold logic gates that covers several decades. In this paper, we will mention early MOS threshold logic solutions and detail numerous very-large-scale integration (VLSI) implementations including capacitive (switched capacitor and floating gate with their variations), conductance/current (pseudo-nMOS and output-wired-inverters, including a plethora of solutions evolved from them), as well as many differential solutions. At the end, we will briefly mention other implementations, e.g., based on negative resistance devices and on single electron technologies.

/pdf/vlsi-implementations-of-threshold-logic-a-comprehensive-vqkiwr19v2.pdf

VLSI implementations of threshold logic-a comprehensive survey

This paper presents a built-in soft error resilience (BISER) technique for correcting radiation-induced soft errors in latches and flip-flops. The presented error-correcting latch and flip-flop designs are power efficient, introduce minimal speed penalty, and employ reuse of on-chip scan design-for-testability and design-for-debug resources to minimize area overheads. Circuit simulations using a sub-90-nm technology show that the presented designs achieve more than a 20-fold reduction in cell-level soft error rate (SER). Fault injection experiments conducted on a microprocessor model further demonstrate that chip-level SER improvement is tunable by selective placement of the presented error-correcting designs. When coupled with error correction code to protect in-pipeline memories, the BISER flip-flop design improves chip-level SER by 10 times over an unprotected pipeline with the flip-flops contributing an extra 7-10.5% in power. When only soft errors in flips-flops are considered, the BISER technique improves chip-level SER by 10 times with an increased power of 10.3%. The error correction mechanism is configurable (i.e., can be turned on or off) which enables the use of the presented techniques for designs that can target multiple applications with a wide range of reliability requirements

Sequential Element Design With Built-In Soft Error Resilience

The design of the high end server processor code named Montecito incorporated several ambitious goals requiring innovation. The most obvious being the incorporation of two legacy cores on-die and at the same time reducing power by 23%. This is an effective 325% increase in MIPS per watt which necessitated a holistic focus on power reduction and management. The next challenge in the implementation was to ensure robust and high frequency circuit operation in the 90-nm process generation which brings with it higher leakage and greater variability. Achieving this goal required new methodologies for design, a greatly improved and tunable clock system and a better understanding of our power grid behavior all of which required new circuits and capabilities. The final aspect of circuit design improvement involved the I/O design for our legacy multi-drop system bus. To properly feed the two high frequency cores with memory bandwidth we needed to ensure frequency headroom in the operation of the bus. This was achieved through several innovations in controllability and tuning of the I/O buffers which are discussed as well.

The implementation of a 2-core, multi-threaded itanium family processor

This 64-b microprocessor is the second-generation design of the new Itanium architecture, termed explicitly parallel instruction computing (EPIC). The design seeks to extract maximum performance from EPIC by optimizing the memory system and execution resources for a combination of high bandwidth and low latency. This is achieved by tightly coupling microarchitecture choices to innovative circuit designs and the capabilities of the transistors and wires in the 0.18-/spl mu/m bulk Al metal process. The key features of this design are: a short eight-stage pipeline, 11 sustainable issue ports (six integer, four floating point, half-cycle access level-1 caches, 64-GB/s level-2 cache and 3-MB level-3 cache), all integrated on a 421 mm/sup 2/ die. The chip operates at over 1 GHz and is built on significant advances in CMOS circuits and methodologies. After providing an overview of the processor microarchitecture and design, this paper describes a few of these key enabling circuits and design techniques.

The implementation of the Itanium 2 microprocessor

An apparatus for performing floating-point division and square root computations according to an IEEE rounding standard includes input data alignment circuitry, core iteration circuitry, remainder compare circuitry, and round and select circuitry. The core iteration circuitry includes digit selector circuitry; remainder registers; quotient logic circuitry; remainder formation circuitry; and quotient registers for storing the quotient Q, incremented quotient Q+1, and decremented quotient Q-1. The remainder formation circuitry produces sum and carry bits of the Pj+1 term, which are in turn fed back to the partial remainder registers and used in subsequent iterations. The quotient logic circuitry builds the quotient Q and maintains the respective quotient Q, Q+1, Q-1 registers. The outputs of these registers are fed back to the quotient logic circuitry for use in subsequent iterations. The remainder compare circuitry comprises a remainder comparator and a logic circuit. The remainder comparator receives the sum and carry bits for the Pj+1 terms and outputs the "Sign" and "Zero" bits. These bits are received by the logic circuit along with a rounding mode signal, which is indicative of the selected rounding mode, e.g., shifted or normalized round to nearest, round to zero, or round to infinity. The logic circuit outputs a round select signal that selects a quotient select signal for selecting, as the final rounded quotient, the output of one of the quotient registers Q, Q+1, or Q-1. The round and select circuitry includes a round block for positive remainders and a round block for negative remainders.

Methods and apparatus for performing division and square root computations in a computer

The invention controls maximum average power dissipation by stalling high power instructions through the pipeline of a pipelined processor. A power dissipation controller stalls the high power instructions in order to control the processor's maximum average power dissipation. Preferably, the controller is modeled after a capacitive system with a constant output rate and a throttled input rate: the output rate represents the steady state maximum average power dissipation; while the input rate is stalled based upon current capacity, representing thermal response time. At start-up, the capacity is initialized. Yet for each high power instruction, the capacity increases by a weighted value. Each clock capacity is also decreased by a variable output rate. In particular, a low power operation is inserted to the stage execution circuit where the stall is desired, creating a low power state for that circuit. This stall effectively creates a “hole” at that pipeline stage, thus temporarily reducing power dissipation. The invention takes advantage of the fact that the presence of an instruction at any stage execution circuit dissipates power and that the absence (i.e., a “hole”) of an instruction at any stage dissipates less power. By controlling where and when a hole occurs within the pipeline, the maximum average power dissipation of the processor is controlled.

Systems and methods for variable control of power dissipation in a pipelined processor

A Multiply Accumulate unit, which may be an FMAC for IEEE 754 format numbers, finds A*B±C faster if the multiplier is allowed to assume that it's A and B inputs are always positive, so that it never has to provide a complemented output, and if the C input for the accumulation with the product is also assumed to be positive. The sign magnitude notation of the IEEE 754 format is temporarily exchanged for a positive two's complement notation of the assumed positive values. Notice is taken of the actual signs, and when there is a difference to be formed, either because of addition between numbers having opposite signs, or because of a subtraction between numbers having the same sign, one of the numbers need to be negated (complemented) prior to the addition of C and the product AB. That number can always be C, provided that correct compensatory negation is available after the addition. Such negations of C are accomplished by performing a one's complement followed by a carry-in to the subsequent adder. Each of the complemented and the non-complemented C values are readily available. Their production and the selection of one or the other are operations that overlap the execution of the multiply, and are done in a way that does not increase the path delay through the shifter. The accumulated result will typically need to be normalized, after which it may need a final complement operation to adjust its sign, in accordance with the original signs and whether the accumulation was addition or subtraction. The result may be converted back to the IEEE 754 format in due course.

Faster multiply/accumulator

A clocking system and method are provided for logic blocks having cascaded self-timed dynamic logic gates. The dynamic logic gates are precharged in parallel and collectively perform self-timed logic evaluation on vector inputs to derive a vector output. An evaluation done detector monitors the output of the logic block and determines when the vector output is valid. An edge detector detects the rising and falling edges of an arbitrary periodic timing signal. Finally, a logic block clock generator is set by the edge detector and reset by the evaluation done detector so as to provide precharging signals to the logic block, thereby defining respective precharge periods, and to provide evaluation periods for the self-timed logic evaluations in the logic block. In a specific implementation, the speed of logic evaluations is twice the speed of the system clock.

Glenn T. Colon-Bonet

Papers

The implementation of the Itanium 2 microprocessor

Methods and apparatus for performing division and square root computations in a computer

Systems and methods for variable control of power dissipation in a pipelined processor

Faster multiply/accumulator

Self-timed clocking system and method for self-timed dynamic logic circuits