Superscalar processors improve performance by exploiting instruction-level parallelism (ILP). ILP in a basic block is, however, not sufficient on non-numerical applications for gaining substantial speedup. Instructions across branches are required to be executed in parallel to dramatically improve performance. That is, speculative execution is strongly required. Boosting is a general solution to achieving speculative execution. Boosting labels an instruction to be speculatively executed, and the hardware handles side-effects. This paper describes the efficient implementation of boosting in terms of cost/performance trade-offs. Our policy in implementation is beneficial in code scheduling heuristics, penalties imposed by code duplication to maintain program semantics, and area cost. This paper also describes a branch scheme which minimizes branch penalty. Branch delay causes crucial penalties on the performance of superscalar processors since multiple delay slots exist even in a single delay cycle. Our scheme is the fetching of both sequential and target instructions, and either of them is selected on a branch. No delay cycle can be imposed. This scheme is realized by a combination of static code movement and hardware support. As a result, we reduce branch penalty with small cost. Simulation results show that our ideas are highly effective in improving the performance of a superscalar processor.

Instruction pre-execution is an effective way to prefetch data. We previously proposed an instruction pre-execution scheme, which we call two-step physical register deallocation (TSD). The TSD realizes pre-execution by exploiting the difference between the amount of instruction-level parallelism available with an unlimited number of physical registers and that available with an actual number of physical registers. Although previous TSD study has successfully improved performance, it still has an inefficient energy consumption. This is because attempts are made for instructions to be pre-executed as much as possible, independently of whether or not they can significantly contribute to load latency reduction, allowing for maximal performance improvement. This paper presents a scheme that improves the energy efficiency of the TSD by pre-executing only those instructions that have great benefit. Our evaluation results using the SPECfp2000 benchmark show that our scheme reduces the dynamic pre-executed instruction count by 76%, compared with the original scheme. This reduction saves 7% energy consumption of the execution core with 2% overhead. Performance degrades by 2%, compared with that of the original scheme, but is still 15% higher than that of the normal processor without the TSD.

Dynamic instruction window resizing (DIWR) is a scheme that effectively exploits both memory-level parallelism and instruction-level parallelism by configuring the instruction window size appropriately for exploiting each parallelism. Although a previous study has shown that the DIWR processor achieves a significant speedup, power consumption has not been explored. The power consumption is increased in DIWR because the instruction window resources are enlarged in memory-intensive phases. If the power consumption exceeds the power budget determined by certain requirements, the DIWR processor must save power and thus, the performance previously presented cannot be achieved. In this paper, we explore to what extent the DIWR processor can achieve improved performance for a given power budget, assuming that dynamic voltage and frequency scaling (DVFS) is introduced as a power saving technique. Evaluation results using the SPEC2006 benchmark programs show that the DIWR processor, even with a constrained power budget, achieves a speedup over the conventional processor over a wide range of given power budgets. At the most important power budget point, i.e., when the power a conventional processor consumes without any power constraint is supplied, DIWR achieves a 16% speedup.

Two-step physical register deallocation (TSD) is an architectural scheme that enhances memory-level parallelism (MLP) by pre-executing instructions. Ideally, TSD allows exploitation of MLP under an unlimited number of physical registers, and consequently only a small register file is needed for MLP. In practice, however, the amount of MLP exploitable is limited, because there are cases where either 1) pre-execution is not performed; or 2) the timing of pre-execution is delayed. Both are due to data dependencies among the pre-executed instructions. This paper proposes the use of value prediction to solve these problems. This paper proposes the use of value prediction to solve these problems. Evaluation results using the SPECfp2000 benchmark confirm that the proposed scheme with value prediction for predicting addresses achieves equivalent IPC, with a smaller register file, to the previous TSD scheme. The reduction rate of the register file size is 21%.

Out-of-order superscalar processors rename register numbers to remove false dependencies between instructions. A renaming logic for register renaming is a high-cost module in a superscalar processor, and it consumes considerable energy. A renamed trace cache (RTC) was proposed for reducing the energy consumption of a renaming logic. An RTC caches and reuses renamed operands, and thus, register renaming can be omitted on RTC hits. However, conventional RTCs suffer from several performance, energy consumption, and hardware overhead problems. We propose a semi-global renamed trace cache (SGRTC) that caches only renamed operands that are short distance from producers outside traces, and solves the problems of conventional RTCs. Evaluation results show that SGRTC achieves 64% lower energy consumption for renaming with a 0.2% performance overhead as compared to a conventional processor.

This paper evaluates the delay of the issue queue in a superscalar processor to aid microarchitectural design, where quick quantification of the complexity of the issue queue is needed to consider the tradeoff between clock cycle time and instructions per cycle. Our study covers two aspects. First, we introduce banking tag RAM, which comprises the issue queue, to reduce the delay. Unlike normal RAM, this is not straightforward, because of the uniqueness of the issue queue organization. Second, we explore and identify the correct critical path in the issue queue. In a previous study, the critical path of each component in the issue queue was summed to obtain the issue queue delay, but this does not give the correct delay of the issue queue, because the critical paths of the components are not connected logically. In the evaluation assuming 32-nm LSI technology, we obtained the delays of issue queues with eight to 128 entries. The process of banking tag RAM and identifying the correct critical path reduces the delay by up to 20% and 23% for 4- and 8-issue widths, respectively, compared with not banking tag RAM and simply summing the critical path delay of each component.

Out-of-order superscalar processors have high performance but consume a large amount of energy for dynamic instruction scheduling. We propose a front-end execution architecture (FXA) for improving the energy efficiency of out-of-order superscalar processors. FXA has two execution units: an out-of-order execution unit (OXU) and an in-order execution unit (IXU). The OXU is the execution core of a common out-of-order superscalar processor. In contrast, the IXU consists only of functional units and a bypass network only. The IXU is placed at the processor front end and executes instructions in order. The IXU functions as a filter for the OXU. Fetched instructions are first fed to the IXU, and the instructions are executed in order if they are ready to execute. The instructions executed in the IXU are removed from the instruction pipeline and are not executed in the OXU. The IXU does not include dynamic scheduling logic, and thus its energy consumption is low. Evaluation results show that FXA can execute more than 50% of the instructions by using IXU, thereby making it possible to shrink the energy-consuming OXU without incurring performance degradation. As a result, FXA achieves both high performance and low energy consumption. We evaluated FXA and compared it with conventional out-of-order/in-order superscalar processors after ARM big.LITTLE architecture. The results show that FXA achieves performance improvements of 7.4% on geometric mean in SPECCPU INT 2006 benchmark suite relative to a conventional superscalar processor (big), while reducing the energy consumption by 17% in the entire processor. The performance/energy ratio (the inverse of the energy-delay product) of FXA is 25% higher than that of a conventional superscalar processor (big) and 27% higher than that of a conventional in-order superscalar processor (LITTLE).

Single-thread performance has not improved much over the past few years, despite an ever increasing transistor budget. One of the reasons for this is that there is a speed gap between the processor and main memory, known as the memory wall. A promising method to overcome this memory wall is aggressive out-of-order execution by extensively enlarging the instruction window resources to exploit memory-level parallelism (MLP). However, simply enlarging the window resources lengthens the clock cycle time. Although pipelining the resources solves this problem, it in turn prevents instruction-level parallelism (ILP) from being exploited because issuing instructions requires multiple clock cycles. This paper proposed a dynamic scheme that adaptively resizes the instruction window based on the predicted available parallelism, either ILP or MLP. Specifically, if the scheme predicts that MLP is available during execution, the instruction window is enlarged and the window resources are pipelined, thereby exploiting MLP. Conversely, if the scheme predicts that less MLP is available, that is, ILP is exploitable for improved performance, the instruction window is shrunk and the window resources are de-pipelined, thereby exploiting ILP. Our evaluation results using the SPEC2006 benchmark programs show that the proposed scheme achieves nearly the best performance possible with fixed-size resources. On average, our scheme realizes a performance improvement of 21% over that of a conventional processor, with additional cost of only 6% of the area of the conventional processor core or 3% of that of the entire processor chip. The evaluation results also show 8% better energy efficiency in terms of 1/EDP (energy-delay product).

The high energy consumption of current processors causes several problems, including a limited clock frequency, short battery lifetime, and reduced device reliability. It is therefore important to reduce the energy consumption of the processor. Among resources in a processor, the issue queue (IQ) is a large consumer of energy, much of which is consumed by the wakeup logic. Within the wakeup logic, the tag comparison that checks source operand readiness consumes a significant amount of energy. This paper proposes an energy reduction scheme for tag comparison, called double-stage tag comparison. This scheme first compares the lower bits of the tag and then, only if these match, compares the higher bits. Because the energy consumption of tag comparison is roughly proportional to the total number of bits compared, energy is saved by reducing this number. However, this sequential comparison increases the delay of the IQ, thereby increasing the clock cycle time. Although this can be avoided by allocating an extra cycle to the issue operation, this in turn degrades the IPC. To avoid IPC degradation, we reconfigure a small number of entries in the IQ, where several oldest instructions that are likely to have an adverse effect on performance reside, to a single stage for tag comparison. Our evaluation results for SPEC2017 benchmark programs show that the double-stage tag comparison achieves on average a 21% reduction in the energy consumed by the wakeup logic (15% when including the overhead) with only 3.0% performance degradation.