• Proceedings of the Korean Information Science Society Conference
• /
• 2000.10c
• /
• pp.636-638
• /
• 2000

제어 독립성(Control Independence)은 슈퍼스칼라 프로세서에서 명령어 수준 병렬성(Instruction-Level Parallelism)을 향상시키기 위한 중요한 요소로 작용하고 있다. 분기 예상기법(Branch Prediction Mechanism)에서 잘못 예상될 경우에는 예상한 분기 방향의 명령어들을 제거하고 올바른 분기 방향의 명령어들을 다시 반입하여 수행해야 한다. 본 논문에서는 컴파일 시 프로파일링을 통한 정적인 방법과 프로그램상의 제어 흐름을 통해 동적으로 제어 독립적인 명령어를 탐지함으로써 분기 명령어의 잘못된 예상으로 인해 제거되는 명령어를 효과적으로 감소시켜 프로세서의 성능을 향상시키는 메커니즘을 제안한다. SPECint95 벤치마크 프로그램에 대해 기존의 방법과 본 논문에서 제안한 방법 사이의 사이클 당 수행된 명령어 수를 분석한 결과, 4-width 프로세서에서 4%~6%, 8-width 프로세서에서 11%~18%, 16-width 프로세서에서 15%~17%의 성능 향상을 보이고 있다.

• The Magazine of the IEIE
• /
• v.28 no.7
• /
• pp.49-55
• /
• 2001

1970년대에 개발된 마이크로 프로세서는 제어기기 분야 및 소형 컴퓨터에서 주로 사용되어 오다가 1980년대에 이르러 RISC(Reduced Instruction Set Computer) 구조의 도입으로 중대형 컴퓨터에 이르기까지 광범위하게 사용되고 있다. 또한 반도체 기술의 급격한 발전으로 슈퍼스칼라 구조가 마이크로 프로세서에서도 적용되고 있으며 동작 속도도 수백 MHz에 이르고 있다. 마이크로 프로세서는 프로그램을 수행하기 위해서 프로그램과 데이터를 메모리로부터 읽어 와야 한다. 그런데 메모리 용량은 빠른 속도로 증가하고 있지만 동작 속도는 마이크로 프로세서의 동작 속도에 크게 미치지 못하고 있다. 1980년에 DRAM의 접근 속도는 250nsec이었으나 1998년에 RDRAM의 동작속도는 300MHz로 70여배 빨라졌다. 그러나 마이크로프로세서는 1980년에 8086의 동작 속도가 8MHz이던 것이 1998년에는 팬티엄-2가 500MHz에 이르고 있다. 더욱이 팬티엄-2는 슈퍼스칼라 구조이므로 이를 감안하면 1GHz 이상에 이르러 120여 배 빨라진 것을 알 수 있다. 이와 같은 메모리 속도와 마이크로 프로세서 속도 차이에 더하여, 메모리와 마이크로 프로세서를 인쇄 회로 기판에서 연결하는데 따른 물리적 특성은 변화하지 않으므로 데이터 전송 폭을 넓히는 것에는 한계가 있다. 따라서 향후 컴퓨터 성능 발달을 제한하는 주요 요소 중 하나는 마이크로 프로세서와 메모리 사이의 데이터 전송 폭이다. 프로그램과 데이터가 메모리에 저장되는 본 뉴먼 방식의 컴퓨터에서 데이터 전송 폭을 줄이기 위해서는 코드 밀도(Code Density)가 높은 컴퓨터 구조를 연구하는 것이 필요하다. 한편 마이크로 프로세서는 실장 제어용으로 거의 모든 전자 제품 및 자동화 기기에서 채용하고 있다. 특히 냉장고, 에어콘, 전축, TV, 세탁기 등 가전기기와 Fax, 복사기, 프린터 등 사무용기기와 자동차, 선박, 자동화기계 등 사무 및 산업용 기기와 PDA(휴대용 정보 기기), NC(Network Computer) 등 정보 기기 그리고 각종 오락기, 노래 반주지 등 정보 기기 등에서 사용하는 실장 제어용 마이크로 프로세서 시장은 매년 10% 이상씩 성장하고 있으며, 21세기 산업을 주도하는 핵심 기술로 자리 매김하고 있다. 이러한 실장 제어용 기기는 마이크로 프로세서와 메모리 및 입출력 자이가 하나의 반도체에 집적되는 경우가 많다. 그런데 반도체 가격은 반도체 크기에 따라 결정되며, 가장 넓은 면적을 차지하는 것은 메모리이다. 따라서 반도체 가격을 낮추기 위해서는 메모리 크기를 줄여야 하며, 이를 위해서 또한 코드 밀도가 높은 컴퓨터 구조에 대한 연구가 필요하다. 최근에는 322비트 RISC 명령어를 16비트 명령어로 축약한 구조가 연구되었다. ARM-7TDMI는 ARM-7의 16비트 축약 명령어 구조이며, TR4101은 MIPS-R3000의 16비트 축약 명령어 구조이다. 이들 16비트 축약 명령어 RISC는 종래 RISC와의 호환성을 위하여 2가지 모드로 동작하므로 구조가 복잡하고, 16비트 명령어에서는 8개의 레지스타만을 접근할 수 있으므로 성능이 크게 떨어지는 단점을 가진다.

• Journal of the Institute of Electronics Engineers of Korea SD
• /
• v.45 no.12
• /
• pp.73-82
• /
• 2008

There are several approaches for reducing the ROB(Reorder Buffer) complexity in processors. The one technique that makes the simplest ROB ports relies on a distributed implementation that spreads the centralized ROB structure across the functional units(FUs). Each distributed buffers are decided on the size of them by workload of the functional units. The performance of the processor depends on the size of distributed ROB. However, most of previous works have depended on the simulation results to decide the optimsize of distributed ROB. In this Paper, we use an analytical model based on the M/M/1 Queuing theory to determine the optimum size of each distributed ROB. Our schemes are evaluated by using the simulation performed by the CPU2000 benchmarks. We are able to choose the optimum size of distributed ROB showing the 99.2% performance compared with existing superscalar processors. We can save 82% hardware resources in ports and reduce more than 30% of delay when ROB and distributed ROB proposed in this paper are designed by HDL.

• Proceedings of the IEEK Conference
• /
• 2000.06c
• /
• pp.55-58
• /
• 2000

To achieve high performance by exploiting instruction level parallelism(ILP) aggressively in superscalar processors, value prediction is used. Value prediction is a technique that breaks data dependences by predicting the outcome of an instruction and executes speculatively it's data dependent instruction based on the predicted outcome. In this paper, the performance of a hybrid value prediction scheme with dynamic classification mechanism is measured and analyzed by using execution-driven simulator for SPECint95 benchmark set.

• The Journal of the Institute of Internet, Broadcasting and Communication
• /
• v.9 no.2
• /
• pp.97-103
• /
• 2009

In this paper, the performance of a hybrid value predictor according to the instruction fetch rate on window size superscalar processors is evaluated. In general, the data dependency relations of instructions are increased with the number of the fetched instructions. Therefore, it is expected that the performance of a value predictor will be higher when the instruction fetch rate is increased. The performance is studied for the machine with collapsing buffer and he one with trace cache as an instruction fetch mechanism. As a result of experiment, it is showed that the performance effect of a value predictor is higher as the instruction fetch rate of instruction window size, IPC, predict rate on apply with non-tc and tc is increased.

• Journal of KIISE:Computer Systems and Theory
• /
• v.27 no.8
• /
• pp.741-751
• /
• 2000

To achieve high performance by exploiting instruction level parallelism aggressively in superscalar processors, it is necessary to overcome the limitation imposed by control dependences and data dependences which prevent instructions from executing parallel. Value prediction is a technique that breaks data dependences by predicting the outcome of an instruction and executes speculatively its data dependent instruction based on the predicted outcome. In this paper, a hybrid value prediction scheme with dynamic classification mechanism is proposed. We design a hybrid predictor by combining the last predictor, a stride predictor and a two-level predictor. The choice of a predictor for each instruction is determined by a dynamic classification mechanism. This makes each predictor utilized more efficiently than the hybrid predictor without dynamic classification mechanism. To show performance improvements of our scheme, we simulate the SPECint95 benchmark set by using execution-driven simulator. The results show that our scheme effect reduce of 45% hardware cost and 16% prediction accuracy improvements comparing with the conventional hybrid prediction scheme and two-level value prediction scheme.

• Journal of KIISE:Computer Systems and Theory
• /
• v.28 no.12
• /
• pp.685-697
• /
• 2001

Cache memories are small fast memories used to temporarily hold the contents of main memory that are likely to be referenced by processors so as to reduce instruction and data access time. In this paper, we represent analytical models of instruction fetch process for four types of instruction cache structures that can be used for superscalar processors. In the models, we define various kinds of architectural parameters and take cache miss and branch misprediction into consideration. To prove the correctness of the proposed models, we performed extensive simulations and compared the results with the analytical models. Simulation results showed that the proposed model can estimate the instruction fetch rate accurately within 10% error in most cases. Both analytical model and simulation show that the increase of cache misses reduces the instruction fetch rate more severely than that of branch misprediction does. However, the analytical model can explain the causes of performance degradation which cannot be uncovered by the simulation method only. The model is also able to provide exact relationship between cache miss and branch misprediction for instruction fetch analysis.

• Journal of KIISE:Computing Practices and Letters
• /
• v.14 no.8
• /
• pp.768-772
• /
• 2008

Today's microprocessor normally includes several processing cores to reduce the energy consumption without losing performance. In this paper, data transfer ordering mechanism can be efficiently used for cache coherence solution in unidirectional ring interconnect. RING-DATA ORDER combines the simplicity of GREEDY-ORDER and the performance of RING-ORDER. RING-DATA ORDER can be easily applicable to multicore processor with unidirectional ring interconnect.

• Journal of KIISE:Computing Practices and Letters
• /
• v.13 no.5
• /
• pp.252-259
• /
• 2007

With the rapid development of multimedia applications and wireless communication networks, consumer demand for video-over-wireless capability on mobile computing systems is growing rapidly. In this regard, this paper introduces YUV-aware instructions that enhance the performance and efficiency in the processing of color image and video. Traditional multimedia extensions (e.g., MMX, SSE, VIS, and AltiVec) depend solely on generic subword parallelism whereas the proposed YUV-aware instructions support parallel operations on two-packed 16-bit YUV (6-bit Y, 5-bits U, V) values in a 32-bit datapath architecture, providing greater concurrency and efficiency for color image and video processing. Moreover, the ability to reduce data format size reduces system cost. Experiment results on a representative dynamically scheduled embedded superscalar processor show that YUV-aware instructions achieve an average speedup of 3.9x over the baseline superscalar performance. This is in contrast to MMX (a representative Intel#s multimedia extension), which achieves a speedup of only 2.1x over the same baseline superscalar processor. In addition, YUV-aware instructions outperform MMX instructions in energy reduction (75.8% reduction with YUV-aware instructions, but only 54.8% reduction with MMX instructions over the baseline).

• Journal of the Institute of Electronics Engineers of Korea CI
• /
• v.43 no.2 s.308
• /
• pp.1-11
• /
• 2006

In this paper, we propose a new register file architecture called the Register File Extension for Multi-words or Long-word Operation (RFEMLO) to accelerate both symmetric and asymmetric cryptographic algorithms. Based on the idea that most of cryptographic algorithms heavily use multi-words or long-word operations, RFEMLO allows multiple contiguous registers to be specified as a single operand. Thus, a single instruction can specify a SIMD-style multi-word operation or a long-word operation. RFEMLO can be applied to general purpose processors by adding instruction set for multi-words or long-word operands and functional units for additional instruction set. To evaluate the performance of RFEMLO, we use Simplescalar/ARM 3.0 (with gcc 2.95.2) and run detailed simulations on various symmetric and asymmetric cryptographic algorithms. By applying RFEMLO, we could get maximum 62% and 70% reductions in the total instruction count of symmetric and asymmetric cryptographic algorithms respectively. Also, performance results show that a speedup of 1.4 to 2.6 can be obtained in symmetric cryptographic algorithms and a speedup of 2.5 to 3.3 can be obtained for asymmetric cryptographic algorithms when we apply RFEMLO to a processor with an in-order pipeline. We also found that RFEMLO can effectively improve the performance of these cryptographic algorithms with much less cost compared to issue-width increase available in Superscalar implementations. Moreover, the RFEMLO can also be applied to Superscalar processor, leading to additional 83% and 138% performance gain in symmetric and asymmetric cryptographic algorithms.