We investigate the utilization of additive prescaling for reducing the input range for determining a divisor reciprocal approximation. In particular, we demonstrate that a single precision (24-bit) ulp accurate monotonic approximate reciprocal operation can be obtained employing three back-to-back three term additions and bipartite table lookup with total table size less than 6 Kbytes.
{"title":"A Prescale-Lookup-Postscale Additive Procedure for Obtaining a Single Precision Ulp Accurate Reciprocal","authors":"D. Matula, Mihai T. Panu","doi":"10.1109/ARITH.2011.31","DOIUrl":"https://doi.org/10.1109/ARITH.2011.31","url":null,"abstract":"We investigate the utilization of additive prescaling for reducing the input range for determining a divisor reciprocal approximation. In particular, we demonstrate that a single precision (24-bit) ulp accurate monotonic approximate reciprocal operation can be obtained employing three back-to-back three term additions and bipartite table lookup with total table size less than 6 Kbytes.","PeriodicalId":272151,"journal":{"name":"2011 IEEE 20th Symposium on Computer Arithmetic","volume":"5 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2011-07-25","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"132865746","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
The implementation of merged floating-point multiply-add operations can be optimized in many ways. For latency sensitive applications, our cascade design reduces the accumulation dependent latency by 2x over a fused design, at a cost of a 13% increase in non-accumulation dependent latency. A simple in-order execution model shows this design is superior in most applications, providing 12% average reduction in FP stalls, and improves performance by up to 6%. Simulations of superscalar out-of-order machines show 4% average improvement in CPI in 2-way machines and 4.6% in 4-way machines. The cascade design has the same area and energy budget as a traditional fused multiple-add FMA.
{"title":"Latency Sensitive FMA Design","authors":"Sameh Galal, M. Horowitz","doi":"10.1109/ARITH.2011.26","DOIUrl":"https://doi.org/10.1109/ARITH.2011.26","url":null,"abstract":"The implementation of merged floating-point multiply-add operations can be optimized in many ways. For latency sensitive applications, our cascade design reduces the accumulation dependent latency by 2x over a fused design, at a cost of a 13% increase in non-accumulation dependent latency. A simple in-order execution model shows this design is superior in most applications, providing 12% average reduction in FP stalls, and improves performance by up to 6%. Simulations of superscalar out-of-order machines show 4% average improvement in CPI in 2-way machines and 4.6% in 4-way machines. The cascade design has the same area and energy budget as a traditional fused multiple-add FMA.","PeriodicalId":272151,"journal":{"name":"2011 IEEE 20th Symposium on Computer Arithmetic","volume":"19 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2011-07-25","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"125719950","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
High performance microprocessors are protected against transient and early end of life failures using a variety of error detection and fault isolation technologies. Execution units can be protected with duplication, parity prediction, or residue checking. Residue checking has an advantage due to its small size. A modulus is selected based on the radix of the numbers being checked. In a decimal floating-point unit there are two types of numbers in different bases. There are base 10 decimal numbers and base 2 integers being used. A residue checking system that makes it easy to check both base 2 and 10 numbers is discussed. Current state of the art designs that are currently in use are described as well as a novel hybrid moduli 9 and 3 residue system. The checking systems for the decimal and binary floating-point units of some recent IBM microprocessors including the Power6, Power7, z10, and z196 microprocessors are detailed.
{"title":"Self Checking in Current Floating-Point Units","authors":"Daniel Lipetz, E. Schwarz","doi":"10.1109/ARITH.2011.18","DOIUrl":"https://doi.org/10.1109/ARITH.2011.18","url":null,"abstract":"High performance microprocessors are protected against transient and early end of life failures using a variety of error detection and fault isolation technologies. Execution units can be protected with duplication, parity prediction, or residue checking. Residue checking has an advantage due to its small size. A modulus is selected based on the radix of the numbers being checked. In a decimal floating-point unit there are two types of numbers in different bases. There are base 10 decimal numbers and base 2 integers being used. A residue checking system that makes it easy to check both base 2 and 10 numbers is discussed. Current state of the art designs that are currently in use are described as well as a novel hybrid moduli 9 and 3 residue system. The checking systems for the decimal and binary floating-point units of some recent IBM microprocessors including the Power6, Power7, z10, and z196 microprocessors are detailed.","PeriodicalId":272151,"journal":{"name":"2011 IEEE 20th Symposium on Computer Arithmetic","volume":"10 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2011-07-25","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"122774776","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
S. Carlough, Adam Collura, S. M. Müller, M. Kroener
Decimal floating-point Arithmetic is widely used in commercial computing applications, such as financial transactions, where rounding errors prevent the use of binary floating-point operations. The revised IEEE Standard for Floating-Point Arithmetic (IEEE-754-2008) defined standardized decimal floating-point (DFP) formats. As more software applications adopt the IEEE decimal floating-point standard, hardware accelerators that support it are becoming more prevalent. This paper describes the second generation decimal floating-point accelerator implemented on the IBM zEnterprise-196 processor. The 4-cycle deep pipeline was designed to optimize the latency of fixed-point decimal operations while significantly improving the bandwidth of DFP operations. A detailed description of the unit and a comparison to previous implementations found in literature is provided.
{"title":"The IBM zEnterprise-196 Decimal Floating-Point Accelerator","authors":"S. Carlough, Adam Collura, S. M. Müller, M. Kroener","doi":"10.1109/ARITH.2011.27","DOIUrl":"https://doi.org/10.1109/ARITH.2011.27","url":null,"abstract":"Decimal floating-point Arithmetic is widely used in commercial computing applications, such as financial transactions, where rounding errors prevent the use of binary floating-point operations. The revised IEEE Standard for Floating-Point Arithmetic (IEEE-754-2008) defined standardized decimal floating-point (DFP) formats. As more software applications adopt the IEEE decimal floating-point standard, hardware accelerators that support it are becoming more prevalent. This paper describes the second generation decimal floating-point accelerator implemented on the IBM zEnterprise-196 processor. The 4-cycle deep pipeline was designed to optimize the latency of fixed-point decimal operations while significantly improving the bandwidth of DFP operations. A detailed description of the unit and a comparison to previous implementations found in literature is provided.","PeriodicalId":272151,"journal":{"name":"2011 IEEE 20th Symposium on Computer Arithmetic","volume":"35 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2011-07-25","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"123775259","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Fixed-point Fast Fourier Transform (FFT) units are widely used in digital communication systems. The twiddle multipliers required for realizing large FFTs are typically implemented with the Coordinate Rotation Digital Computer (CORDIC) algorithm to restrict memory requirements. Recent approaches aiming to optimize the bit-widths of FFT units while satisfying a given maximum bound on Mean-Square-Error (MSE) mostly focus on the architectures with integer multipliers. They ignore the quantization error of coefficients, disabling them to analyze the exact error defined as the difference between the fixed-point circuit and the reference floating-point model. This paper presents an efficient analysis of MSE as well as an optimization algorithm for CORDIC-based FFT units, which is applicable to other Linear-Time-Invariant (LTI) circuits as well.
{"title":"On the Fixed-Point Accuracy Analysis and Optimization of FFT Units with CORDIC Multipliers","authors":"O. Sarbishei, K. Radecka","doi":"10.1109/ARITH.2011.17","DOIUrl":"https://doi.org/10.1109/ARITH.2011.17","url":null,"abstract":"Fixed-point Fast Fourier Transform (FFT) units are widely used in digital communication systems. The twiddle multipliers required for realizing large FFTs are typically implemented with the Coordinate Rotation Digital Computer (CORDIC) algorithm to restrict memory requirements. Recent approaches aiming to optimize the bit-widths of FFT units while satisfying a given maximum bound on Mean-Square-Error (MSE) mostly focus on the architectures with integer multipliers. They ignore the quantization error of coefficients, disabling them to analyze the exact error defined as the difference between the fixed-point circuit and the reference floating-point model. This paper presents an efficient analysis of MSE as well as an optimization algorithm for CORDIC-based FFT units, which is applicable to other Linear-Time-Invariant (LTI) circuits as well.","PeriodicalId":272151,"journal":{"name":"2011 IEEE 20th Symposium on Computer Arithmetic","volume":"13 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2011-07-25","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"126482062","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
This work describes the carry-compact addition (CCA), a novel addition scheme that allows the acceleration of carry-chain computations on contemporary FPGA devices. While based on concepts known from the carry-look ahead addition and from parallel prefix adders, their adaptation by the CCA takes the context of an FPGA as implementation environment into account. These typically provide carry-chain structures to accelerate the simple ripple-carry addition (RCA). Rather than contrasting this scheme with the hierarchical addition approaches favored in hard-core VLSI designs, the CCA combines the benefits of both and uses hierarchical structures to shorten the critical path, which is still left on a core carry chain. In contrast to previous studies examining the asymptotically superior parallel prefix adders on FPGAs, the CCA is shown to outperform the standard RCA already for operand widths starting at 50~bits. Wider adders such as used in extended-precision floating-point units and in cryptographic applications even benefit from increasing speedups. The concrete mapping of the CCA as achieved for current Xilinx and Altera architectures is described and shown to be very favorable so as to yield a high speedup for a very modest investment of additional LUT resources.
{"title":"Accelerating Computations on FPGA Carry Chains by Operand Compaction","authors":"Thomas B. Preußer, M. Zabel, R. Spallek","doi":"10.1109/ARITH.2011.22","DOIUrl":"https://doi.org/10.1109/ARITH.2011.22","url":null,"abstract":"This work describes the carry-compact addition (CCA), a novel addition scheme that allows the acceleration of carry-chain computations on contemporary FPGA devices. While based on concepts known from the carry-look ahead addition and from parallel prefix adders, their adaptation by the CCA takes the context of an FPGA as implementation environment into account. These typically provide carry-chain structures to accelerate the simple ripple-carry addition (RCA). Rather than contrasting this scheme with the hierarchical addition approaches favored in hard-core VLSI designs, the CCA combines the benefits of both and uses hierarchical structures to shorten the critical path, which is still left on a core carry chain. In contrast to previous studies examining the asymptotically superior parallel prefix adders on FPGAs, the CCA is shown to outperform the standard RCA already for operand widths starting at 50~bits. Wider adders such as used in extended-precision floating-point units and in cryptographic applications even benefit from increasing speedups. The concrete mapping of the CCA as achieved for current Xilinx and Altera architectures is described and shown to be very favorable so as to yield a high speedup for a very modest investment of additional LUT resources.","PeriodicalId":272151,"journal":{"name":"2011 IEEE 20th Symposium on Computer Arithmetic","volume":"3 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2011-07-25","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"127989783","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Joppe W. Bos, T. Kleinjung, A. Lenstra, P. L. Montgomery
This paper describes carry-less arithmetic operations modulo an integer 2^M-1 in the thousand-bit range, targeted at single instruction multiple data platforms and applications where overall throughput is the main performance criterion. Using an implementation on a cluster of PlayStation 3 game consoles a new record was set for the elliptic curve method for integer factorization.
{"title":"Efficient SIMD Arithmetic Modulo a Mersenne Number","authors":"Joppe W. Bos, T. Kleinjung, A. Lenstra, P. L. Montgomery","doi":"10.1109/ARITH.2011.37","DOIUrl":"https://doi.org/10.1109/ARITH.2011.37","url":null,"abstract":"This paper describes carry-less arithmetic operations modulo an integer 2^M-1 in the thousand-bit range, targeted at single instruction multiple data platforms and applications where overall throughput is the main performance criterion. Using an implementation on a cluster of PlayStation 3 game consoles a new record was set for the elliptic curve method for integer factorization.","PeriodicalId":272151,"journal":{"name":"2011 IEEE 20th Symposium on Computer Arithmetic","volume":"68 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2011-07-25","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"130783691","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
User requirements for signal processing have increased in line with, or greater than, the increase in FPGA resources and capability. Many current signal processing algorithms require floating point, especially for military applications such as radar. Also, the increasing system complexity of these designs necessitate increased designer productivity, and floating point allows an easier implementation of the system model than the fixed point arithmetic that FPGA devices have been traditionally architected for. This article will review devices and methods for achieving consistent high performance system implementations in floating point. Single device designs at over 200 GFLOPs at the 40nm node, and approaching 1 Teraflop at 28nm will be described.
{"title":"Teraflop FPGA Design","authors":"M. Langhammer","doi":"10.1109/ARITH.2011.32","DOIUrl":"https://doi.org/10.1109/ARITH.2011.32","url":null,"abstract":"User requirements for signal processing have increased in line with, or greater than, the increase in FPGA resources and capability. Many current signal processing algorithms require floating point, especially for military applications such as radar. Also, the increasing system complexity of these designs necessitate increased designer productivity, and floating point allows an easier implementation of the system model than the fixed point arithmetic that FPGA devices have been traditionally architected for. This article will review devices and methods for achieving consistent high performance system implementations in floating point. Single device designs at over 200 GFLOPs at the 40nm node, and approaching 1 Teraflop at 28nm will be described.","PeriodicalId":272151,"journal":{"name":"2011 IEEE 20th Symposium on Computer Arithmetic","volume":"37 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2011-07-25","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"122410238","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
The performance of many cryptographic primitives is reliant on efficient algorithms and implementation techniques for arithmetic in binary fields. While dedicated hardware support for said arithmetic is an emerging trend, the study of software-only implementation techniques remains important for legacy or non-equipped processors. One such technique is that of software-based bit-slicing. In the context of binary fields, this is an interesting option since there is extensive previous work on bit-oriented designs for arithmetic in hardware, such designs are intuitively well suited to bit-slicing in software. In this paper we harness previous work, using it to investigate bit-sliced, software-only implementation arithmetic for binary fields, over a range of practical field sizes and using a normal basis representation. We apply our results to demonstrate significant performance improvements for a stream cipher, and over the frequently employed Ning-Yin approach to normal basis implementation in software.
{"title":"Bit-Sliced Binary Normal Basis Multiplication","authors":"B. Brumley, D. Page","doi":"10.1109/ARITH.2011.36","DOIUrl":"https://doi.org/10.1109/ARITH.2011.36","url":null,"abstract":"The performance of many cryptographic primitives is reliant on efficient algorithms and implementation techniques for arithmetic in binary fields. While dedicated hardware support for said arithmetic is an emerging trend, the study of software-only implementation techniques remains important for legacy or non-equipped processors. One such technique is that of software-based bit-slicing. In the context of binary fields, this is an interesting option since there is extensive previous work on bit-oriented designs for arithmetic in hardware, such designs are intuitively well suited to bit-slicing in software. In this paper we harness previous work, using it to investigate bit-sliced, software-only implementation arithmetic for binary fields, over a range of practical field sizes and using a normal basis representation. We apply our results to demonstrate significant performance improvements for a stream cipher, and over the frequently employed Ning-Yin approach to normal basis implementation in software.","PeriodicalId":272151,"journal":{"name":"2011 IEEE 20th Symposium on Computer Arithmetic","volume":"182 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2011-07-25","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"133670094","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
The hardware implementation of modular exponentiation for very large integers is a well-known topic in digital arithmetic. An effective approach for obtaining parallel and carry-free implementations consists in using the Montgomery exponentiation algorithm and executing the necessary operations in RNS. Two efficient methods for performing the RNS Montgomery exponentiation have been proposed by Kawamura et al. and by Bajard and Imbert. The above approaches mainly differ in the algorithm used for implementing the base extension. This paper presents a modified RNS Montgomery exponentiation algorithm, where several multiplications are moved outside the main execution loop and replaced by an effective pre-processing stage producing a significant saving on the overall delay with respect to state-of-the-art approaches. Since the proposed modification should be applied to both of the above algorithms, two versions are specifically discussed.
{"title":"A General Approach for Improving RNS Montgomery Exponentiation Using Pre-processing","authors":"F. Gandino, F. Lamberti, P. Montuschi, J. Bajard","doi":"10.1109/ARITH.2011.35","DOIUrl":"https://doi.org/10.1109/ARITH.2011.35","url":null,"abstract":"The hardware implementation of modular exponentiation for very large integers is a well-known topic in digital arithmetic. An effective approach for obtaining parallel and carry-free implementations consists in using the Montgomery exponentiation algorithm and executing the necessary operations in RNS. Two efficient methods for performing the RNS Montgomery exponentiation have been proposed by Kawamura et al. and by Bajard and Imbert. The above approaches mainly differ in the algorithm used for implementing the base extension. This paper presents a modified RNS Montgomery exponentiation algorithm, where several multiplications are moved outside the main execution loop and replaced by an effective pre-processing stage producing a significant saving on the overall delay with respect to state-of-the-art approaches. Since the proposed modification should be applied to both of the above algorithms, two versions are specifically discussed.","PeriodicalId":272151,"journal":{"name":"2011 IEEE 20th Symposium on Computer Arithmetic","volume":"15 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2011-07-25","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"134339839","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
京公网安备 11010802042870号
京ICP备2023020795号-1
扫码关注我们
求助内容：
应助结果提醒方式：