Performance of field-programmable gate arrays (FPGAs) used for floating-point applications is poor due to the complexity of floating-point arithmetic. Implementing floating-point units (FPUs) on FPGAs consume a large amount of resources. This makes FPGAs less attractive for use in floating-point intensive applications. Therefore, there is a need for embedded FPUs in FPGAs. However, if unutilized, embedded FPUs waste space on the FPGA die. To overcome this issue, we propose a flexible multimode embedded FPU for FPGAs that can be configured to perform a wide range of operations. The floating-point adder and multiplier in our embedded FPU can each be configured to perform one double-precision operation or two single-precision operations in parallel. To increase flexibility further, access to the large integer multiplier, adder and shifters in the FPU is provided. Benchmark circuits were implemented on both a standard Xilinx Virtex-II FPGA and on our FPGA with embedded FPU blocks. The results using our embedded FPUs showed a mean area improvement of 5.5 times and a mean delay improvement of 5.8 times for the double-precision benchmarks, and a mean area improvement of 3.8 times and a mean delay improvement of 4.2 times for the single-precision benchmarks. The embedded FPUs were also shown to provide significant area and delay benefits for fixed-point and integer circuits.

Performance of Field Programmable Gate Arrays (FPGAs) used for floating-point applications is poor due to the complexity of floating-point arithmetic. Implementing floating-point units on FPGAs consume a large amount of resources. This makes FPGAs less attractive for use in floating-point intensive applications. Therefore, there is a need for embedded floating-point units (FPUs) in FPGAs. However, if unutilized, embedded FPUs waste space on the FPGA die. To overcome this issue, we propose a flexible multi-mode embedded FPU for FPGAs that can be configured to perform a wide range of operations. The floating-point adder and multiplier in our embedded FPU can each be configured to perform one double-precision operation or two single-precision operations in parallel. To increase flexibility further, access to the large integer multiplier, adder and shifters in the FPU is provided. Benchmark circuits were implemented on both a standard Xilinx Virtex-II FPGA and on our FPGA with embedded FPU blocks. The results using our embedded FPUs showed a mean area improvement of 5.2 times and a mean delay improvement of 5.8 times for the double-precision benchmarks, and a mean area improvement of 4.4 times and a mean delay improvement of 4.2 times for the single-precision benchmarks.

While application-specific instruction-set processors (ASIPs) have allowed designers to create processors with custom instructions to target specific applications, floating-point (FP) units (FPUs) are still instantiated as noncustomizable general-purpose units, which, if underutilized, wastes area and performance. Therefore, there is a need for custom FPUs for embedded systems. To create a custom FPU, the subset of FP instructions that should be implemented in hardware has to be determined. Implementing more instructions in hardware reduces the cycle count of the application but may lead to increased latency if the critical delay of the FPU increases. Therefore, a balance between the hardware-implemented and the software-emulated instructions, which produces the best performance, must be found. In order to find this balance, a rapid design space exploration was performed to explore the tradeoffs between the area and the performance. In order to reduce the area of the custom FPU, it is desirable to merge the datapaths for each of the FP operations so that redundant hardware is minimized. However, FP datapaths are complex and contain components with varying bit widths; hence, sharing components of different bit widths is necessary. This introduces the problem of bit alignment, which involves determining how smaller resources should be aligned within larger resources when merged. A novel algorithm for solving the bit-alignment problem during datapath merging was developed. Our results show that adding more FP hardware does not necessarily equate to lower runtime if the delays associated with the additional hardware overcomes the cycle count reductions. We found that, with the Mediabench applications, datapath merging with bit alignment reduced area by an average of 22.5%, compared with an average of 14.1% without bit alignment. With the Standard Performance Evaluation Corporation (SPEC) CPU2000 FP (CFP2000) applications, datapath merging with bit alignment reduced area by an average of 7.6%, compared with an average of 3.9% without bit alignment. The less pronounced improvement with the SPEC CFP2000 benchmarks occurs because the SPEC CFP2000 applications predominantly use double-precision operations only. Therefore, there are fewer resources with different bit widths, which benefit less from bit alignment.

While ASIPs have allowed designers to create processors with custom instructions to target specific applications, floating point units are still instantiated as fixed general-purpose units, which wastes area if not fully utilized. Therefore, there is a need for custom FPUs for embedded systems. The creation of a custom FPU requires the selection of a subset of the full floating-point instruction set and the implementation of this subset in hardware, such that the runtime of the application is minimized. To minimize area, it is desirable to merge the datapaths for each of the floating-point operations, so that redundant hardware is minimized. Floating-point datapaths are complex and contain components with varying bit-widths, so sharing components of different bit-widths is necessary. However, this introduces the problem of bit-alignment, which involves determining how smaller resources should be aligned within larger resources when merged. This is a problem that has been largely neglected in previous work. Thus, this paper presents a novel algorithm for solving the bit-alignment problem, which neatly integrates into the datapath merging process. By solving this bit-alignment problem, automatic datapath merging can be made available for FPU generation. To explore the trade-offs between area and performance, a rapid design space exploration was performed to determine which FP operations should be implemented in hardware rather than emulated. Our results show that more floating-point hardware does not necessarily equate to lower run-time if the additional hardware increases delay. We found that bit-alignment reduced area by an average of 22.5% in our benchmarks, compared to an average of 14.1% without bit-alignment.

This paper describes the creation of custom floating point units (FPUs) for application specific instruction set processors (ASIPs). ASIPs allow the customization of processors for use in embedded systems by extending the instruction set, which enhances the performance of an application or a class of applications. These extended instructions are manifested as separate hardware blocks, making the creation of any necessary floating point instructions quite unwieldy. On the other hand, using a predefined FPU includes a large monolithic hardware block with considerable number of unused instructions. A customized FPU will overcome these drawbacks, yet the manual creation of one is a time consuming, error prone process. This paper presents a methodology for automatically generating floating-point units (FPUs) that are customized for specific applications at the instruction level. Generated FPUs comply with the IEEE754 standard, which is an advantage over FP format customization. Custom FPUs were generated for several Mediabench applications. Area savings over a fully-featured FPU without resource sharing of 26%-80% without resource sharing and 33%-87% with resource sharing, were obtained. Clock period increased in some cases by up to 9.5% due to resource sharing