公开(公告)号：GB2511314A

公开(公告)日：2014-09-03

申请号：GB201303464

申请日：2013-02-27

Applicant: IBM

Inventor： TRONG SON DAO ,  MULLER SILVIA MELITTA ,  KLEIN MICHAEL ,  LAYER CHRISTOPHE

IPC: G06F7/483

Abstract: Disclosed is a method of operating a fast fused-multiply-add pipeline in a floating point unit of a processor, using a plurality of operands 32, 34, 90 as an input. A data formatting step is performed after a partial product reduction in the pipeline instead of conversion of the input operands 32, 34, 90 to an internal floating point format. The method may comprise the steps of receiving a first and second input operands to be multiplied together, receiving a third operand to be added to the result of the multiplication. The multiplication is performed in a multiplier block to give partial product results, which are input into a carry-save adder block. Next a partial product reduction is performed on the partial product results to generate a carry-save result comprising a sum term and a carry term. Then a data formatting step is carried out on the carry-save result and a carry-out bit is generated. Finally the result is generated by adding the carry-save result to the third operand.

公开(公告)号：GB2506871A

公开(公告)日：2014-04-16

申请号：GB201218112

申请日：2012-10-10

Applicant: IBM

Inventor： LAYER CHRISTOPHE ,  MUELLER SILVIA MELITTA ,  KROENER MICHAEL ,  SCHELM KERSTIN

IPC: G06F7/544 ,  G06F7/49

Abstract: An arithmetic operation, such as an SRT computation of a division, square root, addition, subtraction or multiplication, in a data processing unit (216), preferably by iterative digit accumulations, is proposed. An approximate result of the arithmetic operation is computed iteratively. Concurrently, at least two supplementary values of the approximate result of the arithmetic operation are computed, and the final result selected from one of the values of the approximate result and the at least two supplementary values of the arithmetic operation depending on the results of the last iteration step. A multiplexing unit may select the final result. Iteration may use accumulating digit values concatenated to previous results using a radix.