A method is disclosed for efficiently multiplying de-normalized floating-point numbers without necessarily incurring additional delay over the multiplication of normalized numbers, wherein the de-normalized numbers are initially assumed to be normalized. The method includes providing multiplier and multiplicand floating-point numbers, each defining a fraction and an exponent; encoding the multiplier fraction; multiplexing the multiplier fraction; and multiplying the multiplier fraction by the multiplicand fraction to form a first set of partial products. The method further includes detecting a zero-value multiplier exponent characteristic indicative of de-normalization; computing a last partial product according to the value of the multiplier exponent characteristic; detecting a zero-value multiplicand exponent characteristic indicative of de-normalization; selecting a de-normalized multiplicand correction term according to the detected multiplicand exponent characteristic and the value of the multiplier fraction; and adding each of the first set and the last partial products to the de-normalized multiplicand correction term to form a final product.

A floating point unit 10 provides a multiply-accumulate operation to determine a result B+(A*C). The multiplier 20 takes several processing cycles to determine the product (A*C). Whilst the multiplier 20 and its subsequent carry-save-adder 26 operate, an aligned value B' of the addend B is generated by an alignment-shifter 34. The aligned-addend B' may only partially overlap with the product (A*C) to which it is to be added using an adder 44. Any high-order-portion HOP of the aligned-addend B' that does not overlap with the product (A*C) must be subsequently concatenated with the output of the adder 44 that sums the product (A*C) with the overlapping portion of the aligned-addend B'. If the sum performed by the adder 44 generates a carry then it is an incremented version IHOP of the high-order-portion that should be concatenated with the output of the adder 44. This incremented-high-order-portion is generated by the adder 44 during otherwise idle processing cycles present due to the multiplier 20 operating over multiple cycles.

Aspects for performing a multiply-accumulate operation on floating point numbers in a single clock cycle are described. These aspects include mantissa logic for combining a mantissa portion of floating point inputs and exponent logic coupled to the mantissa logic. The exponent logic adjusts the combination of an exponent portion of the floating point inputs by a predetermined value to produce a shift amount and allows pipeline stages in the mantissa logic, wherein an unnormalized floating point result is produced from the mantissa logic on each clock cycle.

An enhanced floating point unit that supports floating point, integer, and graphics operations by combining the units into a single functional unit is disclosed. The enhanced floating point unit comprises a register file coupled to a plurality of bypass multiplexers. Coupled to the bypass multiplexers are an aligner and a multiplier. And, coupled to the multiplier is an adder that further couples to a normalizer/rounder unit. The normalizer/rounder unit may comprise a normalizer and a rounder coupled in series and or a parallel normalizer/rounder. The enhanced floating point unit supports both integer operations and graphics operations with one functional unit.

A processor having a conditional branch extension of an instruction set architecture which incorporates a set of high performance floating point operations. The instruction set architecture incorporates a variety of data formats including single precision and double precision data formats, as well as the paired-single data format that allows two simultaneous operations on a pair of operands. The extension includes instructions directed to branching if, for example, either one of two condition codes is false or true, if any of three condition codes are false or true, or if any one of four condition codes are false or true.