[ Previous | Next | Contents | Home | Search ]
AIX Version 4.3 Assembler Language Reference

Floating-Point Processor

The POWER and PowerPC floating-point processors have the same register set for nonprivileged instructions. The registers are:

The floating-point processor provides high-performance execution of floating-point operations. Instructions are provided to perform arithmetic, comparison, and other operations in floating-point registers, and to move floating-point data between storage and the floating-point registers.

PowerPC and POWER2 also support conversion operations in floating-point registers.

Floating-Point Numbers

A floating-point number consists of a signed exponent and a signed significand, and expresses a quantity that is the product of the signed fraction and the number 2**exponent. Encodings are provided in the data format to represent:

Operations involving infinities produce results obeying traditional mathematical conventions. NaNs have no mathematical interpretation. Their encoding permits a variable diagnostic information field. They may be used to indicate uninitialized variables and can be produced by certain invalid operations.

Interpreting the Contents of a Floating-Point Register

There are thirty-two 64-bit floating-point registers, numbered from floating-point register 0-31. All floating-point instructions provide a 5-bit field that specifies which floating-point registers to use in the execution of the instruction. Every instruction that interprets the contents of a floating-point register as a floating-point value uses the double-precision floating-point format for this interpretation.

All floating-point instructions other than loads and stores are performed on operands located in floating-point registers and place the results in a floating-point register. The Floating-Point Status and Control Register and the Condition Register maintain status information about the outcome of some floating-point operations.

Load and store double instructions transfer 64 bits of data without conversion between storage and a floating-point register in the floating-point processor. Load single instructions convert a stored single floating-format value to the same value in double floating format and transfer that value into a floating-point register. Store single instructions do the opposite, converting valid single-precision values in a floating-point register into a single floating-format value, prior to storage.

Floating-Point Load and Store Instructions

Floating-point load instructions for single and double precision are provided. Double-precision data is loaded directly into a floating-point register. The processor converts single-precision data to double precision prior to loading the data into a floating-point register, since the floating-point registers support only floating-point double-precision operands.

Floating-point store instructions for single and double precision are provided. Single-precision stores convert floating-point register contents to single precision prior to storage.

POWER2 provides load and store floating-point quad instructions. These are primarily to improve the performance of arithmetic operations on large volumes of numbers, such as array operations. Data access is normally a performance bottleneck for these types of operations. These instructions transfer 128 bits of data, rather than 64 bits, in one load or store operation (that is, one storage reference). The 128 bits of data is treated as two doubleword operands, not as one quadword operand.

Floating-Point Move Instructions

Floating-point move instructions copy data from one FPR to another, with data modification as described for each particular instruction. These instructions do not modify the FPSCR.

Floating-Point Arithmetic Instructions

Floating-point arithmetic instructions perform arithmetic operations on floating-point data contained in floating-point registers.

Floating-Point Multiply-Add Instructions

Floating-point multiply-add instructions combine a multiply operation and an add operation without an intermediate rounding operation. The fractional part of the intermediate product is 106 bits wide, and all 106 bits are used in the add or subtract portion of the instruction.

Floating-Point Compare Instructions

Floating-point compare instructions perform ordered and unordered comparisons of the contents of two FPRs. The CR field specified by the BF field is set based on the result of the comparison. The comparison sets one bit of the designated CR field to 1, and sets all other bits to 0. The Floating-Point Condition Code (FPCC) (bits 16:19) is set in the same manner.

The CR field and the FPCC are interpreted as follows:

Condition-Register Field and Floating-Point Condition Code Interpretation
Bit Name Description
0 FL (FRA) < (FRB)
1 FG (FRA) > (FRB)
2 FE (FRA) = (FRB)
3 FU (FRA) ? (FRB) (unordered)

Floating-Point Conversion Instructions

Floating-point conversion instructions are only provided for PowerPC and POWER2. These instructions convert a floating-point operand in an FPR into a 32-bit signed fixed-point integer. The CR1 field and the FPSCR are altered.

Floating-Point Status and Control Register Instructions

Floating-Point Status and Control Register Instructions manipulate data in the FPSCR.

Related Information

Processing and Storage: Overview.

Overview of POWER and PowerPC Architecture.

Branch Processor.

Fixed-Point Processor.

PowerPC Architecture.

[ Previous | Next | Contents | Home | Search ]