The arm floating point architecture presents an

Info iconThis preview shows page 1. Sign up to view the full content.

View Full Document Right Arrow Icon
This is the end of the preview. Sign up to access the rest of the document.

Unformatted text preview: ons). FPAIOdata types The ARM FPA10 hardware floating-point accelerator supports single, double and extended double precision formats. Packed decimal formats are supported only by software. The coprocessor registers are all extended double precision, and all internal calculations are carried out in this, the highest precision format, except that there are faster versions of some instructions which do not produce the full 80-bit accuracy. However loads and stores between memory and these registers can convert the precision as required. 164 Architectural Support for High-Level Languages This is similar to the treatment of integers in the ARM integer architecture, where all internal operations are on 32-bit quantities, but memory transfers can specify bytes and half-words. Load and store floating instructions Since there are only eight floating-point registers, the register specifier field in the coprocessor data transfer instruction (shown in Figure 5.16 on page 138) has a spare bit which is used here as an additional data size specifier: Figure 6.7 Load and store floating binary encoding. The other fields in this format are described in Section 5.18 on page 138. The X and Y bits allow one of four precisions to be specified, choosing between single, double, double extended and packed decimal. (The choice between packed decimal and extended packed decimal is controlled by a bit in the FPSR.) Load and store multiple floating The load and store multiple floating-point registers instructions are used to save and restore the floating-point register state. Each register is saved using three memory words, and the precise format is not defined; it is intended that the only use for the saved values will be to reload them using the equivalent load multiple floating instruction to restore the context. 'FRd' specifies the first register to be transferred, and 'X' and 'Y' encode the number of registers transferred which can be from one to four. Note that these instructions use coprocessor number 2, whereas the other floating-point instructions use coprocessor number 1. 31 Figure 6.8 Load and store multiple floating binary en...
View Full Document

Ask a homework question - tutors are online