RCPDP Double-Precision Floating-Point Reciprocal Approximation

Note:

1)If src2 is SNaN, NaN_out is placed in dst and the INVAL and NAN2 bits are set.

2)If src2 is QNaN, NaN_out is placed in dst and the NAN2 bit is set.

3)If src2 is a signed denormalized number, signed infinity is placed in dst and the DIV0, INFO, OVER, INEX, and DEN2 bits are set.

4)If src2 is signed 0, signed infinity is placed in dst and the DIV0 and INFO bits are set.

5)If src2 is signed infinity, signed 0 is placed in dst.

6)If the result underflows, signed 0 is placed in dst and the INEX and UNDER bits are set. Underflow occurs when 21022 t src2 t infinity.

Pipeline

 

 

 

Pipeline

E1

E2

 

 

Stage

 

 

 

 

 

Read

src2_l

 

 

 

src2_h

 

 

Written

dst_l

dst_h

 

Unit in use

.S

 

 

 

 

 

If dst is used as the source for the ADDDP, CMPEQDP, CMPLTDP, CMPGTDP, MPYDP, or SUBDP instruction, the number of delay slots can be reduced by one, because these instructions read the lower word of the DP source one cycle before the upper word of the DP source.

Instruction Type

2-cycle DP

 

 

 

 

 

Delay Slots

1

 

 

 

 

 

 

 

Functional Unit

1

 

 

 

 

 

 

 

Latency

 

 

 

 

 

 

 

 

 

See Also

 

RCPSP, RSQRDP

 

 

 

 

 

Example

 

RCPDP .S1 A1:A0,A3:A2

 

 

 

 

 

Before instruction

 

2 cycles after instruction

 

A1:A0

 

 

 

 

 

 

 

 

 

4010 0000h

 

0000 0000h

 

A1:A0

4010 0000h

0000

0000h

4.00

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

A3:A2

xxxx xxxxh

 

xxxx xxxxh

 

A3:A2

3FD0 0000h

0000

0000h

0.25

 

 

 

 

 

 

 

 

 

 

3-198

Instruction Set

SPRU733

Page 258
Image 258
Texas Instruments TMS320C67X/C67X+ DSP manual RCPSP, Rsqrdp, Unit in use