Table of Contents
Advertisement
FLOATING-POINT INSTRUCTIONS
When dual operation instructions are used in single-precision mode, all 64 bits of the T,
KR, and KI registers are updated, but the values stored there are not converted to
double-precision format (the exponent bias is not adjusted for double precision).
Instead, zeros are inserted as pads in exponent bits 11:9 and as the fraction's least
significant 29 bits (bits 28:0). All 64 bits of the T, KR, and KI registers can be initialized
to zero using 3 single-precision r2apt.ss ID,ID,ID instructions and 1 i2apt.ss ID,ID,ID.
Because single-precision values are stored in these 64-bit registers in a format which
does not conform to the standard for double-precision numbers, leaving a valid single-
precision value in T, KR, or KI can cause floating-point traps if a double-precision
operation is later performed referencing one of these registers. Likewise, valid double-
precision values left in T, KR, or KI can cause traps if a single precision operation is
later performed using one of these registers. Therefore, programs should clear T, KR,
and KI before switching precisions.
fsrc1
fsrc2 fdest
MULTIPLIER UNIT
RESULT
ADDER UNIT
RESULT
240329i
Figure 6-2. Dual-Operation Data Paths
6-16
Advertisement
Table of Contents
