RX610 Group
2. CPU
R01UH0032EJ0120 Rev.1.20
Page 63 of 1006
Feb 20, 2013
Explanation of Floating-Point Rounding Modes
•
Rounding to the nearest value (the default behavior): An inexact result is rounded to the available value that is
closest to the result which would be obtained with an infinite
number of digits. If two available values are equally close,
rounding is to the even alternative.
•
Rounding towards 0:
An inexact result is rounded to the smallest available absolute
value, i.e. in the direction of zero (simple truncation).
•
Rounding t
∞
:
An inexact result is rounded to the nearest available value in
the direction of positive infinity.
•
Rounding towards –
∞
:
An inexact result is rounded to the nearest available value in
the direction of negative infinity.
(1) Rounding to the nearest value is specified as the default mode and returns the most accurate value.
(2) Modes such as rounding towards 0, rounding t
∞
, and rounding towards –
∞
are used to ensure precision when
interval arithmetic is employed.
CV Flag (Invalid Operation Cause Flag), CO Flag (Overflow Cause Flag),
CZ Flag (Division-by-Zero Cause Flag), CU Flag (Underflow Cause Flag),
CX Flag (Inexact Cause Flag), and CE Flag (Un-Implemented Processing Cause Flag)
Floating-point exceptions include the five specified in the IEEE754 standard, namely overflow, underflow, inexact,
division-by-zero, and invalid operation. For a further floating-point exception that is generated upon detection of
unimplemented processing, the corresponding flag (CE) is set to 1.
• The bit that has been set to 1 is cleared to 0 when the FPU instruction is executed.
• When 0 is written to the bit by the MVTC and POPC instructions, the bit is set to 0; the bit retains the previous value
when 1 is written by the instruction.
DN Flag (0 Flush Bit of Denormalized Number)
When this bit is set to 0, a denormalized number is handled as a denormalized number. When this bit is set to 1, a
denormalized number is handled as 0.
EV Bit (Invalid Operation Exception Enable), EO Bit (Overflow Exception Enable),
EZ Bit (Division-by-Zero Exception Enable), EU Bit (Underflow Exception Enable), and
EX Bit (Inexact Exception Enable)
When any of five floating-point exceptions specified in the IEEE754 standard is generated by the FPU instruction, the bit
decides whether the CPU will start handling the exception. When the bit is set to 0, the exception handling is masked;
when the bit is set to 1, the exception handling is enabled.
Summary of Contents for RX600 Series
Page 1006: ...RX610 Group R01UH0032EJ0120 ...