Non-Confidential | PDF version | ARM DUI0472M | ||
| ||||
Home > Compiler Command-line Options > --fpmode=model |
Specifies floating-point standard conformance. This controls which floating-point optimizations the compiler can perform, and also influences library selection.
--fpmode=
model
Where
is one of:model
ieee_full
All facilities, operations, and representations guaranteed by the IEEE standard are available in single and double-precision. Modes of operation can be selected dynamically at runtime.
This defines the symbols:
__FP_IEEE __FP_FENV_EXCEPTIONS __FP_FENV_ROUNDING __FP_INEXACT_EXCEPTION
ieee_fixed
IEEE standard with round-to-nearest and no inexact exceptions.
This defines the symbols:
__FP_IEEE __FP_FENV_EXCEPTIONS
ieee_no_fenv
IEEE standard with round-to-nearest and no exceptions. This mode is stateless and is compatible with the Java floating-point arithmetic model.
This defines the symbol __FP_IEEE
.
none
The compiler permits --fpmode=none
as an alternative to
--fpu=none
, indicating that source code is not permitted to use
floating-point types of any kind.
std
IEEE finite values with denormals flushed to zero, round-to-nearest, and no exceptions. This is compatible with standard C and C++ and is the default option.
Normal finite values are as predicted by the IEEE standard. However:
NaNs and infinities might not be produced in all circumstances defined by the IEEE model. When they are produced, they might not have the same sign.
The sign of zero might not be that predicted by the IEEE model.
Using NaNs in arithmetic operations with --fpmode=std
causes undefined behavior.
fast
Perform more aggressive floating-point optimizations that might cause a small loss
of accuracy to provide a significant performance increase. This option defines the
symbol __FP_FAST
.
This option results in behavior that is not fully compliant with the ISO C or C++ standard. However, numerically robust floating-point programs are expected to behave correctly.
A number of transformations might be performed, including:
Double-precision math functions might be converted to single precision equivalents if all floating-point arguments can be exactly represented as single precision values, and the result is immediately converted to a single-precision value.
This transformation is only performed when the selected library contains the single-precision equivalent functions, for example, when the selected library is armcc or aeabi_glibc.
For example:
float f(float a) { return sqrt(a); }
is transformed to
float f(float a) { return sqrtf(a); }
Double-precision floating-point expressions that are narrowed to
single-precision are evaluated in single-precision when it is beneficial to do so.
For example,
is evaluated as
float
y =
(float
)(x + 1.0)
.float
y =
(float
)x + 1.0f
Division by a floating-point constant is replaced by multiplication with the
inverse. For example, x / 3.0
is evaluated as x * (1.0 /
3.0)
.
It is not guaranteed that the value of errno
is compliant with
the ISO C or C++ standard after math functions have been called. This enables the
compiler to inline the VFP square root instructions in place of calls to
sqrt()
or sqrtf()
.
Using a NaN with --fpmode=fast
can produce undefined
behavior.
Initialization code might be required to enable the VFP.
The default is --fpmode=std
.