This option enables you to specify the target FPU architecture.
If you specify this option, it overrides any implicit FPU
option that appears on the command line, for example, where you
use the --cpu option.
To obtain a full list of FPU architectures use the --fpu=list option.
Syntax
--fpu=name
Where name
noneSelects no floating-point option. No floating-point code is to be used. This produces an error if your code contains float types.
vfpvThis is a synonym for
vfpv2.vfpv2Selects a hardware vector floating-point unit conforming to architecture VFPv2.
Note
If you enter
armcc --thumb --fpu=vfpv2on the command line, the compiler compiles as much of the code using the Thumb instruction set as possible, but hard floating-point sensitive functions are compiled to ARM code. In this case, the value of the predefine__thumbis not correct.vfpv3Selects a hardware vector floating-point unit conforming to architecture VFPv3. VFPv3 is backwards compatible with VFPv2 except that VFPv3 cannot trap floating-point exceptions.
vfpv3_fp16Selects a hardware vector floating-point unit conforming to architecture VFPv3 that also provides the half-precision extensions.
vfpv3_d16Selects a hardware vector floating-point unit conforming to VFPv3-D16 architecture.
vfpv3_d16_fp16Selects a hardware vector floating-point unit conforming to VFPv3-D16 architecture, that also provides the half-precision extensions.
vfpv4Selects a hardware floating-point unit conforming to the VFPv4 architecture.
vfpv4_d16Selects a hardware floating-point unit conforming to the VFPv4-D16 architecture.
fpv4-spSelects a hardware floating-point unit conforming to the single precision variant of the FPv4 architecture.
softvfpSelects software floating-point support where floating-point operations are performed by a floating-point library,
fplib. This is the default if you do not specify a--fpuoption, or if you select a CPU that does not have an FPU.softvfp+vfpv2Selects a hardware vector floating-point unit conforming to VFPv2, with software floating-point linkage. Select this option if you are interworking Thumb code with ARM code on a system that implements a VFP unit.
If you select this option:
Compiling with
--thumbbehaves in a similar way to--fpu=softvfpexcept that it links with floating-point libraries that use VFP instructions.Compiling with
--armbehaves in a similar way to--fpu=vfpv2except that all functions are given software floating-point linkage. This means that functions pass and return floating-point arguments and results in the same way as--fpu=softvfp, but use VFP instructions internally.
Note
If you specify
softvfp+vfpv2with the--armor--thumboption for C code, it ensures that your interworking floating-point code is compiled to use software floating-point linkage.softvfp+vfpv3Selects a hardware vector floating-point unit conforming to VFPv3, with software floating-point linkage. Select this option if you are interworking Thumb code with ARM code on a system that implements a VFPv3 unit.
softvfp+vfpv3_fp16Selects a hardware vector floating-point unit conforming to VFPv3-fp16, with software floating-point linkage.
softvfp+vfpv3_d16Selects a hardware vector floating-point unit conforming to VFPv3-D16, with software floating-point linkage.
softvfp+vfpv3_d16_fp16Selects a hardware vector floating-point unit conforming to VFPv3-D16-fp16, with software floating-point linkage.
softvfp+vfpv4Selects a hardware floating-point unit conforming to FPv4, with software floating-point linkage.
softvfp+vfpv4_d16Selects a hardware floating-point unit conforming to VFPv4-D16, with software floating-point linkage.
softvfp+fpv4-spSelects a hardware floating-point unit conforming to FPv4-SP, with software floating-point linkage.
Any FPU explicitly selected using the --fpu option
always overrides any FPU implicitly selected using the --cpu option.
For example, the option --cpu=ARM1136JF-S --fpu=softvfp generates
code that uses the software floating-point library fplib,
even though the choice of CPU implies the use of architecture VFPv2.
To control floating-point linkage without affecting the choice
of FPU, you can use --apcs=/softfp or --apcs=/hardfp.
The compiler only permits hardware VFP architectures (for
example, --fpu=vfpv3, --fpu=softvfp+vfpv2),
to be specified when MRRC and MCRR instructions
are supported in the processor instruction set. MRRC and MCRR instructions
are not supported in 4, 4T, 5T and 6-M. Therefore, the compiler
does not allow the use of these CPU architectures with hardware
VFP architectures.
Other than this, the compiler does not check that --cpu and --fpu combinations
are valid. Beyond the scope of the compiler, additional architectural
constraints apply. For example, VFPv3 is not supported with architectures
prior to ARMv7. Therefore, the combination of --fpu and --cpu options
permitted by the compiler does not necessarily translate to the
actual device in use.
The compiler only generates scalar floating-point operations. If you want to use VFP vector operations, you must do this using assembly code.
NEON support is disabled for softvfp.
The default target FPU architecture is derived from use of
the --cpu option.
If the CPU specified with --cpu has a VFP
coprocessor, the default target FPU architecture is the VFP architecture
for that CPU. For example, the option --cpu ARM1136JF-S implies
the option --fpu vfpv2. If a VFP coprocessor
is present, VFP instructions are generated.
If you are building ARM Linux applications with --arm_linux or --arm_linux_paths,
the default is always software floating-point linkage. Even if you
specify a CPU that implies an FPU (for example, --cpu=ARM1136JF-S),
the compiler still defaults to --fpu=softvfp+vfp,
not --fpu=vfp.
If there is no VFP coprocessor, the compiler generates code
that makes calls to the software floating-point library fplib to
carry out floating-point operations.
Vector Floating-Point (VFP) architectures in Using the Compiler
Compiler support for floating-point computations and linkage in Using the Compiler
ARM and Thumb floating-point build options (ARMv6 and earlier) in Developing Software for ARM Processors
ARM and Thumb floating-point build options (ARMv7 and later) in Developing Software for ARM Processors
MRC, MRC2, MRRC and MRRC2 in the Assembler Reference
MCR, MCR2, MCRR, and MCRR2 in the Assembler Reference.
PDF version