hardfloat

Hardfloat refers to the use of hardware floating-point support in a processor. It contrasts with softfloat, where floating-point arithmetic is implemented in software. When hardfloat is available and enabled, the processor’s floating-point unit (FPU) performs operations such as addition, subtraction, multiplication, division, and conversion between floating-point formats, typically resulting in higher performance for numeric workloads.

In practice, hardfloat is also associated with compiler toolchains and application binary interfaces (ABIs). An ABI described as hardfloat expects the target system to have an FPU and uses the FPU’s registers and instructions. Conversely, softfloat indicates that floating-point arithmetic is emulated in software, which improves portability to systems without an FPU but incurs a performance penalty.

Common discussions of hardfloat appear in ARM and MIPS toolchains. For ARM, the hard-float ABI uses an FPU and hardware calling conventions, while the soft-float ABI relies on software routines. For MIPS, a hard-float configuration enables code and libraries compiled to use the hardware FPU; soft-float relies on software emulation. Some toolchains offer a softfp mode to enable interoperability between libraries compiled with different floating-point ABIs.

Implications of choosing hardfloat include improved runtime performance for floating-point workloads and the need for all linked libraries (and the target hardware) to support the same FPU ABI. If the target system lacks an FPU, or if libraries are built for softfloat, using hardfloat can cause runtime errors or incompatibilities. Overall, hardfloat is about leveraging hardware capabilities for faster floating-point computation while balancing portability and compatibility concerns.