Double-precision FPU probe – codegen witness for the RA8P1 DP-FPU. More...

This graph shows which files directly or indirectly include this file:

Functions
double	ra8_fpu_dp_madd (double a, double b, double c)
	Evaluate the double-precision product-sum a * b + c.

Detailed Description

Double-precision FPU probe – codegen witness for the RA8P1 DP-FPU.

Tag: [Ring 3 / HAL] {World: S}

The two supported RA8 parts differ in FPU width even though both are a Cortex-M85:

RA8D2: single-precision FPv5 (__FPU_DP == 0). The RA8D2 toolchain file compiles with -mfpu=fpv5-sp-d16, so double arithmetic cannot run on the FPU and the compiler lowers it to soft-float library calls (__aeabi_dmul, __aeabi_dadd, ...).
RA8P1: double-precision FPv5 (the M85 DP-FPU). The RA8P1 toolchain file (cmake/toolchain-ra8p1.cmake) overrides to -mfpu=fpv5-d16, so the same double arithmetic compiles to hardware DP-FPU opcodes (vmul.f64, vadd.f64 / vfma.f64).

ra8_fpu_dp_madd() is a deliberately tiny double computation that both device builds compile. Disassembling its object (arm-none-eabi-objdump -d) is the witness: the RA8P1 build shows .f64 VFP opcodes; the RA8D2 build shows soft-float bl __aeabi_d* calls. It doubles as a runtime DP-FPU sanity check whose numeric result is validated by the host unit test.

Note: Host-friendly: pure numeric leaf function, touches no hardware, so it runs unchanged under RA8_OFF_TARGET and in the host unit tests (which compute it on the host's native binary64 hardware).

See also: cmake/toolchain-ra8p1.cmake Overrides -mfpu to the DP-FPU for RA8P1.; ra8_device.h RA8D2/RA8P1 compile-time device switch.

Copyright: Copyright (c) 2026 Brighton Sikarskie SPDX-License-Identifier: MIT

Since: 0.1.0

Definition in file ra8_fpu_probe.h.

Function Documentation

◆ ra8_fpu_dp_madd()

double ra8_fpu_dp_madd	(	double	a,
		double	b,
		double	c )

Evaluate the double-precision product-sum a * b + c.

A minimal IEEE-754 binary64 (double) computation used as a compile-target witness for the FPU width (see the file header). On a DP-FPU build the body lowers to hardware .f64 opcodes; on an SP-FPU build it lowers to soft-float calls. The math itself – multiply then add, or a fused multiply-add – is identical in both cases and is exercised for numeric correctness by the host unit test.

Parameters

[in]	a	First multiplicand, any finite double.
[in]	b	Second multiplicand, any finite double.
[in]	c	Addend, any finite double.

Returns: The product-sum evaluated in double precision.

Return values

a*b+c The IEEE-754 binary64 result (subject to a single rounding, or fewer if the compiler contracts to a fused multiply-add).

Precondition: a, b, c are ordinary double values (no NaN/Inf constraint – they propagate per IEEE-754).; The build selected an FPU via -mfpu (hard-float ABI in this project).

Postcondition: No hardware or global state is modified.; For finite inputs whose exact product-sum is representable, the result equals a * b + c.

Note: Thread-safe: pure function, no shared or static state.

See also: cmake/toolchain-ra8p1.cmake

Since: 0.1.0

Definition at line 19 of file ra8_fpu_probe.c.

Functions

Detailed Description

Function Documentation

◆ ra8_fpu_dp_madd()