FPU Emulation Revival for i486SX on NetBSD

This retro-computing project restores support for x87 floating-point unit (FPU) emulation in the NetBSD kernel, targeting legacy 486SX-class processors without hardware FPUs. It brings back the original MATH_EMULATE option into NetBSD 10.x and beyond, as well as reverts and reworks the changes introduced in commit dfe83e0, which removed FPU emulation support from the kernel.

Disclaimer

This project is a work in progress and may contain bugs or incomplete functionality. Use at your own risk. The author is not responsible for any issues caused by its use.

Notes

• fyl2x seems to work properly on FPU, but when executed from libc, log functions bring incorrect results
• Some operations have hard-to-detect issues with precision

x87 FPU Emulated Instructions

🧠 Control & Initialization

Instruction	Status	Description	Opcode	Example
fninit	✅ OK	Initialize FPU	9B DB E3	fninit

📤 Load to FPU Stack

Instruction	Status	Description	Opcode	Example
fld	✅ OK	Load floating-point value	D9 /0	fld st(1)
fldt	✅ OK	Load 80-bit extended precision	DB /5	fldt [mem]
filds	✅ OK	Load integer (short)	DB /0	filds [mem]
fildl	✅ OK	Load long integer	DB /A	fildl [mem]

📤 Store from FPU Stack

Instruction	Status	Description	Opcode	Example
fstps	✅ OK	Store and pop single precision	D9 /3	fstps [mem]
fstpt	✅ OK	Store 80-bit extended precision & pop	DB /7	fstpt [mem]

➕➖✖️➗ Arithmetic Operations

Instruction	Status	Description	Opcode	Example
fadd	✅ OK	Add floating-point values	D8 /0	fadd st(1), st
faddl	✅ OK	Add long double from memory	DA /0	faddl [mem]
fsub	✅ OK	Subtract floating-point values	D8 /4	fsub st(1), st
fsubp	✅ OK	Subtract with pop	DE /5	fsubp st(1), st(0)
fmul	✅ OK	Multiply floating-point values	D8 /1	fmul st(1), st
fdiv	✅ OK	Divide floating-point values	D8 /6	fdiv st(1), st
fdivp	✅ OK	Divide with pop	DE /7	fdivp st(1), st(0)
fscale	✅ OK	Scale ST(0) by ST(1)	D9 FD	fscale

🔍 Comparison

Instruction	Status	Description	Opcode	Example
fcom	✅ OK	Compare floating-point values	D8 /2	fcom st(1)
fucom	✅ OK	Unordered compare ST(0) with ST(i)	DD E0+i	fucom st(1)
fucomp	✅ OK	Unordered compare and pop	DD E8+i	fucomp st(1)
ftst	✅ OK	Compare ST(0) with 0.0	D9 E4	ftst

🔁 Stack Manipulation & Modification

Instruction	Status	Description	Opcode	Example
fxch	✅ OK	Exchange ST(0) with ST(i)	D9 C8+i	fxch st(1)
fchs	✅ OK	Change sign of ST(0)	D9 E0	fchs
fabs	✅ OK	Absolute value of ST(0)	D9 E1	fabs
frndint	✅ OK	Round ST(0) to integer	D9 FC	frndint

📈 Logarithmic and Special Math Operations

Instruction	Status	Description	Opcode	Example
fyl2x	⏳ TBD	Compute y × log₂(x) (ST(1) × log₂(ST(0))) and pop	D9 F1	fyl2x
fyl2xp1	❌ N/A	Compute y × log₂(x+1) and pop	D9 F9	fyl2xp1
fxtract	❌ N/A	Extract: ST(0) → ST = significand, ST+1 = exponent	D9 F4	fxtract
f2xm1	✅ OK	Compute 2^x - 1 for ST(0)	D9 F0	f2xm1
fpatan	❌ N/A	Compute arctangent(ST(1)/ST(0)) and pop	D9 F3	fpatan
fsqrt	❌ N/A	Compute square root of ST(0)	D9 FA	fsqrt

Installation

• Read Chapter 32. Obtaining the sources
• Read Chapter 34. Compiling the kernel
• Add the repository contents under /src/sys/arch on NetBSD 10.x machine, then run:

$ cd /usr/src/sys/arch/i386/conf/
$ vi GENERIC_TINY_486SX# or GENERIC_PS2TINY_486SX or create your own kernel configuration with "options MATH_EMULATE"
$ config GENERIC_TINY_486SX
$ cd ../compile/GENERIC_TINY_486SX
$ make depend
$ make

# If everythig good, install the new kernel under root:
# mv /netbsd /netbsd.old
# mv netbsd /

Prebuilt drive image

Check Releases for the compiled netbsd kernel with FPU-emulation as well, as the drive image

Using 86Box to Emulate i486SX

To try FPU emulation on NetBSD it's convenient to use 86Box — a high-accuracy emulator of IBM PC-compatible systems.

• Install 86Box
• If required, download ROM set from 86Box ROM repository, then extract to the 86Box roms directory
• Download NetBSD-10.1 prebuilt drive image from Releases
• Edit sample 86Box configuration to specify correct paths and host parameters. Don't forget to select i486SX and check that no FPU present.

Contacts

If you have an idea, a question, or have found a problem, feel free to open an issue or contact me directly: Mikhail Zakharov.

My changes to the original code by Linus Torvalds, William "Bill" Jolitz, and NetBSD are licensed under the BSD-2-Clause license.