FPU Emulation Revival for i486SX on NetBSD

This retro-computing project aims to restore support for x87 floating-point unit (FPU) emulation in the NetBSD kernel, specifically targeting legacy 486SX-class processors that lack hardware FPUs. By integrating the original MATH_EMULATE option into NetBSD 10.x and beyond, this initiative reverts and enhances the alterations made in commit dfe83e0, which previously removed FPU emulation support from the kernel.

Project Status

This project is in active development and may exhibit bugs or incomplete features. Users are encouraged to proceed with caution, as the developer is not liable for any issues resulting from its usage.

Known Issues

• The fyl2x instruction appears to function properly; however, invoking log functions from libc yields incorrect results.
• Certain operations may have subtle precision discrepancies.

Supported x87 FPU Instructions

Control & Initialization

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

Load Instructions

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 short integer	DB /0	filds [mem]
fildl	✅ OK	Load long integer	DB /A	fildl [mem]

Store Instructions

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

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) and pop	D9 F1	fyl2x
fyl2xp1	❌ N/A	Compute y × log₂(x+1) and pop	D9 F9	fyl2xp1
fxtract	❌ N/A	Extract significand and exponent from ST(0)	D9 F4	fxtract
f2xm1	✅ OK	Compute 2^x - 1 for ST(0)	D9 F0	f2xm1
fpatan	❌ N/A	Compute arctangent and pop	D9 F3	fpatan
fsqrt	❌ N/A	Compute square root of ST(0)	D9 FA	fsqrt

Visit the Releases section to find precompiled kernel images with FPU emulation capability, as well as related drive images.