amiga-bootcamp/15_cpu_and_mmu/68040_68060_libraries.md

[← Home](../README.md) · [CPU & MMU](README.md)

# 68040.library and 68060.library — CPU Support Libraries

## Overview

The 68040 and 68060 processors **removed certain instructions** that the 68020/68030 supported in hardware. These "unimplemented" instructions cause a Line-F exception when executed. The `68040.library` and `68060.library` are **trap handler libraries** that catch these exceptions and **emulate the missing instructions in software**, providing transparent backward compatibility.

Without these libraries, any program using the affected instructions would crash with a Line-F exception on 040/060 hardware.

---

## What Instructions Are Emulated?

### 68040.library

The 68040 removed several FPU instructions that the 68881/68882 supported:

| Category | Missing Instructions | Description |
|---|---|---|
| Transcendental FPU | `FSIN`, `FCOS`, `FTAN`, `FASIN`, `FACOS`, `FATAN` | Trig functions |
| Transcendental FPU | `FSINH`, `FCOSH`, `FTANH`, `FATANH` | Hyperbolic functions |
| Transcendental FPU | `FLOG2`, `FLOG10`, `FLOGN`, `FLOGNP1` | Logarithms |
| Transcendental FPU | `FETOX`, `FETOXM1`, `FTWOTOX`, `FTENTOX` | Exponentials |
| Other FPU | `FMOD`, `FREM` | Modulo/remainder |
| Other FPU | `FGETEXP`, `FGETMAN` | Get exponent/mantissa |
| Other FPU | `FSGLDIV`, `FSGLMUL` | Single-precision ops |
| Integer | `MOVEP` | Move peripheral (not on all 040 revisions) |

### 68060.library

The 68060 removed **everything the 040 removed** plus additional instructions:

| Category | Additionally Missing | Description |
|---|---|---|
| Integer | `MOVEP` | Move peripheral (byte-strided) |
| Integer | `CAS2` | Compare-and-swap dual |
| Integer | `CHK2`, `CMP2` | Range check |
| Integer | `MULU.L (64-bit)` | 64-bit unsigned multiply |
| Integer | `MULS.L (64-bit)` | 64-bit signed multiply |
| Integer | `DIVU.L (64-bit)` | 64-bit unsigned divide |
| Integer | `DIVS.L (64-bit)` | 64-bit signed divide |
| FPU | All 68040-missing FPU ops | Same as above |
| FPU | `FMOVECR` | Move constant ROM |
| FPU | `FDABS`, `FDSQRT`, etc. | Some double-precision ops |

---

## How They Work

```
1. Program executes FSIN (opcode $F200 xxxx)
2. 68040/060 CPU has no microcode for this → Line-F exception (#11)
3. CPU vectors to the Line-F exception handler
4. 68040.library's handler decodes the opcode from the stack frame
5. Software emulates FSIN using basic FADD/FMUL/FDIV
6. Result is placed in the correct FPU register
7. Handler returns → program continues as if nothing happened
```

---

## Installation

These libraries are **loaded at boot time** as resident modules. They install themselves as the Line-F exception vector handler.

```
; In startup-sequence or user-startup:
LIBS:68040.library    ; for 68040 systems
; or:
LIBS:68060.library    ; for 68060 systems (replaces 68040.library)
```

The library is typically loaded by `SetPatch` or an explicit `C:LoadModule`:

```
C:LoadModule LIBS:68040.library
; or for 68060:
C:LoadModule LIBS:68060.library
```

---

## struct (ROM Tag)

Both libraries register as `RTF_COLDSTART` resident modules with high priority to ensure they are initialised before any user code runs:

```c
/* Typical RomTag for 68040.library: */
static struct Resident romtag = {
    RTC_MATCHWORD,         /* $4AFC */
    &romtag,
    &endskip,
    RTF_COLDSTART,         /* flags: cold start */
    40,                    /* version */
    NT_LIBRARY,
    105,                   /* priority: very high */
    "68040.library",
    "68040.library 40.1 (1.1.93)\r\n",
    initRoutine
};
```

---

## Detection

```c
/* Check which CPU is present: */
if (SysBase->AttnFlags & AFF_68040)  /* 68040 */
if (SysBase->AttnFlags & AFF_68060)  /* 68060 */

/* AttnFlags bits: */
#define AFB_68010   0
#define AFB_68020   1
#define AFB_68030   2
#define AFB_68040   3
#define AFB_68060   7   /* added by 68060.library */
#define AFB_68881   4   /* FPU present */
#define AFB_68882   5
#define AFB_FPU40   6   /* 040 internal FPU */
```

---

## Performance Impact

Software emulation of transcendental FPU instructions is **10–100x slower** than the 68881/68882 hardware implementation. Performance-critical code should:
- Use lookup tables for trig functions
- Use polynomial approximations (Chebyshev, CORDIC)
- Avoid `FSIN`/`FCOS` in tight loops

---

## Common Sources

| Library | Source |
|---|---|
| 68040.library 37.4 | Commodore (OS 3.0 distribution) |
| 68040.library 40.1 | Commodore (OS 3.1 distribution) |
| 68060.library 40.1 | Phase5 (original 68060 accelerators) |
| 68060.library 46.x | Motorola reference implementation |

---

## References

- Motorola: *MC68040 User's Manual* — unimplemented instruction list
- Motorola: *MC68060 User's Manual* — unimplemented instruction list  
- NDK39: `exec/execbase.h` — `AttnFlags`
- Phase5: 68060.library source (public domain)