readme

linux 内核源代码

 +---------------------------------------------------------------------------+
 |  wm-FPU-emu   an FPU emulator for 80386 and 80486SX microprocessors.      |
 |                                                                           |
 | Copyright (C) 1992,1993,1994,1995,1996,1997,1999                          |
 |                       W. Metzenthen, 22 Parker St, Ormond, Vic 3163,      |
 |                       Australia.  E-mail [e-mail]              |
 |                                                                           |
 |    This program is free software; you can redistribute it and/or modify   |
 |    it under the terms of the GNU General Public License version 2 as      |
 |    published by the Free Software Foundation.                             |
 |                                                                           |
 |    This program is distributed in the hope that it will be useful,        |
 |    but WITHOUT ANY WARRANTY; without even the implied warranty of         |
 |    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the          |
 |    GNU General Public License for more details.                           |
 |                                                                           |
 |    You should have received a copy of the GNU General Public License      |
 |    along with this program; if not, write to the Free Software            |
 |    Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.              |
 |                                                                           |
 +---------------------------------------------------------------------------+



wm-FPU-emu is an FPU emulator for Linux. It is derived from wm-emu387
which was my 80387 emulator for early versions of djgpp (gcc under
msdos); wm-emu387 was in turn based upon emu387 which was written by
DJ Delorie for djgpp.  The interface to the Linux kernel is based upon
the original Linux math emulator by Linus Torvalds.

My target FPU for wm-FPU-emu is that described in the Intel486
Programmer's Reference Manual (1992 edition). Unfortunately, numerous
facets of the functioning of the FPU are not well covered in the
Reference Manual. The information in the manual has been supplemented
with measurements on real 80486's. Unfortunately, it is simply not
possible to be sure that all of the peculiarities of the 80486 have
been discovered, so there is always likely to be obscure differences
in the detailed behaviour of the emulator and a real 80486.

wm-FPU-emu does not implement all of the behaviour of the 80486 FPU,
but is very close.  See "Limitations" later in this file for a list of
some differences.

Please report bugs, etc to me at:
       [e-mail]
or     [e-mail]

For more information on the emulator and on floating point topics, see
my web pages, currently at  http://www.suburbia.net/~billm/


--Bill Metzenthen
  December 1999


----------------------- Internals of wm-FPU-emu -----------------------

Numeric algorithms:
(1) Add, subtract, and multiply. Nothing remarkable in these.
(2) Divide has been tuned to get reasonable performance. The algorithm
    is not the obvious one which most people seem to use, but is designed
    to take advantage of the characteristics of the 80386. I expect that
    it has been invented many times before I discovered it, but I have not
    seen it. It is based upon one of those ideas which one carries around
    for years without ever bothering to check it out.
(3) The sqrt function has been tuned to get good performance. It is based
    upon Newton's classic method. Performance was improved by capitalizing
    upon the properties of Newton's method, and the code is once again
    structured taking account of the 80386 characteristics.
(4) The trig, log, and exp functions are based in each case upon quasi-
    "optimal" polynomial approximations. My definition of "optimal" was
    based upon getting good accuracy with reasonable speed.
(5) The argument reducing code for the trig function effectively uses
    a value of pi which is accurate to more than 128 bits. As a consequence,
    the reduced argument is accurate to more than 64 bits for arguments up
    to a few pi, and accurate to more than 64 bits for most arguments,
    even for arguments approaching 2^63. This is far superior to an
    80486, which uses a value of pi which is accurate to 66 bits.

The code of the emulator is complicated slightly by the need to
account for a limited form of re-entrancy. Normally, the emulator will
emulate each FPU instruction to completion without interruption.
However, it may happen that when the emulator is accessing the user
memory space, swapping may be needed. In this case the emulator may be
temporarily suspended while disk i/o takes place. During this time
another process may use the emulator, thereby perhaps changing static
variables. The code which accesses user memory is confined to five
files:
    fpu_entry.c
    reg_ld_str.c
    load_store.c
    get_address.c
    errors.c
As from version 1.12 of the emulator, no static variables are used
(apart from those in the kernel's per-process tables). The emulator is
therefore now fully re-entrant, rather than having just the restricted
form of re-entrancy which is required by the Linux kernel.

----------------------- Limitations of wm-FPU-emu -----------------------

There are a number of differences between the current wm-FPU-emu
(version 2.01) and the 80486 FPU (apart from bugs).  The differences
are fewer than those which applied to the 1.xx series of the emulator.
Some of the more important differences are listed below:

The Roundup flag does not have much meaning for the transcendental
functions and its 80486 value with these functions is likely to differ
from its emulator value.

In a few rare cases the Underflow flag obtained with the emulator will
be different from that obtained with an 80486. This occurs when the
following conditions apply simultaneously:
(a) the operands have a higher precision than the current setting of the
    precision control (PC) flags.
(b) the underflow exception is masked.
(c) the magnitude of the exact result (before rounding) is less than 2^-16382.
(d) the magnitude of the final result (after rounding) is exactly 2^-16382.
(e) the magnitude of the exact result would be exactly 2^-16382 if the
    operands were rounded to the current precision before the arithmetic
    operation was performed.
If all of these apply, the emulator will set the Underflow flag but a real
80486 will not.

NOTE: Certain formats of Extended Real are UNSUPPORTED. They are
unsupported by the 80486. They are the Pseudo-NaNs, Pseudoinfinities,
and Unnormals. None of these will be generated by an 80486 or by the
emulator. Do not use them. The emulator treats them differently in
detail from the way an 80486 does.

Self modifying code can cause the emulator to fail. An example of such
code is:
          movl %esp,[%ebx]
	  fld1
The FPU instruction may be (usually will be) loaded into the pre-fetch
queue of the CPU before the mov instruction is executed. If the
destination of the 'movl' overlaps the FPU instruction then the bytes
in the prefetch queue and memory will be inconsistent when the FPU
instruction is executed. The emulator will be invoked but will not be
able to find the instruction which caused the device-not-present
exception. For this case, the emulator cannot emulate the behaviour of
an 80486DX.

Handling of the address size override prefix byte (0x67) has not been
extensively tested yet. A major problem exists because using it in
vm86 mode can cause a general protection fault. Address offsets
greater than 0xffff appear to be illegal in vm86 mode but are quite
acceptable (and work) in real mode. A small test program developed to
check the addressing, and which runs successfully in real mode,
crashes dosemu under Linux and also brings Windows down with a general
protection fault message when run under the MS-DOS prompt of Windows
3.1. (The program simply reads data from a valid address).

The emulator supports 16-bit protected mode, with one difference from
an 80486DX.  A 80486DX will allow some floating point instructions to
write a few bytes below the lowest address of the stack.  The emulator
will not allow this in 16-bit protected mode: no instructions are
allowed to write outside the bounds set by the protection.

----------------------- Performance of wm-FPU-emu -----------------------

Speed.
-----

The speed of floating point computation with the emulator will depend
upon instruction mix. Relative performance is best for the instructions
which require most computation. The simple instructions are adversely
affected by the FPU instruction trap overhead.


Timing: Some simple timing tests have been made on the emulator functions.
The times include load/store instructions. All times are in microseconds
measured on a 33MHz 386 with 64k cache. The Turbo C tests were under
ms-dos, the next two columns are for emulators running with the djgpp
ms-dos extender. The final column is for wm-FPU-emu in Linux 0.97,
using libm4.0 (hard).

function      Turbo C        djgpp 1.06        WM-emu387     wm-FPU-emu

   +          60.5           154.8              76.5          139.4
   -          61.1-65.5      157.3-160.8        76.2-79.5     142.9-144.7
   *          71.0           190.8              79.6          146.6
   /          61.2-75.0      261.4-266.9        75.3-91.6     142.2-158.1

 sin()        310.8          4692.0            319.0          398.5
 cos()        284.4          4855.2            308.0          388.7
 tan()        495.0          8807.1            394.9          504.7
 atan()       328.9          4866.4            601.1          419.5-491.9

 sqrt()       128.7          crashed           145.2          227.0
 log()        413.1-419.1    5103.4-5354.21    254.7-282.2    409.4-437.1
 exp()        479.1          6619.2            469.1          850.8


The performance under Linux is improved by the use of look-ahead code.
The following results show the improvement which is obtained under
Linux due to the look-ahead code. Also given are the times for the
original Linux emulator with the 4.1 'soft' lib.

 [ Linus' note: I changed look-ahead to be the default under linux, as
   there was no reason not to use it after I had edited it to be
   disabled during tracing ]

            wm-FPU-emu w     original w
            look-ahead       'soft' lib
   +         106.4             190.2
   -         108.6-111.6      192.4-216.2
   *         113.4             193.1
   /         108.8-124.4      700.1-706.2

 sin()       390.5            2642.0
 cos()       381.5            2767.4
 tan()       496.5            3153.3
 atan()      367.2-435.5     2439.4-3396.8