Exact timing

[Hans Ruegg]

I was always looking for a way to control the timing of a program more
exactly than it is possible with TIMER (e.g. for animations, music, etc).
MTIMER did not provide any solution because you can call this function only
once and must then reset it (which includes, as I found, waiting for the
next "regular" timer tick). Finally I decided to write my own Interrupt 8
handler in order to control the frequency of the timer tick (see code below).
It works quite fine so far, but I still have some questions concerning
running these routines under Windows 95/98:
- In my experience it was not possible to set the timer to a higher frequency
than about 1000 Hz. When setting it to any value between 1000 and 2000 Hz,
the real frequency remained at 1000 Hz. Above 2000 Hz, a message box appears:
"This program probably does not run well under Windows." (CPU speed cannot
be the reason; the speed of my system is 266 MHz),
- I tried the same code WITHOUT calling the original INT 8 handler. It would
be expected that this would freeze the system time at the point when the new
interrupt handler was installed. Under DOS, it actually does, but under
Windows it did not affect the system time at all (although it affected the
value of TIMER in PowerBASIC).
- In conclusion: Could anybody explain me what exactly Windows is doing with
the timer, and if this could result in unexpected problems using the code
posted below? ...and maybe somebody has a hint how to increase the clock
frequency to more than 1000 Hz (under Windows)?

Hans Ruegg.


Code for Timer routines:
========================
DEFINT A-Z

DECLARE SUB setpithandler (count??, pithandler??())
DECLARE FUNCTION getpit() AS DWORD
DECLARE FUNCTION pitcheck() AS WORD
DECLARE SUB pitfreq (freq)
DECLARE SUB resetpithandler ()

DIM tmr??(28) 'array which holds code for the new INT 8 handler
CLS
setpithandler (&H1234DD\1000), tmr??() 'set clock frequency to 1000Hz
PRINT "New handler installed."
DO:
LOCATE 3,1: PRINT USING "##########"; getpit;
LOCATE 3,40: PRINT USING "#####.##"; TIMER;
LOOP WHILE INKEY$=""
resetpithandler
LOCATE 24,1: PRINT "Handler successfully restored."
END

'========================================================================
'Following routines are only auxiliary routines
'for setting up the INT 8 handler

SUB loadASM (matrix??(), hexcode$)
'used to convert a string containing hexadecimal code into real code
length?? = LEN(hexcode$)
posic?? = VARPTR(matrix??(0))
DEF SEG = VARSEG(matrix??(0))
FOR i?? = 1 TO length??-1 STEP 2
d? = VAL("&H" + MID$(hexcode$, i??, 2))
POKE posic??, d?
INCR posic??
NEXT i??
DEF SEG
END SUB

SUB putintvector (intno%, start??)
'Lets an interrupt vector point to an array element specified by "start??"
DEF SEG = 0
POKEI 4*intno%, VARPTR(start??)
POKEI 4*intno%+2, VARSEG(start??)
DEF SEG
END SUB

SUB putintaddress (intno%, segm??, offs??)
'Establishes a segment and offset address as an interrupt vector
DEF SEG = 0
POKEI 4*intno%, offs??
POKEI 4*intno%+2, segm??
DEF SEG
END SUB

SUB getintvector (intno%, segm??, offs??)
'Returns segment and offset of an interrupt vector
DEF SEG = 0
segm?? = peeki(4*intno%+2)
offs?? = peeki(4*intno%)
DEF SEG
END SUB

'=========================================================================
'Now following the proper timer routines:

SUB setpithandler (count??, pithandler??()) PUBLIC
'Substitutes INT 8 handler
'pithandler??(28) holds code
'count?? = new counter value (0 to 65535)
'Output: pithandler??(0) = LSB, pithandler??(1) = MSB of new tick count
' pithandler??(6) = segment, pithandler??(7) = offset of BIOS handler
' pithandler??(2) = count??
cod$ = "00000000000000000000000000000000"
cod$ = cod$ + "50FA2E83060000012E83160200002EA104002E01060A007302"
cod$ = cod$ + "CDF6FBCDF5B020E62058CFCF9050495448"
loadASM pithandler??(), cod$ '(Source code see below)
getintvector 8, sg??, offs??
pithandler??(6) = sg??: pithandler??(7) = offs?? 'load variables
pithandler??(2) = count??
! CLI
putintaddress &HF6, sg??, offs?? 'Relocate original INT 8 handler
'to INT F6
putintvector &HF5, pithandler??(26) 'points to IRET instruction
'(INT F5 will be available to hook
'a routine which needs to be called
'at each tick)
putintvector 8, pithandler??(8) 'Let INT 8 vector point to the entry
'point of the new handler
OUT &H43, &H34 'Set clock frequency
OUT &H40, count?? MOD 256
OUT &H40, count??\256
! STI
END SUB

FUNCTION getpit() PUBLIC AS DWORD
'returns tick count since "setpithandler" was initialized.
DEF SEG = 0
tickptr?? = PEEKI(32)
tickseg?? = PEEKI(34)
DEF SEG = tickseg??
getpit = PEEKL (tickptr??-16)
DEF SEG
END FUNCTION

FUNCTION pitcheck() PUBLIC AS WORD
'Installation check for pithandler
'Output: 1 = pithandler not installed,
' other values: = counter constant in pithandler
DEF SEG = 0
s?? = PEEKI(34): o?? = PEEKI(32)
DEF SEG = s??
Id$ = PEEK$ (o??+38, 4)
IF Id$="PITH" THEN pitcheck = PEEKI (o??-12) ELSE pitcheck = 1
DEF SEG
END FUNCTION

SUB pitfreq (freq) PUBLIC
'Changes timer frequency; freq = frequency in Hz
IF freq<19 OR freq>2000 THEN EXIT SUB 'because Windows does not accept
'higher frequencies
counter?? = &H1234DD\freq
DEF SEG = 0
s?? = PEEKI(34): o?? = PEEKI(32)
DEF SEG = s??
POKEI o??-12, counter?? 'Set counter constant in interrupt handler
DEF SEG
! CLI
OUT &H43, &H34 'Set new clock frequency
OUT &H40, counter?? MOD 256
OUT &H40, counter??\256
! STI
END SUB

SUB resetpithandler () PUBLIC
'retrieves original handler, resets original frequency.
! CLI
getintvector &HF6, sg??, offs??
putintaddress 8, sg??, offs??
OUT &H43, &H34 'Restore original clock frequency
OUT &H40, 0
OUT &H40, 0
! STI
END SUB

==============================
Source code for INT 8 handler:
==============================
Byte Machine code ASM
00 00 00 ;var: Tick count (low)
02 00 00 ;var: Tick count (high)
04 00 00 ;var: counter constant
06 00 00 ;var: real clockticks (low)
08 00 00 ;var: real clockticks (high)
0A 00 00 ;var: sum for real tick count
0C 00 00 ;var: BIOS Handler Segment
0E 00 00 ;var: BIOS Handler Offset

;Byte 10h = Routine Entry point
10 50 PUSH AX
11 FA CLI
12 2E CS: ; update Tick count
13 83 06 00 00 01 ADD WORD [0000], 1
18 2E CS:
19 83 16 02 00 00 ADC WORD [0002], 0
1E 2E CS: ; update real tick counter
1F A1 04 00 MOV AX, [0004]
22 2E CS:
23 01 06 0A 00 ADD WORD [000A], AX
27 73 02 JNC 002B ; overflow?
29 CD F6 INT F6 ; yes, then call original INT 8 ; handler (relocated at INT F6)

2B FB STI
2C CD F5 INT F5 ; INT F5 can be used to hook further
; routines
2E B0 20 MOV AL, 20 ; acknowledge interrupt
30 E6 20 OUT 20, AL ; (may not be necessary ?)
32 58 POP AX
33 CF IRET
34 CF IRET ; repeated at an EVEN byte position
; so the INT F5 vector can point to
; this instruction
35 90 NOP

36 50 49 54 48 ;Signature "PITH" for installation check

(58 bytes)

[Bobby Gage]

I bild programs for interfaces that I bild. I need very good timing between the external interface and my programming. But it's hard in windows, and as you know everything is windows now!

So I finly found a good timeing mothed that will give a very good 1000th / sec accuracy. But it would not work all that good in windows. Then I found WinLib.zip it's fantastic. It allows you to completly shutdown windows so it give your dos program full controll.

I now have this accuracy in windows every time. I just lock windows up when I need to use the timer and turn it back on when I don't need to use the timer any more. EMail me if you need help with the timers and WinLib support.

I am very very bizzy but I can try to help you out as much as I can.



------------------

[Hans Ruegg]

Thank you, Bobby, for your hint and for WINLIB. I found that the
function for giving the DOS box "total control" is a simple interrupt
call which declares certain program sections as "time critical" to
Windows: Interrupt &H2F, AX = &H1681 for Start Critical Section and AX
= &H1682 for End Critical Section. According to Ralf Brown's Interrupt
List, this prevents Windows from task switching during the critical
section (which should not be too long in time). - The problem is that
even using this call, the DOS program does not get "absolute" control,
for example regarding my question, Windows still does NOT allow me
timer frequencies above 1000Hz.

I found other suggestions as, for example, using a combination of the
TIMER value and INP(&H40) (reading the internal timer count directly,
which should decrease from 65535 to 0 during each timer tick). For
using this method, you must first determine if you got the high byte or
the low byte first; then you can use every high byte to get a timer
resolution of 18.2*256 = 4659 Hz, or use both bytes to get an even
higher resolution. But there are also several problems I found:

- It seems that the first byte you get when you first read INP(&H40) is
neither the high nor the low byte, but garbage (at least on my system).
- I found that the counter decreased TWICE from 65536 to 0 during each
timer tick, contrary to every documentation I read. (Maybe somebody can
explain this to me?)
- If you use this method, you do not need to reprogram the timer, but
you can not connect an interrupt to the count. Therefore the program
must constantly check for the timer value, which uses much CPU time and
may slow down the rest of the program. On slow systems, it may be
impossible to check this value as fast as required while performing
other tasks.

Hans Ruegg

[Bobby Gage]

I have not found any code that will do better then 1000Hz.

The only thing I can recomend is some sort of hardware that will work for you. Make some sort of device that pluges into your ISA or PCI slots that will give you the timer you need.

Good luck man,
Bobby

------------------
Bobby Gage II

[Paul Dixon]

Hans,
you're always going to have trouble in windows with timing. Since you are using PB/DOS, why don't you run in DOS? You will then have NO problems with timing. I've had accurate timings to 10us without difficulty in DOS using similar methods to you.


<<the first byte you get when you first read INP(&H40) is neither the high nor the low byte, but garbage>>

The first byte is the low byte of the timer which increments very quickly so if you want to make sense of it you must read it very quickly. Also, in mode 3 it increments twice on each clock, see below.


<<I found that the counter decreased TWICE>>

Your PIT timer is set to the wrong mode. In mode 3 it counts as you describe to give a square wave output on the interrupt line. Change it to mode 2 and it will count as you want it to (65535->0) without any problem and give a pulse output on the interrupt line. Since the PC uses an edge triggerred interrupts then pulse or square wave makes no difference to the interrupts.

Paul.


------------------

[Hans Ruegg]

Thank you, Paul, for your hints - you were right about the timer mode.
So this would be another possibility for a timer with a higher
resolution (1/256 of a normal timer tick):

OUT &H43, &B00110100 'setting timer to mode 2
'(use this before reading the timer)
ticks& = peektimer

FUNCTION peektimer AS LONG
DEF SEG = 0
low? = INP(&H40) 'discard low byte
a& = PEEKL (&H46C)
SHIFT LEFT a&, 8
FUNCTION = a&+256-INP(&H40)
DEF SEG
END FUNCTION

As for running in DOS mode, I personally had much trouble when trying to configure my
system in order to support mouse, soundcard, etc. in DOS mode, and I wonder if other
users have the same problems. (I am aware this is outside the scope of this subject,
but I did not yet find a DOS compatible mouse driver on the Windows CD, and even EMM386
returns an error message when I try to put it into the CONFIG.SYS file ... there are still many
things I do not understand yet.)
As I am not a professional programmer, I do not have the time either to search for and read
the technical information necessary for making good Windows programs. So I am staying
with DOS which seems to me much easier to understand, but I do not know if Windows users
will like any programs which suggest them to be run in DOS mode. This is the reason why I am
trying to find a solution which runs under Windows.

Hans Ruegg.

[Matthew Berg]

As for running in DOS mode, I personally had much trouble when trying to configure my system in order to support mouse, soundcard, etc. in DOS mode, and I wonder if other users have the same problems. (I am aware this is outside the scope of this subject, but I did not yet find a DOS compatible mouse driver on the Windows CD, and even EMM386 returns an error message when I try to put it into the CONFIG.SYS file ... there are still many
things I do not understand yet.)

I'm not surprised you couldn't find a DOS mouse driver on your Windows CD. You can download v11.0 of the standard Micro$oft MS-DOS Mouse Driver from their web site. For DOS support of your sound card, you will need specific drivers from the card manufacturer. I find it useful to use MEMMAKER to optimise my startup files for DOS.