Detecting the Hardware: One Routine for PAL/NTSC, OS, Memory, Stereo and VBXE
Modern Atari software often adapts to the machine it finds: music players switch to stereo, games correct their speed for PAL, demos use extended memory or a VBXE when one is present. All of that starts with a short detection routine run at start-up. This guide builds one that checks seven things in one pass and prints a report. Each test is small and explained on its own, so you can lift the ones you need. The modern utility HW-Detect does the same job in much more depth (see HW-Detect: One Program to Identify Your Atari's Hardware). This guide is about how such tests work.
What you'll learn: how to read the video standard from GTIA and from ANTIC (and why both), how to tell the 400/800 OS from the XL/XE OS without crashing either, where the OS records the top of RAM, a safe 130XE bank test, a second-POKEY test, and how a VBXE identifies itself.
What you need: MADS and an emulator. The program was assembled with MADS and run in atari800 as four machines: an NTSC 800XL (genuine XL/XE OS Rev. 2), a PAL 130XE with stereo, a 400/800 with OS-B, and a 320K RAMBO XL with BASIC and stereo. All four reports were correct. We had no VBXE to test with: that test follows the documented register layout, and we only saw its "not found" result.
1. The rules for detection code
- Read before you write. Some tests have to write to hardware, but a write to the wrong place on the wrong machine does damage. PORTB is the best example: on an XL/XE it controls memory, but on a 400/800 it is the joystick port for sticks 3 and 4. The program only touches it after the OS test has said "XL/XE".
- Put everything back. Every byte the tests overwrite is saved and restored, and so are interrupt settings.
- Keep the code out of the way. The bank test switches $4000–$7FFF, so the routine sits at $3000–$330E, below that window.
2. Video standard: ask both chips
GTIA has a read-only register at $D014 (53268), PAL. Bits 1–3 read 111 on an NTSC GTIA and 000 on a PAL or SECAM one. In practice NTSC chips return $0F and PAL chips $01, so testing AND #$0E is safe.
GTIA only tells you about the colour encoding. It is ANTIC that sets the frame timing, 262 lines (about 60 Hz) or 312 lines (about 50 Hz), and the two chips can be mismatched. Machines exist with an NTSC ANTIC and a PAL GTIA (a 60 Hz PAL picture) and the reverse (50 Hz NTSC). In such a machine, ANTIC sets the frame timing and GTIA decides what the PAL register says. For timing music and game speed, ANTIC is what matters, and ANTIC has no ID register. Instead, the program watches VCOUNT ($D40B) for three frames and keeps the highest value seen. VCOUNT is the scan line divided by two, so it peaks at $82 on a 262-line ANTIC and $9B on a 312-line one.
3. Which OS?
The program needs to know whether this is an XL/XE OS before it dares to touch PORTB. It uses a fixed difference between the ROMs. On the XL/XE OS, $E480 is the start of the new jump vector PUPDIV, so it holds a JMP opcode ($4C). On the 400/800 OS, the same address holds the initial value for the DLI vector: we read $90 there on OS-B. For the 400/800 case, the program goes one step further: $E40F reads 56 on OS-A and 0 on OS-B. We could only check the OS-B branch.
These tests identify the OS, not the machine. An XL running the Translator's 400/800 OS from RAM, or a 400/800 fitted with an XL-compatible replacement OS, will report the OS it is running. For most programs, that is exactly the right answer.
4. How much RAM?
At power-up the OS searches for the end of RAM and stores the result, as a number of 256-byte pages, in RAMSIZ ($02E4, 740). A cartridge or the built-in BASIC sits at $A000, so the value is $A0 (40K) when one is present and $C0 (48K) on a machine without. The RAMBO run above was started with BASIC enabled and reported $A000, the others $C000.
Extended memory is a separate question. On an XL/XE, the program does a minimal 130XE-style test with interrupts and NMIs off:
- Save byte $4000 of each of the four banks selected by PORTB bits 2–3 (with bit 4 = 0 to open the window), and of main memory.
- Write a different tag into each bank, then $FF into main memory.
- Count the banks that still hold their own tag.
- Restore every saved byte, PORTB, NMIEN and the I flag.
On a 64K XL every write lands in main memory, the final $FF overwrites the tag, and the count is 0. A 130XE gives 4, as does the 320K RAMBO, because this test only looks at the two bank bits a 130XE has. For larger expansions, use the 64-combination counter in Memory Expansions Beyond the 130XE: 192K to 1MB and How to Use Them.
5. A second POKEY
The stereo upgrade puts a second POKEY at the odd 16-byte mirrors of the POKEY page: $D210–$D21F. On a mono machine those addresses are mirrors of the first chip. So the program sets up timer 1 with the shortest period, enables its interrupt and starts it, all through the $D21x addresses, with IRQs masked. Then it reads IRQST on the first chip. If timer 1's bit has gone to 0 (pending), the writes reached the first chip and the machine is mono. If nothing happened there, a second chip took the writes. This is the principle behind the timer-based detector in Stereo POKEY Programming: Detecting and Driving a Second POKEY. Finally the program switches the test timer off and restores IRQEN from POKMSK.
6. A VBXE
The VideoBoard XE has a 32-byte register block at either $D640 or $D740, depending on how it is configured. Its first register, CORE_REVISION, is read-only. Its high nibble holds the major version (1 for current cores), and the low nibble is 0 for the full FX core and 1 for a GTIA-only core. FX core 1.26 reads $11 there, with MINOR_REVISION ($Dx41) reading $26. The program checks both addresses for a value whose high nibble is 1. On our test machines without a VBXE, $D640 and $D740 read as something else and the report said "NOT FOUND".
7. The program
; One-pass hardware report: video standard (GTIA and ANTIC), OS family,
; top of RAM, 130XE-style extended banks, second POKEY, VBXE.
POKMSK = $10
RTCLOK = $12
RAMSIZ = $02E4
ICCOM = $0342
ICBLL = $0348
PAL = $D014
IRQEN = $D20E
IRQST = $D20E
PORTB = $D301
VCOUNT = $D40B
NMIEN = $D40E
CIOV = $E456
SPTR = $CB
org $3000
start jsr say
dta c'GTIA VIDEO: ',0
lda PAL ; bits 1-3: 111 = NTSC, 000 = PAL
and #$0E
bne gnt
jsr say
dta c'PAL',$9B,0
jmp antic
gnt jsr say
dta c'NTSC',$9B,0
antic lda #0 ; ANTIC: highest VCOUNT seen in 3 frames
sta maxv
lda RTCLOK+2
clc
adc #3
sta endf
av lda VCOUNT
cmp maxv
bcc av1
sta maxv
av1 lda RTCLOK+2
cmp endf
bne av
jsr say
dta c'ANTIC LINES: ',0
lda maxv
cmp #$90 ; $82 = 262 lines, $9B = 312 lines
bcs apal
jsr say
dta c'262 (NTSC)',$9B,0
jmp osid
apal jsr say
dta c'312 (PAL)',$9B,0
osid jsr say
dta c'OS: ',0
lda $E480 ; XL/XE OS: JMP PUPDIV here; 400/800: data
cmp #$4C
bne os800
lda #1
sta isxl
jsr say
dta c'XL/XE',$9B,0
jmp ram
os800 lda $E40F ; 400/800: 56 on OS-A, 0 on OS-B
bne osa
jsr say
dta c'400/800 OS-B',$9B,0
jmp ram
osa jsr say
dta c'400/800 OS-A (OR OTHER)',$9B,0
ram jsr say
dta c'RAM TOP: $',0
lda RAMSIZ ; pages of RAM found below any cartridge
jsr hex
jsr say
dta c'00',$9B,0
jsr say
dta c'XE BANKS: ',0
lda isxl ; PORTB is a joystick port on the 400/800:
bne banks ; never write it there
jsr say
dta c'N/A ON 400/800',$9B,0
jmp pokey
banks jsr xebanks
txa
jsr hex
lda #$9B
jsr putc
pokey jsr say
dta c'POKEYS: ',0
jsr stereo
bcc mono
jsr say
dta c'2 (STEREO)',$9B,0
jmp vbxe
mono jsr say
dta c'1',$9B,0
vbxe jsr say
dta c'VBXE: ',0
ldx #$D6 ; look at $D640, then $D740
vb1 stx vbx+2
vbx lda $D640 ; CORE_REVISION: high nibble = major version
and #$F0
cmp #$10
beq vbyes
inx
cpx #$D8
bne vb1
jsr say
dta c'NOT FOUND',$9B,0
jmp *
vbyes jsr say
dta c'CORE AT $',0
lda vbx+2
jsr hex
jsr say
dta c'40, REV $',0
lda vbx+1
sta vbm+1
lda vbx+2
sta vbm+2
ldx #0
vbm lda $D640 ; (patched to the found address)
jsr hex
lda #$9B
jsr putc
jmp *
; ---- 130XE-style banks: bits 2-3 select, bit 4 = 0 opens the window ----
xebanks lda PORTB
sta oldpb
sei
lda #0
sta NMIEN ; nothing may run while banks come and go
ldx #3
xb1 jsr setbank ; save $4000 of each bank
lda $4000
sta save,x
dex
bpl xb1
lda oldpb
ora #$10 ; main memory
sta PORTB
lda $4000
sta savm
ldx #3
xb2 jsr setbank ; tag bank X with $A0+X
txa
ora #$A0
sta $4000
dex
bpl xb2
lda oldpb
ora #$10
sta PORTB
lda #$FF ; main memory gets a different value
sta $4000
ldy #0 ; Y = banks that kept their tag
ldx #3
xb3 jsr setbank
txa
ora #$A0
cmp $4000
bne xb4
iny
xb4 dex
bpl xb3
ldx #3 ; put everything back
xb5 jsr setbank
lda save,x
sta $4000
dex
bpl xb5
lda oldpb
ora #$10
sta PORTB
lda savm
sta $4000
lda oldpb
sta PORTB
lda #$40
sta NMIEN
cli
tya
tax
rts
setbank txa ; PORTB = old value, bit 4 = 0, bits 2-3 = X
asl
asl
sta tmp
lda oldpb
and #%11100011
ora tmp
sta PORTB
rts
; ---- second POKEY: start timer 1 "on chip 2" and see if chip 1 notices ----
stereo sei
lda #0
sta IRQEN ; chip 1: no interrupts
sta $D21E ; chip 2 (or chip 1 again on a mono machine)
sta $D218 ; AUDCTL
sta $D211 ; AUDC1
sta $D210 ; AUDF1 = 0: fastest timer
lda #$01
sta $D21E ; enable timer 1 IRQ at the $D21x address
sta $D219 ; STIMER at the $D21x address
ldx #0
st1 dex ; give the timer time to run out
bne st1
lda IRQST ; chip 1: bit 0 = 0 means its timer fired
and #$01
sta tmp
lda #0
sta $D21E ; switch the test timer off again
lda POKMSK
sta IRQEN
cli
lda tmp
lsr ; carry = 1: chip 1 never saw it -> stereo
rts
; ---- output helpers ----
say pla ; print the zero-terminated string that
sta SPTR ; follows the JSR, then return past it
pla
sta SPTR+1
sy1 inc SPTR
bne sy2
inc SPTR+1
sy2 ldy #0
lda (SPTR),y
beq sy3
jsr putc
jmp sy1
sy3 lda SPTR+1
pha
lda SPTR
pha
rts
putc ldx #0 ; one character via CIO (length 0 = use A)
stx ICBLL
stx ICBLL+1
pha
lda #11
sta ICCOM
pla
jmp CIOV
hex pha
lsr
lsr
lsr
lsr
jsr hex1
pla
and #$0F
hex1 tax
lda hexc,x
jmp putc
hexc dta c'0123456789ABCDEF'
maxv dta 0
endf dta 0
isxl dta 0
oldpb dta 0
savm dta 0
tmp dta 0
save .ds 4
run start
Notes on the code:
sayprints the string that follows theJSR, then returns past it. The 6502 pushes the address of the last byte of theJSR, so the routine increments before reading, and leaves the pointer on the terminating 0 so the finalRTSlands just after it.putcuses CIO's PUT CHARACTERS command with a buffer length of zero, which makes CIO send the byte in A.- The VBXE check patches its own operand (
vbx+2) to step from $D640 to $D740. On a found core, the second read is patched the same way to print the revision.
Troubleshooting
- The bank test crashes: the code or its data sits inside $4000–$7FFF, or interrupts were left on while a bank was mapped.
- Banks reported on a 64K machine: you skipped the final write to main memory, so main RAM kept the last tag and matched.
- Stereo detected on a mono machine: check that the test writes go to $D21x and the read comes from $D20E, and that IRQs are masked during the test.
- Wrong video result in an emulator: emulators let you mix PAL and NTSC settings in odd ways. Compare both the GTIA and ANTIC lines of the report.
- The program hangs on a 400/800: something wrote to PORTB. Only do it after the XL/XE test.
Further reading
- Avery Lee, Altirra Hardware Reference Manual: GTIA's PAL register (chapter 6), ANTIC display timing and VCOUNT (chapter 4), PORTB bank switching (chapter 2) and the VideoBoard XE register map (chapter 12).
- Ian Chadwick, Mapping the Atari: PAL (53268), RAMSIZ (740), PORTB (54017) and the OS-A/OS-B test at 58383.
- On this site: HW-Detect: One Program to Identify Your Atari's Hardware, Stereo POKEY Programming: Detecting and Driving a Second POKEY, CTIA, GTIA and FGTIA: Which Graphics Chip Do You Have?, Using the 130XE's Extra 64K: PORTB Bank Switching, The OS Jump Table: Calling the Atari Operating System the Supported Way.
Credits
- Picture: The detection program on an emulated PAL 130XE with the stereo upgrade switched on. — Program authors; screenshot by the site
- Altirra Hardware Reference Manual — 6.1 GTIA NTSC/PAL detection ($0F/$01), 4.2 mixed PAL/NTSC systems and VCOUNT, 4.10 VCOUNT range, 2.6-2.7 PORTB/extended memory, 12.4 VideoBoard XE register map (CORE_REVISION, MINOR_REVISION, $D640/$D740) — Avery Lee (Altirra)
- Mapping the Atari — 740 RAMSIZ, 53268 PAL, 54017 PORTB (XL), 58383 ($E40F) OS-A/OS-B test, Appendix 12 XL vector PUPDIV at $E480 — AtariArchives.org
