| |
Frostbyte
Registered: Aug 2003
Posts: 171 |
Most efficient (in cycles) 25Hz music player
Last night, just for funsies and obviously inspired by the 25Hz music compo, I began wondering what is the fastest way to skip playing every 2nd frame. I came up with this.
First in the music init, I also init a toggle counter on zeropage address $02 to 00:
lda #$00
sta $02
Then, in the music play IRQ, I do this:
....
lda #$ff // 2 cycles - set accu to FF
eor $02 // 3 cycles - XOR with toggle: accu becomes 00 if toggle is FF, FF if 00
sta $02 // 3 cycles - update toggle counter to either FF or 00
bne end_irq // 2 cycles - skip playing if counter is not 00
jsr music.play
end_irq:
....
Of course it'd be much more efficient to use this or some other code to skip the music IRQ completely every 2nd frame, but you guys get the point. :)
This was a beginner's 10 cycle approach to the problem. Any faster solutions? |
|
| |
chatGPZ
Registered: Dec 2001
Posts: 11136 |
A much better approach would be splitting the workload into two (more or less) equal portions that you call alternating. Unless you totally insist on calling the player at 25Hz for whatever reason :) |
| |
Frostbyte
Registered: Aug 2003
Posts: 171 |
Quote: A much better approach would be splitting the workload into two (more or less) equal portions that you call alternating. Unless you totally insist on calling the player at 25Hz for whatever reason :)
Sorry I wasn't very clear in what I was after. I was thinking more in the lines of e.g. playing a Goattracker 25Hz tune, i.e. calling the unmodified GT playroutine every second frame. I'm guessing by splitting the workload you mean splitting the work a playrountine itself does? This would require modifying the playroutines of the tunes themselves. I'm just after an easy and fast solution to call e.g. $1003 every other frame. :) |
| |
Conrad
Registered: Nov 2006
Posts: 833 |
Another approach? ... (won't work on raster-lines #$00-$7f, but that's not what this is about, right?)
This obviously includes the code for raster compare. Tested in vice monitor.
lda #$00
jsr music_init
play:
lda #$80 <-- bit 7 MUST be set in accumulator, therefore rasterline has to be between #$80 and #$ff
!: cmp $d012
bne !-
play_25hz:
anc #$aa // 2 cycles / 2 bytes (#$aa = #%10101010)
rol play_25hz+1 // 4 cycles / 3 bytes
bpl play_25hz_skip // 2 cycles / 2 bytes
jsr music_play
play_25hz_skip:
!: bit $d012
bmi !-
bpl play
EDIT:
With this you could make use of other ANC #$-- bit values to do other things than just play every second frame:
#% 11111111 = Play every frame ("50hz")
#% 10101010 = Play every second frame ("25hz")
#% 10001000 = Play every fourth frame ("12.5hz")
#% 01110111 = "Skip" every fourth frame
#% 01111011 = "Skip" every 1st and 6th (out of 8th) frame :)
#% 10110110 = Experimental, etc. ;) |
| |
Krill
Registered: Apr 2002
Posts: 2852 |
Set up a spare CIA timer. Deciding whether to call the player or not should then be just a matter of a CIA register read plus a branch. Bonus points for folding both into a JMP (CIAREG). But then, really... but why. =) |
| |
Copyfault
Registered: Dec 2001
Posts: 466 |
Maybe I still didn't get the point, but wouldn't putting the music_play-call in a timer irq that is triggered every 312*63*2 cycles be the shortest way to do it? |
| |
Oswald
Registered: Apr 2002
Posts: 5025 |
copyfault wins |
| |
Krill
Registered: Apr 2002
Posts: 2852 |
Quoting Jojelior some other code to skip the music IRQ completely every 2nd frame, but you guys get the point. :) Oswald didn't get the point. But i'm not sure anyone besides Jojeli did. |
| |
Oswald
Registered: Apr 2002
Posts: 5025 |
he doesnt excludes irq :) his example happens in an irq in fact. |
| |
Frostbyte
Registered: Aug 2003
Posts: 171 |
Apologies, even though the context of my question was very clear in my head, I completely failed to communicate that in my initial post. :)
So, in my head, the scenario is to play a 25Hz Goattracker (or some other tracker) tune in e.g. a demo. I was thinking of using a raster interrupt mainly because I haven't done any CIA timer based players, in fact as a n00b it didn't even cross my mind. Given this scenario, and the fact that I was using a raster interrupt which usually triggers 50 times per second, I wanted to come up with some piece of code that skips the jump to the music playroutine (or bypasses the music player IRQ completely) every second execution, with the least overhead compared to the 50Hz "jsr $1003 every time" player.
So in my example, the code is used to skip every 2nd jump to the playroutine within the player IRQ. I guess it could also be used in a previous IRQ to choose which IRQ gets executed next (that being either player IRQ, or the IRQ that comes after the player IRQ), but I'm guessing the overhead of the deciding/branching code itself would still be the same.
Anyway, a CIA timer IRQ (or just reading the CIA register in a raster IRQ, as Krill probably suggested? Some of this stuff goes already above my noob head...) seems like the least cycle-consuming way to get something done 25 times per second.
Sorry for this messy ramble... I know it is quite a pointless coding exercise anyway, given the benefits of maybe saving a few cycles per frame, but I just like this kind of tinkering. :) I was also inspired by the previous discussion about shortest code for stable raster timer setup, albeit a) it was perhaps a more useful coding exercise, and b) I admit I didn't understand half of it. |
| |
Oswald
Registered: Apr 2002
Posts: 5025 |
its quite simple you can tell CIA to generate an interrupt at every xth cpu cycle, evey xth can be a 16 bit number, you can read in the manuals how exactly this have to be done. |
| |
Stryyker
Registered: Dec 2001
Posts: 465 |
Or for you playing music each frame:
jsr $1003
lda $1003
eor #$2c
sta $1003
This assumes $1003 has a jmp. Basically toggle $1003 opcode between whatever it has a $60 to do a rts. |
| |
chatGPZ
Registered: Dec 2001
Posts: 11136 |
not that it saves anything over the solution provided in the first post :) |
| |
Stryyker
Registered: Dec 2001
Posts: 465 |
If you want to save more cycles, 2 raster interrupts in the same page so you only need to do lda #<rasterirq2 sta $0314 or $fffe and lda #<raster1 sta $0314 or $fffe so you have 6 cycles changing at the expense of more memory use |
| |
Copyfault
Registered: Dec 2001
Posts: 466 |
Yes, I had the feeling a RASTER IRQ was kinda set. But since the options for this had more or less been posted already, I just wanted to put up the timer-irq approach;)
At the end it's all for fun, and I'm happy that the 25Hz-Compo also sparked something for the coders (be it this discussion or some disco... aehh. funky releases :)=) ) |
| |
Oswald
Registered: Apr 2002
Posts: 5025 |
Quote: If you want to save more cycles, 2 raster interrupts in the same page so you only need to do lda #<rasterirq2 sta $0314 or $fffe and lda #<raster1 sta $0314 or $fffe so you have 6 cycles changing at the expense of more memory use
you should have read all the posts, the ultimate solution was found already. |
| |
TWW
Registered: Jul 2009
Posts: 541 |
I'm also probably missing the point, but if it's cycles / frame, why is:
jsr $1000
lda $1000 // 4 cycles
eor #$2c // 2 cycles
sta $1000 // 4 cycles
using more cycles than
IRQ // 7 cycles
jsr $1003
rti // 7 cycles
or was it ment something like this (6,5 cycles):
bit $dc0x
bpl !+
jsr $1003
!:
Considering the overhead on the IRQ method there is no memory saved.
If memory is no concern, you could easily make it make it 0 cycles (add more raster-IRQ chains). <- New ultimate solution?
What was the point here? |
| |
Oswald
Registered: Apr 2002
Posts: 5025 |
bit $dc0x
bpl !+
jsr $1003
!:
you will also need a jmp !- |
| |
TWW
Registered: Jul 2009
Posts: 541 |
Quote: bit $dc0x
bpl !+
jsr $1003
!:
you will also need a jmp !-
We are probably thinking about this differently?
I was thinking to set the timer so it is alternating between negative and positive each frame (so it skipps the call to the music routine every 2nd frame). i didn't do the math though and sure as hell didn't test it so it's out of my ass only.
Edit:
and IRQ chains = 0 cycles:
IRQ1
:Vsync
Play music
Do stuff
IRQ pointer => IRQ2
IRQ2
:Vsync
Do not play music
Do stuff
IRQ pointer => IRQ1 |
| |
Copyfault
Registered: Dec 2001
Posts: 466 |
Another idea if we assume RASTER IRQ as fixed component of the concept:
lda #<msx_call
call:
jsr msx_call*
sta* call+1
and one has to ensure that the msx_call-routine offers a byte (not necessarily accu, hence the sta*) that is free to use in the msx_call-page; an RTS has to go there ofcourse.
Unfortunately, this is short on additional bytes, but since we're doing the JSR every frame (every other just used for skipping), it effectively increases the no. of cycles.
More elegant would be
call:
jsr $20a9
sta call
and taking the "bitter pill" to add an "LDA #$EA" to the end of the msx-call, which in turn must be located at $20a9.
Best in terms of additional cycles is to use a timer irq, impossible to get below when RASTER IRQ-only is a must;) |
| |
Copyfault
Registered: Dec 2001
Posts: 466 |
Quote:[...]More elegant would be
call:
jsr $20a9
sta call
and taking the "bitter pill" to add an "LDA #$EA" to the end of the msx-call, which in turn must be located at $20a9.[...] Ah, silly me, ofcourse it's not necessary to tweak the music routine itself, just:
call:
jsr $0c20
lda #$a9
sta call
There are some othter candidates for the hi-byte ($0c = NOP abs was chosen here), depending on the desired location of the music routine (relocation needed either way). |
