User:Ulfalizer: Difference between revisions
(Mystery tick for sprite zero should read as 1 if ignoring propagation delays) |
(Read/write timing) |
||
| (One intermediate revision by the same user not shown) | |||
| Line 62: | Line 62: | ||
* Values appear to be read during '''φ2''' ('''db''x''''' pins buffered on '''cclk'''), and '''db''' changes during '''φ2''' for writes too. | * Values appear to be read during '''φ2''' ('''db''x''''' pins buffered on '''cclk'''), and '''db''' changes during '''φ2''' for writes too. | ||
* [[Visual_circuit_tutorial#Clocks|Clocks section of tutorial]]. | * [[Visual_circuit_tutorial#Clocks|Clocks section of tutorial]]. | ||
=== Read/write timing === | |||
"The address is guaranteed to be stable 300 nanoseconds after the leading edge of Phase One, and the data must be stable 100 nanoseconds before the trailing edge of Phase Two. At 1.0 MHz operation, this allows the memory devices approximately 575 ns to make data available on the data bus.", from http://users.telenet.be/kim1-6502/6502/hwman.html. | |||
== [[CPU_interrupts|Interrupts]] == | == [[CPU_interrupts|Interrupts]] == | ||
| Line 283: | Line 287: | ||
CPU: [ Low ][ High ] -> 1 | CPU: [ Low ][ High ] -> 1 | ||
CPU: [ Low ][ High ] -> 1 | CPU: [ Low ][ High ] -> 1 | ||
</pre> | |||
= [[PPU sprite evaluation|Sprite overflow bug]] = | |||
Appears to be caused by a timing issue. | |||
Signal trace for a "good" sprite skip (eight sprites not found yet, current sprite not in range): | |||
<pre> | |||
hpos spr_addr_clear_low_bump_high_setup spr_addr_load_next_value /spr_load_next_value_or_write_2003_reg | |||
04c 0 0 1 | |||
04c 0 0 1 | |||
04c 0 0 1 | |||
04c 0 0 1 | |||
04c 0 0 1 | |||
04c 0 0 1 | |||
04c 0 0 1 | |||
04c 0 0 1 | |||
04b 1 0 1 | |||
04b 1 0 1 | |||
04b 1 0 1 | |||
04b 1 0 1 | |||
04b 1-----------------------------------1 0 | |||
04b 1 Overlap 1 0 | |||
04b 1 Clears low and bumps high 1 0 | |||
04b 1-----------------------------------1 0 | |||
04a 1 0 1 | |||
04a 1 0 1 | |||
04a 1 0 1 | |||
04a 1 0 1 | |||
04a 1 0 1 | |||
04a 1 0 1 | |||
04a 1 0 1 | |||
04a 1 0 1 | |||
049 0 0 1 | |||
049 0 0 1 | |||
049 0 0 1 | |||
049 0 0 1 | |||
049 0 1 0 | |||
049 0 1 0 | |||
049 0 1 0 | |||
049 0 1 0 | |||
048 0 0 1 | |||
</pre> | |||
Signal trace for a glitchy sprite skip (eight sprites found, current sprite not in range): | |||
<pre> | |||
hpos spr_addr_clear_low_bump_high_setup spr_addr_load_next_value /spr_load_next_value_or_write_2003_reg | |||
08e 1 0 1 | |||
08e 1 0 1 | |||
08e 1 0 1 | |||
08e 1 0 1 | |||
08e 1 0 1 | |||
08e 1 0 1 | |||
08e 1 0 1 | |||
08e 1 0 1 | |||
08d 0 0 1 | |||
08d 0 0 1 | |||
08d 0 0 1 | |||
08d 0 0 1 | |||
08d 0-----------------------------------1 0 | |||
08d 0 No overlap 1 0 | |||
08d 0 Bumps low and high 1 0 | |||
08d 0-----------------------------------1 0 | |||
08c 1 0 1 | |||
08c 1 0 1 | |||
08c 1 0 1 | |||
08c 1 0 1 | |||
08c 1 0 1 | |||
08c 1 0 1 | |||
08c 1 0 1 | |||
08c 1 0 1 | |||
08b 0 0 1 | |||
08b 0 0 1 | |||
08b 0 0 1 | |||
08b 0 0 1 | |||
08b 0 1 0 | |||
08b 0 1 0 | |||
08b 0 1 0 | |||
08b 0 1 0 | |||
08a 1 0 1 | |||
</pre> | </pre> | ||
Latest revision as of 21:37, 20 August 2013
Misc. timing stuff brought together in one place
Reads and writes
Read (LDA $1234, NOP)
Tick-by-tick from http://nesdev.org/6502_cpu.txt interleaved with steps from Visual 2A03:
# address R/W description
--- ------- --- -------------------------------------------------
1 PC R fetch opcode, increment PC
ab db rw pc phi2
0000 ad 1 0000 0
0000 ad 1 0000 1
2 PC R fetch low byte of address, increment PC
ab db rw pc phi2
0001 34 1 0001 0
0001 34 1 0001 1
3 PC R fetch high byte of address, increment PC
ab db rw pc phi2
0002 12 1 0002 0
0002 12 1 0002 1
4 address R read from effective address
ab db rw pc phi2
1234 00 1 0003 0
1234 00 1 0003 1
Write (LDA #$AB, STA $1234, NOP)
Tick-by-tick from http://nesdev.org/6502_cpu.txt interleaved with steps from Visual 2A03:
# address R/W description
--- ------- --- ------------------------------------------
1 PC R fetch opcode, increment PC
ab db rw pc phi2
0002 8d 1 0002 0
0002 8d 1 0002 1
2 PC R fetch low byte of address, increment PC
ab db rw pc phi2
0003 34 1 0003 0
0003 34 1 0003 1
3 PC R fetch high byte of address, increment PC
ab db rw pc phi2
0004 12 1 0004 0
0004 12 1 0004 1
4 address W write register to effective address
ab db rw pc phi2
1234 12 0 0005 0
1234 ab 0 0005 1
Read/write observations
- Address bus and rw changes right away (during φ1).
- Values appear to be read during φ2 (dbx pins buffered on cclk), and db changes during φ2 for writes too.
- Clocks section of tutorial.
Read/write timing
"The address is guaranteed to be stable 300 nanoseconds after the leading edge of Phase One, and the data must be stable 100 nanoseconds before the trailing edge of Phase Two. At 1.0 MHz operation, this allows the memory devices approximately 575 ns to make data available on the data bus.", from http://users.telenet.be/kim1-6502/6502/hwman.html.
Interrupts
Sampled on the falling edge of φ2. IRQ detection depends on nodes nnT2BR (branch-related) and 646 (maybe a "sampling points" signal) being low.
Detection intervals for various instructions
Smallest IRQ assertion interval that will trigger an IRQ for LDA #FF: http://visual6502.org/JSSim/expert.html?logmore=Execute,irq,IRQP,State,nnT2BR,646&a=0&d=58A9FF1890FE&irq0=5&irq1=6&steps=20
Ditto for LSR $AB: http://visual6502.org/JSSim/expert.html?logmore=Execute,irq,IRQP,State,nnT2BR,646&a=0&d=5846AB1890FE&irq0=11&irq1=12&steps=20
Ditto for LSR $AB with NMI instead of IRQ: http://visual6502.org/JSSim/expert.html?logmore=Execute,nmi,~NMIG,State&a=0&d=58461890FE&nmi0=11&nmi1=12&steps=20
IRQ triggered by assertion during first cycle of BCC - no page crossing, offset 0, pad with SEDs: http://visual6502.org/JSSim/expert.html?logmore=Execute,irq,IRQP,State,nnT2BR,646&a=0&d=58189000F8F8&irq0=9&irq1=10&steps=30
IRQ triggered by assertion during first cycle of BCC - page crossing, offset E0 (backwards jump), pad with SEDs: http://visual6502.org/JSSim/expert.html?logmore=Execute,irq,IRQP,State,nnT2BR,646&a=0&d=581890E0&a=ffe4&d=F8F8&irq0=9&irq1=10&steps=30
IRQ triggered by assertion before fixup cycle of BCC - page crossing, offset E0 (backwards jump), pad with SEDs: http://visual6502.org/JSSim/expert.html?logmore=Execute,irq,IRQP,State,nnT2BR,646&a=0&d=581890E0&a=ffe4&d=F8F8&irq0=13&irq1=14&steps=30
IRQ during JMP Absolute (to after JMP instruction): http://visual6502.org/JSSim/expert.html?logmore=Execute,irq,IRQP,State,nnT2BR,646&a=0&d=584C0400F8F8&irq0=7&irq1=8&steps=30
IRQ during JMP Relative (address stored after JMP instruction, to next location after that): http://visual6502.org/JSSim/expert.html?logmore=Execute,irq,IRQP,State,nnT2BR,646&a=0&d=586C04000600F8F8&irq0=11&irq1=12&steps=30
RTI - interrupt flag changes *before* poll location: http://visual6502.org/JSSim/expert.html?logmore=Execute,irq,IRQP,State,nnT2BR,646&a=0&d=40&irq0=9&irq1=10&steps=30
CLI+SEI - interrupt flag changes *after* poll location for both: http://visual6502.org/JSSim/expert.html?logmore=Execute,irq,IRQP,State,nnT2BR,646&a=0&d=5878&irq0=5&irq1=6&steps=30
PLP - interrupt flag changes *after* poll location: http://visual6502.org/JSSim/expert.html?logmore=Execute,irq,IRQP,State,nnT2BR,646&a=0&d=28F8F8&irq0=5&irq1=6&steps=30
Interrupt hijacking
http://en.wikipedia.org/wiki/Interrupts_in_65xx_processors
BRK hijacked by NMI: http://visual6502.org/JSSim/expert.html?logmore=Execute,nmi,~NMIG,State,nnT2BR,646&a=0&d=5800&nmi0=11&nmi1=12&steps=30
IRQ hijacked by NMI: http://visual6502.org/JSSim/expert.html?logmore=Execute,irq,IRQP,nmi,~NMIG,State,nnT2BR,646&a=0&d=58F8F8&irq0=5&irq1=6&nmi0=15&nmi1=16&steps=30
Reset
Seems to take 1+7 cycles, where the last 7 is the ordinary interrupt sequence.
M2 duty cycle
The M2 duty cycle is 5/8 - forum post and CPU pin-out page. Low for 9/8 PPU cycles, high for 15/8 PPU cycles. Low for 4.5 master cycles, high for 7.5 master cycles. Low for ~210 ns, high for ~349 ns.
PPU interface
Input map:
Binding | Interpretation ----------------+-------------------------------------- 0 -> 15 | 0 -> /Enable second demultiplexer a15 -> 13 | a15 -> A_1b a14 -> 3 | a14 -> A_1a a13 -> 2 | a13 -> A_0a M2 -> 14 | M2 -> A_0b 11 -> 1 | /O_1b -> /Enable first demultiplexer 9 -> /ROM SEL | /O_3b -> /ROM SEL 5 -> /DBE | /O_1a -> /DBE 4 -> RAM CS | /O_0a -> RAM CS
Input/output map:
Inputs | Outputs ---------+---------------- E A0 A1 | O_0 O_1 O_2 O_3 ---------+---------------- 1 x x | 1 1 1 1 0 0 0 | 0 1 1 1 0 1 0 | 1 0 1 1 0 0 1 | 1 1 0 1 0 1 1 | 1 1 1 0
Hence:
ROM SEL = M2 && a15 <Enable first demultiplexer> = M2 && !a15 DBE = <Enable first demultiplexer> && !a14 && a13 RAM CS = <Enable first demultiplexer> && !a14 && !a13
Summary:
When M2 high,
A15-13 | Signal -------+-------- 1xx | ROM SEL 001 | DBE 000 | RAM CS
- rw signal together with ab lines directly generates signals like /r2002.
- _io_ce follows inverted M2.
- vbl_flag is set via a vpos was 241 and vpos is not 241 comparison. Goes high at hpos = 1 during at pclk0.
- The set_vbl_flag signal is high during pclk0 of vpos=241/hpos=1.
- spr0_hit goes high at tick x+2, during pclk1. Reading uses same _io_ce behavior as vbl_flag.
- spr_overflow reading behavior just like spr0_hit. TODO: set timing.
- $2005/$2006 and $2000 nametable bits writes use the same timing.
Timing charts
Around VBL flag setting:
set_vbl_flag signal active during this interval
|
[++]
Master: 0101010101010101010101010101010101010101010101010101010101010101010101010
PPU: [p0][p1][p0][p1][p0][p1][p0][p1][p0][p1][p0][p1][p0][p1][p0][p1][p0][p1]
CPU: [ Low ][ High ] -> 0, NMI (-8)
CPU: [ Low ][ High ] -> 0, NMI (-7)
CPU: [ Low ][ High ] -> 0, no NMI (-6)
CPU: [ Low ][ High ] -> 0, no NMI (-5)
CPU: [ Low ][ High ] -> 0, no NMI (-4)
CPU: [ Low ][ High ] -> 0, no NMI (-3)
CPU: [ Low ][ High ] -> 0, no NMI (-2)
CPU: [ Low ][ High ] -> 0, no NMI (-1)
CPU: [ Low ][ High ] -> 1, no NMI (0)
CPU: [ Low ][ High ] -> 1, no NMI (1)
CPU: [ Low ][ High ] -> 1, no NMI (2)
CPU: [ Low ][ High ] -> 1, no NMI (3)
CPU: [ Low ][ High ] -> 1, no NMI (4)
CPU: [ Low ][ High ] -> 1, no NMI (5)
CPU: [ Low ][ High ] -> 1, no NMI (6)
CPU: [ Low ][ High ] -> 1, NMI (7)
CPU: [ Low ][ High ] -> 1, NMI (8)
(NOTE: Read buffer latches the data, so value when high phase starts matters. Value at end probably matters for the rest.)
Around VBL flag clearing: cleared during this interval
[+++++++++++...
Master: 0101010101010101010101010101010101010101010101010101010101010101010101010
PPU: [p0][p1][p0][p1][p0][p1][p0][p1][p0][p1][p0][p1][p0][p1][p0][p1][p0][p1]
CPU: [ Low ][ High ] -> 1
CPU: [ Low ][ High ] -> 1
CPU: [ Low ][ High ] -> 1
CPU: [ Low ][ High ] -> 1
CPU: [ Low ][ High ] -> 1
CPU: [ Low ][ High ] -> 1
CPU: [ Low ][ High ] -> 1
CPU: [ Low ][ High ] -> 0 (?)
CPU: [ Low ][ High ] -> 0
CPU: [ Low ][ High ] -> 0
CPU: [ Low ][ High ] -> 0
CPU: [ Low ][ High ] -> 0
CPU: [ Low ][ High ] -> 0
CPU: [ Low ][ High ] -> 0
CPU: [ Low ][ High ] -> 0
CPU: [ Low ][ High ] -> 0
CPU: [ Low ][ High ] -> 0
Sprite zero/overflow flag clearing: cleared during this interval
[+++++++++++....
Master: 0101010101010101010101010101010101010101010101010101010101010101010101010
PPU: [p0][p1][p0][p1][p0][p1][p0][p1][p0][p1][p0][p1][p0][p1][p0][p1][p0][p1]
CPU: [ Low ][ High ] -> 1
CPU: [ Low ][ High ] -> 1
CPU: [ Low ][ High ] -> 1
CPU: [ Low ][ High ] -> 1
CPU: [ Low ][ High ] -> 0 (?)
CPU: [ Low ][ High ] -> 0
CPU: [ Low ][ High ] -> 0
CPU: [ Low ][ High ] -> 0
CPU: [ Low ][ High ] -> 0
CPU: [ Low ][ High ] -> 0
CPU: [ Low ][ High ] -> 0
CPU: [ Low ][ High ] -> 0
CPU: [ Low ][ High ] -> 0
CPU: [ Low ][ High ] -> 0
Pixel location Flag goes high
Sprite 0 setting: -------- -----------------...
Master: 0101010101010101010101010101010101010101010101010101010101010101010101010
PPU: [p0][p1][p0][p1][p0][p1][p0][p1][p0][p1][p0][p1][p0][p1][p0][p1][p0][p1]
CPU: [ Low ][ High ] -> 0
CPU: [ Low ][ High ] -> 0
CPU: [ Low ][ High ] -> 0
CPU: [ Low ][ High ] -> 0
CPU: [ Low ][ High ] -> 0
CPU: [ Low ][ High ] -> 0
CPU: [ Low ][ High ] -> 1 (?)
CPU: [ Low ][ High ] -> 1
CPU: [ Low ][ High ] -> 1
CPU: [ Low ][ High ] -> 1
CPU: [ Low ][ High ] -> 1
CPU: [ Low ][ High ] -> 1
CPU: [ Low ][ High ] -> 1
CPU: [ Low ][ High ] -> 1
CPU: [ Low ][ High ] -> 1
CPU: [ Low ][ High ] -> 1
CPU: [ Low ][ High ] -> 1
Sprite overflow bug
Appears to be caused by a timing issue.
Signal trace for a "good" sprite skip (eight sprites not found yet, current sprite not in range):
hpos spr_addr_clear_low_bump_high_setup spr_addr_load_next_value /spr_load_next_value_or_write_2003_reg 04c 0 0 1 04c 0 0 1 04c 0 0 1 04c 0 0 1 04c 0 0 1 04c 0 0 1 04c 0 0 1 04c 0 0 1 04b 1 0 1 04b 1 0 1 04b 1 0 1 04b 1 0 1 04b 1-----------------------------------1 0 04b 1 Overlap 1 0 04b 1 Clears low and bumps high 1 0 04b 1-----------------------------------1 0 04a 1 0 1 04a 1 0 1 04a 1 0 1 04a 1 0 1 04a 1 0 1 04a 1 0 1 04a 1 0 1 04a 1 0 1 049 0 0 1 049 0 0 1 049 0 0 1 049 0 0 1 049 0 1 0 049 0 1 0 049 0 1 0 049 0 1 0 048 0 0 1
Signal trace for a glitchy sprite skip (eight sprites found, current sprite not in range):
hpos spr_addr_clear_low_bump_high_setup spr_addr_load_next_value /spr_load_next_value_or_write_2003_reg 08e 1 0 1 08e 1 0 1 08e 1 0 1 08e 1 0 1 08e 1 0 1 08e 1 0 1 08e 1 0 1 08e 1 0 1 08d 0 0 1 08d 0 0 1 08d 0 0 1 08d 0 0 1 08d 0-----------------------------------1 0 08d 0 No overlap 1 0 08d 0 Bumps low and high 1 0 08d 0-----------------------------------1 0 08c 1 0 1 08c 1 0 1 08c 1 0 1 08c 1 0 1 08c 1 0 1 08c 1 0 1 08c 1 0 1 08c 1 0 1 08b 0 0 1 08b 0 0 1 08b 0 0 1 08b 0 0 1 08b 0 1 0 08b 0 1 0 08b 0 1 0 08b 0 1 0 08a 1 0 1
