Timing: Difference between revisions

From SNESdev Wiki
Jump to navigationJump to search
m (→‎CPU: prefer master clocks to master cycles)
(→‎Master Clock: add more precision and relationship to the chroma subcarrier)
 
(3 intermediate revisions by one other user not shown)
Line 4: Line 4:


The SNES master clock:
The SNES master clock:
* NTSC 21.447 MHz
* NTSC: 945/44 MHz ≈ 21.4773 MHz (6 times chroma).
* PAL 21.281 MHz
* PAL: 21.281370 MHz (4.8 times chroma). This is synthesized by multiplying a 17.734475 MHz (4 times chroma) crystal by 6/5.


== CPU ==
== CPU ==
Line 12: Line 12:


This gives some commonly quoted SNES CPU speeds, though none of them tell a complete story:
This gives some commonly quoted SNES CPU speeds, though none of them tell a complete story:
* 3.58 MHz fast-ROM (6-cycle)
* 3.58 MHz fast-ROM (6-clocks per cycle)
* 2.68 MHz slow-ROM (8-cycle)
* 2.68 MHz slow-ROM (8-clocks per cycle)
* 1.79 MHz other (12-cycle)
* 1.79 MHz other (12-clocks per cycle)


The speed of access depends on the memory region:<ref>[https://problemkaputt.de/fullsnes.htm#cpuclockcycles Fullsnes]: CPU Clock Cycles</ref>
The speed of access depends on the memory region:<ref>[https://problemkaputt.de/fullsnes.htm#cpuclockcycles Fullsnes]: CPU Clock Cycles</ref>
* 6-cycles for fast-ROM access, enabled via MEMSEL and accessed at an address of $800000 of higher.
* 6-clocks for fast-ROM access, enabled via MEMSEL and accessed at an address of $800000 of higher.
* 8-cycles for slow-ROM access.
* 8-clocks for slow-ROM access.
* 8-cycles for internal S-WRAM.
* 8-clocks for internal S-WRAM.
* 6-cycles for most [[MMIO registers]].
* 6-clocks for most [[MMIO registers]].
* 12-cycles for [[MMIO registers#JOYSER0|JOYSER0 and JOYSER1]].
* 12-clocks for [[MMIO registers#JOYSER0|JOYSER0 and JOYSER1]].
* 6-cycles for "internal" cycles not accessing memory (e.g. 2nd cycle of NOP)
* 6-clocks for "internal" cycles not accessing memory (e.g. 2nd cycle of NOP)


== Video ==
== Video ==


===Scanline===
===Scanline===
* 1364 master cycles = 341 dot cycles.
* 1364 master clocks = 341 dot cycles.
* The CPU pauses for 40 master cycles in the middle of each scanline for DRAM refresh, leaving 1324 active cycles per line.
* The CPU pauses for 40 master clocks in the middle of each scanline for DRAM refresh, leaving 1324 active clocks per line.
* The left edge of the picture begins at cycle 88 of the line and continues until 1112.
* The left edge of the picture begins at clock 88 of the line and continues until 1112.
* Scanline 0 is the end of vblank and beginning of rendering. It is hidden, and displays as a blank line.
* Scanline 0 is the end of vblank and beginning of rendering. It is hidden, and displays as a blank line.
* Scanlines 1-224 or 239 will normally render the visible image, unless force blanking is applied.
* Scanlines 1-224 or 239 will normally render the visible image, unless force blanking is applied.
Line 36: Line 36:


====Short and Long Scanlines====
====Short and Long Scanlines====
Scanlines are normally 1364 master cycles, but there are two special cases:
Scanlines are normally 1364 master clocks, but there are two special cases:
* Short scanline: NTSC with interlace off, field=1, V=240.
* Short scanline: NTSC with interlace off, field=1, V=240.
* Long scanline: PAL with interlace on, field=1, V=311.
* Long scanline: PAL with interlace on, field=1, V=311.
Line 45: Line 45:


Timings:
Timings:
* Normal: 1364 cycles, 340 dots. 336 dots of 4-cycles, 4 dots of 5-cycles.
* Normal: 1364 clocks, 340 dots. 336 dots of 4-clocks, 4 dots of 5-clocks.
* Long: 1368 cycles, 341 dots. 337 dots of 4-cycles, 4 dots of 5-cycles.
* Long: 1368 clocks, 341 dots. 337 dots of 4-clocks, 4 dots of 5-clocks.
* Short: 1360 cycles, 340 dots. 340 dots of 4-cycles.
* Short: 1360 clocks, 340 dots. 340 dots of 4-clocks.


===Vertical Blank===
===Vertical Blank===
Line 53: Line 53:
* Ends after vertical line V=261 (NTSC) or V=311 (PAL). The next line is V=0.
* Ends after vertical line V=261 (NTSC) or V=311 (PAL). The next line is V=0.
* This allows 37 (NTSC) or 87 (PAL) lines of vertical blank normally, or 22 (NTSC) / 72 (PAL) with overscan.
* This allows 37 (NTSC) or 87 (PAL) lines of vertical blank normally, or 22 (NTSC) / 72 (PAL) with overscan.
* With 1324 master cycles per line available, and [[DMA]] taking 8 cycles per byte, this gives an upper bound on DMA bandwidth per blank without [[PPU registers#INIDISP|forced blanking]]:
* With 1324 master clocks per line available, and [[DMA]] taking 8 clocks per byte, this gives an upper bound on DMA bandwidth per blank without [[PPU registers#INIDISP|forced blanking]]:


{|class="wikitable sortable"
{|class="wikitable sortable"
! Region !! [[PPU registers#SETINI|Overscan]] !! VBlank Lines !! VBlank Cycles !! DMA Bandwidth
! Region !! [[PPU registers#SETINI|Overscan]] !! VBlank Lines !! VBlank Clocks !! DMA Bandwidth
|-
|-
| NTSC || 224 || 37 || 48,988 || 5.9 kb
| NTSC || 224 || 37 || 48,988 || 5.9 kb
Line 68: Line 68:


===Frame===
===Frame===
The total cycles per frame is dependent on many factors:
The total clocks per frame is dependent on many factors:
* Some scanlines are slightly short or long.
* Some scanlines are slightly short or long.
* [[PPU registers#SETINI|interlacing]] adds an extra scanline on even frames.
* [[PPU registers#SETINI|Interlacing]] adds an extra scanline on even frames.
* DRAM refresh uses 40 cycles in the middle of each line.
* DRAM refresh uses 40 clocks in the middle of each line.
* [[HDMA]] uses varying amounts of cycles just past the active part of each line.
* [[HDMA]] uses varying amounts of clocks just past the active part of each line.


Approximating a total, ignoring the factors above and assuming a constant number of cycles per scanline:
Approximating a total, ignoring the factors above and assuming a constant number of clocks per scanline:


{|class="wikitable sortable"
{|class="wikitable sortable"
! Region !! Scanlines !! Master Cycles !! Available Cycles
! Region !! Scanlines !! Master Clocks !! Available Clocks
|-
|-
| NTSC || 262 || 357,368 || 346,888
| NTSC || 262 || 357,368 || 346,888

Latest revision as of 01:46, 28 March 2024

Timing of the SNES hardware.

Master Clock

The SNES master clock:

  • NTSC: 945/44 MHz ≈ 21.4773 MHz (6 times chroma).
  • PAL: 21.281370 MHz (4.8 times chroma). This is synthesized by multiplying a 17.734475 MHz (4 times chroma) crystal by 6/5.

CPU

A 65816 CPU "cycle" can take 6, 8 or 12 master clocks on the SNES, depending on the memory region accessed, and the MEMSEL fast-ROM setting.

This gives some commonly quoted SNES CPU speeds, though none of them tell a complete story:

  • 3.58 MHz fast-ROM (6-clocks per cycle)
  • 2.68 MHz slow-ROM (8-clocks per cycle)
  • 1.79 MHz other (12-clocks per cycle)

The speed of access depends on the memory region:[1]

  • 6-clocks for fast-ROM access, enabled via MEMSEL and accessed at an address of $800000 of higher.
  • 8-clocks for slow-ROM access.
  • 8-clocks for internal S-WRAM.
  • 6-clocks for most MMIO registers.
  • 12-clocks for JOYSER0 and JOYSER1.
  • 6-clocks for "internal" cycles not accessing memory (e.g. 2nd cycle of NOP)

Video

Scanline

  • 1364 master clocks = 341 dot cycles.
  • The CPU pauses for 40 master clocks in the middle of each scanline for DRAM refresh, leaving 1324 active clocks per line.
  • The left edge of the picture begins at clock 88 of the line and continues until 1112.
  • Scanline 0 is the end of vblank and beginning of rendering. It is hidden, and displays as a blank line.
  • Scanlines 1-224 or 239 will normally render the visible image, unless force blanking is applied.
  • Scanline 261 or 311 is the last line of vertical blank (NTSC or PAL), after which the next frame begins rendering.
  • With interlacing on, 1 extra scanline will appear with each even frame, and one scanline outside the visible picture will be slightly shortened or lengthened for color synchronization.

Short and Long Scanlines

Scanlines are normally 1364 master clocks, but there are two special cases:

  • Short scanline: NTSC with interlace off, field=1, V=240.
  • Long scanline: PAL with interlace on, field=1, V=311.

This means every odd frame may have one adjusted scanline, depending on the interlacing setting and region. It affects the same line regardless of the overscan setting.

These adjusted scanlines are always during vertical blank, so they do not affect the visible picture directly, but they are necessary to maintain synchronization with the colour signal.

Timings:

  • Normal: 1364 clocks, 340 dots. 336 dots of 4-clocks, 4 dots of 5-clocks.
  • Long: 1368 clocks, 341 dots. 337 dots of 4-clocks, 4 dots of 5-clocks.
  • Short: 1360 clocks, 340 dots. 340 dots of 4-clocks.

Vertical Blank

  • Begins on vertical line V=225 or V=240 based on (SETINI overscan setting).
  • Ends after vertical line V=261 (NTSC) or V=311 (PAL). The next line is V=0.
  • This allows 37 (NTSC) or 87 (PAL) lines of vertical blank normally, or 22 (NTSC) / 72 (PAL) with overscan.
  • With 1324 master clocks per line available, and DMA taking 8 clocks per byte, this gives an upper bound on DMA bandwidth per blank without forced blanking:
Region Overscan VBlank Lines VBlank Clocks DMA Bandwidth
NTSC 224 37 48,988 5.9 kb
NTSC 239 22 29,128 3.5 kb
PAL 224 87 115,188 14.0 kb
PAL 239 72 95,328 11.6 kb

Frame

The total clocks per frame is dependent on many factors:

  • Some scanlines are slightly short or long.
  • Interlacing adds an extra scanline on even frames.
  • DRAM refresh uses 40 clocks in the middle of each line.
  • HDMA uses varying amounts of clocks just past the active part of each line.

Approximating a total, ignoring the factors above and assuming a constant number of clocks per scanline:

Region Scanlines Master Clocks Available Clocks
NTSC 262 357,368 346,888
PAL 312 425,568 413,088

Tools

References

  1. Fullsnes: CPU Clock Cycles