Difference between revisions of "DSP Memory Region"

From 3dbrew
Jump to navigation Jump to search
 
(17 intermediate revisions by 3 users not shown)
Line 1: Line 1:
'''Some of this is known to be incorrect. I'll correct this in a few months after a bit more RE work. [[User:Merry|Merry]] ([[User talk:Merry|talk]]) 21:10, 28 January 2016 (CET)'''
+
''Note that everything below may vary depending on the exact DSP firmware used and different variants have slightly different behaviours.''
 
 
----
 
 
 
*Note that everything below will vary depending on the exact DSP firmware used and different variants have slightly different behaviours.*
 
  
 
The DSP communicates with the application through two shared memory areas 0x8000 bytes long each (at 0x1FF50000 and 0x1FF70000 respectively). The DSP alternates between the use of these two areas (much like a double-buffer). Each area has 15 structures within it. The location of these structures can be obtained by [[DSP:ReadPipe|reading channel 2 of the DSP pipe]]. A list of structures in the order the DSP addresses are read from the pipe follows:
 
The DSP communicates with the application through two shared memory areas 0x8000 bytes long each (at 0x1FF50000 and 0x1FF70000 respectively). The DSP alternates between the use of these two areas (much like a double-buffer). Each area has 15 structures within it. The location of these structures can be obtained by [[DSP:ReadPipe|reading channel 2 of the DSP pipe]]. A list of structures in the order the DSP addresses are read from the pipe follows:
Line 33: Line 29:
 
13. Unknown Coefficients
 
13. Unknown Coefficients
  
14. Unknown Coefficients
+
14. Surround sound biquad filter 1
  
15. Unknown Coefficients
+
15. Surround sound biquad filter 2
  
 
The DSP has 24 inputs, each of which are individually configurable. These 24 inputs each produce three sets of 4 audio channels (two left, two right).  
 
The DSP has 24 inputs, each of which are individually configurable. These 24 inputs each produce three sets of 4 audio channels (two left, two right).  
Line 42: Line 38:
  
 
== Frame Count ==
 
== Frame Count ==
 +
 +
The frame count of the first region must be even and the frame count of the second region must be odd.
 +
 +
The frame with the higher count is the "current region".
 +
 +
The DSP firmware only responds if the first frame count is 4.
  
 
== Input Config ==
 
== Input Config ==
Line 123: Line 125:
 
| 180
 
| 180
 
| u16
 
| u16
| bits[0:1]: 1 = mono, 2 = stereo; bits[2:3]: buffer format, bits[5]: fade in
+
| Format
 
|-
 
|-
 
| 182
 
| 182
Line 136: Line 138:
 
| u16
 
| u16
 
| Buffer Id of embedded buffer
 
| Buffer Id of embedded buffer
 +
|}
 +
 +
There is a 2 sample delay in this preprocessing stage, likely due to the interpolation step.
 +
 +
=== Format ===
 +
 +
This is a u16.
 +
 +
{| class="wikitable"
 +
|-
 +
| Bits
 +
| Desciption
 +
|-
 +
| 0-1
 +
| Number of channels: 0,1,3 = mono; 2 = stereo
 +
|-
 +
| 2-3
 +
| Buffer codec: [[BCWAV#Encoding|0:PCM8; 1:PCM16; 2:ADPCM]]
 +
|-
 +
| 5
 +
| Fade
 
|}
 
|}
  
Line 213: Line 236:
 
| 2
 
| 2
 
| s1.15
 
| s1.15
| a1
+
| a1 (negated)
 +
|}
 +
 
 +
=== Biquad Filter ===
 +
 
 +
This is a [[wikipedia:Digital_biquad_filter|biquadratic filter]].
 +
 
 +
{| class="wikitable"
 +
|-
 +
| Offset
 +
| Type
 +
| Description
 +
|-
 +
| 0
 +
| s2.14
 +
| a2 (negated)
 +
|-
 +
| 2
 +
| s2.14
 +
| a1 (negated)
 +
|-
 +
| 4
 +
| s2.14
 +
| b2
 +
|-
 +
| 6
 +
| s2.14
 +
| b1
 +
|-
 +
| 8
 +
| s2.14
 +
| b0
 
|}
 
|}
  
Line 232: Line 286:
 
| 1
 
| 1
 
| u8
 
| u8
| Dirty flag for previous buffer id.
+
| Dirty flag for buffer id, set to 1 when buffer (after the first) starts playing
 
|-
 
|-
 
| 2
 
| 2
Line 244: Line 298:
 
| 8
 
| 8
 
| u16
 
| u16
| Buffer id of last buffer that was played
+
| Buffer id of the buffer that's just started playing.
 
|-
 
|-
 
| 10
 
| 10
Line 277: Line 331:
  
 
=== Reverb Effect ===
 
=== Reverb Effect ===
 +
 +
Reverb consists of two comb filters and one all-pass filter in standard configuration.
  
 
== DSP status ==
 
== DSP status ==
  
Read only.
+
Read only. A 32 byte long structure.
 +
 
 +
{| class="wikitable"
 +
|-
 +
| Offset
 +
| Type
 +
| Description
 +
|-
 +
| 0
 +
| u16
 +
| ?
 +
|-
 +
| 2
 +
| u16
 +
| Number of dropped frames
 +
|-
 +
| 4
 +
| 28 bytes
 +
| ?
 +
|}
  
 
== Output samples ==
 
== Output samples ==
  
Read only. This structure is 320 bytes long.
+
Read only. This structure is 640 bytes long. Output is stereo (the 3DS has two speakers).
  
 
{| class="wikitable"
 
{| class="wikitable"
Line 294: Line 369:
 
| 0
 
| 0
 
| s16[160]
 
| s16[160]
| Samples
+
| Left-channel Samples
 +
|-
 +
| 320
 +
| s16[160]
 +
| Right-channel Samples
 
|}
 
|}
 +
 +
This structure has separate arrays for the left and right channels.
  
 
== Intermediate mix samples ==
 
== Intermediate mix samples ==
Line 301: Line 382:
 
Read/Write.
 
Read/Write.
  
PCM32. Also serves an aux function, allowing the ARM11 to apply custom effects to audio.
+
PCM32. Also serves an aux function, allowing the ARM11 to apply custom effects to audio. Internal format of the DSP firmware is quadraphonic audio.
  
 
This structure is 5120 bytes long.
 
This structure is 5120 bytes long.
Line 312: Line 393:
 
|-
 
|-
 
| 0
 
| 0
| s32[2][4][160]
+
| IntermediateSample[160]
 
| Samples
 
| Samples
 +
|}
 +
 +
In contrast to the final output samples, this structure has the left/right channels interleaved.
 +
 +
=== Intermediate sample ===
 +
 +
A quadraphonic sample.
 +
 +
{| class="wikitable"
 +
|-
 +
| Offset
 +
| Type
 +
| Description
 +
|-
 +
| 0
 +
| s32
 +
| Left Channel A
 +
|-
 +
| 4
 +
| s32
 +
| Right Channel A
 +
|-
 +
| 8
 +
| s32
 +
| Left Channel B
 +
|-
 +
| 12
 +
| s32
 +
| Right Channel B
 
|}
 
|}
  
Line 319: Line 429:
  
 
A precomputed response curve lookup table for the compressor.
 
A precomputed response curve lookup table for the compressor.
 +
 +
[[Category:DSP]]

Latest revision as of 22:00, 28 September 2024

Note that everything below may vary depending on the exact DSP firmware used and different variants have slightly different behaviours.

The DSP communicates with the application through two shared memory areas 0x8000 bytes long each (at 0x1FF50000 and 0x1FF70000 respectively). The DSP alternates between the use of these two areas (much like a double-buffer). Each area has 15 structures within it. The location of these structures can be obtained by reading channel 2 of the DSP pipe. A list of structures in the order the DSP addresses are read from the pipe follows:

1. Frame count

2. Input configurations

3. Input status

4. Input ADPCM coefficients

5. DSP configuration

6. DSP status

7. Output samples

8. Intermediate mix samples

9. Compressor table

10. DSP debug statistics

11. Unknown Coefficients

12. Unknown Coefficients

13. Unknown Coefficients

14. Surround sound biquad filter 1

15. Surround sound biquad filter 2

The DSP has 24 inputs, each of which are individually configurable. These 24 inputs each produce three sets of 4 audio channels (two left, two right).

These four audio channels feed into three intermediate mixers. Two of these intermediate mixers are used for effects and aux.

Frame Count[edit]

The frame count of the first region must be even and the frame count of the second region must be odd.

The frame with the higher count is the "current region".

The DSP firmware only responds if the first frame count is 4.

Input Config[edit]

A 192 byte long structure. There are 24 of them.

Offset Type Description
0 u32 Dirty flags
4 f32[3][2][2] Input Gain (Each input has 12 channels)
52 f32 Rate multiplier (1.0x == native DSP rate)
56 u8 Interpolation mode
57 u8 Polyphase filter select
58 u16 bit[0]: Simple Filter enabled, bit[1]: Biquadratic Filter enabled
60 SimpleFilter Simple Filter (One pole normalized recursive linear filter)
64 BiquadFilter Biquadratic Filter (Two poles two zeros normalized recursive linear filter)
74 u16 Bitmap of which buffers in queue are dirty
76 Buffer[4] Buffer queue
156 u32 -
160 u16 Is Active
162 u16 Sync Count
164 u32 Play position
168 4 ?
172 u32 Physical address of embedded buffer
176 u32 Number of samples in embedded buffer
180 u16 Format
182 AdpcmData ADPCM data associated with embedded buffer
188 u16 bit[0]: ADPCM updated?; bit[1]: Is looping?r
190 u16 Buffer Id of embedded buffer

There is a 2 sample delay in this preprocessing stage, likely due to the interpolation step.

Format[edit]

This is a u16.

Bits Desciption
0-1 Number of channels: 0,1,3 = mono; 2 = stereo
2-3 Buffer codec: 0:PCM8; 1:PCM16; 2:ADPCM
5 Fade

Buffer[edit]

Offset Type Description
0 u32 Physical Address
4 u32 Sample Count
8 AdpcmData ADPCM data
14 u8 ADPCM data dirty?
15 u8 Looping?
16 u16 Buffer Id
18 u16 -

Adpcm Data[edit]

0 u8 ADPCM predictor/scale
1 u8 -
2 s16 ADPCM y[n-1]
4 s16 ADPCM y[n-2]


Simple Filter[edit]

This is a standard single-pole filter. The fall-off is 6dB per octave as you would expect.

Offset Type Description
0 s1.15 b0
2 s1.15 a1 (negated)

Biquad Filter[edit]

This is a biquadratic filter.

Offset Type Description
0 s2.14 a2 (negated)
2 s2.14 a1 (negated)
4 s2.14 b2
6 s2.14 b1
8 s2.14 b0

Input status[edit]

Read only. This structure is set by the DSP. This structure is 12 bytes long and there are 24 of them.

Offset Type Description
0 u8 Input Enabled?
1 u8 Dirty flag for buffer id, set to 1 when buffer (after the first) starts playing
2 u16 Sync count
4 u32 Position (number of samples) into current buffer playback
8 u16 Buffer id of the buffer that's just started playing.
10 u16 -

Input ADPCM coefficients[edit]

This is a 32 byte long structure. There are 24 of them.

Offset Type Description
0 s5.11[16] ADPCM coefficents

DSP configuration[edit]

Delay Effect[edit]

Delay with feedback.

Length of delay is expressed in terms of number of audio frames (there are 160 samples per audio frame).

Feedback arm only has a gain on it. Under the feedback arm is a single-pole filter with the delay.

Reverb Effect[edit]

Reverb consists of two comb filters and one all-pass filter in standard configuration.

DSP status[edit]

Read only. A 32 byte long structure.

Offset Type Description
0 u16 ?
2 u16 Number of dropped frames
4 28 bytes ?

Output samples[edit]

Read only. This structure is 640 bytes long. Output is stereo (the 3DS has two speakers).

Offset Type Description
0 s16[160] Left-channel Samples
320 s16[160] Right-channel Samples

This structure has separate arrays for the left and right channels.

Intermediate mix samples[edit]

Read/Write.

PCM32. Also serves an aux function, allowing the ARM11 to apply custom effects to audio. Internal format of the DSP firmware is quadraphonic audio.

This structure is 5120 bytes long.

Offset Type Description
0 IntermediateSample[160] Samples

In contrast to the final output samples, this structure has the left/right channels interleaved.

Intermediate sample[edit]

A quadraphonic sample.

Offset Type Description
0 s32 Left Channel A
4 s32 Right Channel A
8 s32 Left Channel B
12 s32 Right Channel B

Compressor table[edit]

A precomputed response curve lookup table for the compressor.