ARM-Mali-Bifrost.md

Bifrost is a 4th generation of Mali GPU architecture.

Content:

• Gen1
• Gen2
• Gen3
• All gens

Bifrost Gen1

Examples

• Mali-G31
• Mali-G51
• Mali-G71

References

1.1. ARM Unveils Next Generation Bifrost GPU Architecture & Mali-G71
1.2. Arm Mali-G71 Performance Counters Reference Guide, [backup]

Notes

• implement a single texel-per-clock and single pixel-per-clock shader core.
• G71: 4-wide SIMD per quad; 3x quads per core. 12 FMAs at the same time. These 3 quads are managed by a core's quad manager. [4]
• In both the Mali-G71 and G72, a quad is just that: a 4-wide SIMD unit, with each lane possessing separate FMA and ADD/SF pipes. [4]

Bifrost Gen2

Examples

• Mali-G52
• Mali-G72

References

2.1. ARM Announces Mali-G72
2.2. Arm Mali-G72 Performance Counters Reference Guide, [backup]
2.3. Mali-G52

Notes

• implement a single texel-per-clock and single pixel-per-clock shader core.

Bifrost Gen3

Examples

• Mali-G76

References

3.1. Arm Mali-G76 Performance Counters Reference Guide, [backup]

Notes

• Implement a two texel-per-clock and two pixel-per-clock shader core, with an increase in arithmetic performance to compensate. Not every GPU doubled the available performance though.
• G76: 8-wide SIMD per quad, 3 quads per core. [4]

Bifrost (all gens)

References

1. The Bifrost Shader Core, [backup]
2. The Bifrost GPU architecture
3. Mesa driver details
4. Everything we learnt from hacking Arm Mali GPUs

Notes

• scalar

• Mali GPUs can contain many identical shader cores. Each shader core supports hundreds of concurrently executing threads. [4]

• Each shader core contains: [4]

  • One to three arithmetic pipelines or execution engines.
  • One load-store pipeline.
  • One texture pipeline.

• AFBC (v1.2) with 4x4 block.

• Transaction Elimination with 16x16 pixel block size.