- connecting hundreds of temporary registers to places where they’ll eventually be used eats up about half the power budget in a CPU.
- solution to this problem is replacing the registers in a CPU with something called a ‘belt’
- a weird combination of a stack and a shift register.
- CPU can take data from any position on the belt, perform an operation, and places the result at the front of the belt.
- No general registers or rename
registers >> fast, small, low-power bypass
- No issue, dispatch, or retire stages
short pipe, low mispredict penalty
- No encoded result addresses
compact code
Instructions and Pipelines
The Belt
Memory
- a 64-bit Single Address Space (SAS) .
- On-CPU caches all use virtual addressing
- virtualisation is performed by a Translation Lookaside Buffer (TLB).
- TLB allocate virtual memory in HW
The Belt
Introduction:
Mill CPU Architecture
- The Mill is a new general-purpose
commercial CPU family.
- Invented by Out-of-the-Box Computing.
- This presentation from interview with Ivan Godard.
- his group just knew they could "do better".
- They worked for 10 years.
- The Mill manages roughly 10% of the power
usage of other chips for the same
computations.
Metadata
Outline:
What is the Mill CPU?
Why it is different?
The belt.
Metadata
Memory
Instruction and pipeline
Instruction execution
References
Advantages
Advantages:
- Metadata reduces the number of distinct opcodes by a factor of seven.
- Metadata enables speculative execution without fix-up code.
- Metadata eliminates flag-control overheads in floating point.
- Metadata permits vectorizing of
while-loops.
References:
Instruction and pipelines contd.
Mill has lots of pipelines, and each one issue an operation each cyle
Thanks for listening