blob: a0b53ce2b9a21fc14894b9724402398374ce8f2a [file] [log] [blame]
Background
==========
- Priority scale: High, Medium and Low
- Complexity scale: C1, C2, C4 and C8. The complexity scale is exponential,
with complexity 1 being the lowest complexity. Complexity is a function
of both task 'complexity' and task 'scope'.
The general rule of thumb is that a complexity 1 task should take 1-2 weeks
for a person very familiar with oFono codebase. Higher complexity tasks
require more time and have higher uncertainty.
Higher complexity tasks should be refined into several lower complexity tasks
once the task is better understood.
Encoder optimizations
=====================
- Currently, only the decoder is optimized to take advantage of advanced
processor instruction sets. In use cases like HFP 1.6, optimizing the
encoder will bring a significant latency, power and performance advantage.
For example, the MMX encoder is 3 to 6 time faster than the SIMD encoder.
Priority: High
Complexity: C8
- Use a log2 table for byte integer scale factors calculation (sum log2
results for high and low bytes) fill bitpool by 16 bits instead of one
at a time in bits allocation/bitpool generation port to the dsp
Priority: Medium
Complexity: C2
- The neon optimization code is split in two parts. Sample reordering and
blocks encoding. There is a neon optimization for encoding SBC. But mSBC
is not supported by this optimizer because the reordering has been
specifically for mSBC.
Priority: Low
Complexity: C2
Decoder optimizations
=====================
- The decoder is optimized to take advantage of advanced processor
instruction sets. Currently implemented are MMX, arm neon, arm v6
and iwmmxt. SSE3 is available since almost 10 years now, on a large
range of Intel processors. It should be interesting to implement it
and to compare with MMX implementation on Intel processors.
Priority: Medium
Complexity: C4