RISE RP005 QEMU weekly report 2024-10-02

Work completed since last report

WP2
- Complete SiFive library function set
  - added strcat.
  - work in progress on the remaining 8 functions.
- Single instruction benchmarking relevant to SiFive library functions
  - on hold.
- Resume work on AArch64 comparison.
  - working on the single instruction comparison between a Aarch64 vector load and vle8.v.
  - on hold to prioritize the separation of the patches.
Other
- Separate Embecosm's and Max's patches:
  - work in progress. This is a substantial effort because all the changes since July have been assumed to build on Max's work, and hence we need to write code to allow the patch to work standalone.
- Updated small load/store loop optimization upstream with user mode condition.

WP2
- Complete SiFive library function set.
- Single instruction benchmarking relevant to SiFive library functions.
- Continue to work on AArch64 comparison.
Other:
- Max: share patches applied to our benchmarking scripts.
- Max: run SPECCPU 2017 results again with VLEN=128 and LMUL=1.

It is useful to keep an idea of what we have achieved since 1 May 2024. We use ns/instruction as our metric.

Metric	Start	Now	Speedup	Notes
`memcpy` worst case			2%
`memcpy` best case			6,869%
SPEC CPU 2017 average	13.3	11.3	15%
SPEC CPU 2017 best case	12.0	7.4	38%	`628.pop2_s`

Percentage changes relative to "now". SPEC CPU 2017 results are from August, since we have stopped measuring this.