Closed chriselrod closed 3 years ago
chriselrod
commented
3 years ago The image doesn't show up because I hardcoded the path, and this PR will have to be merged before there are actually images there.
I'm not sure if ".svg" images are supported in Markdown on GitHub, so for now it's pointing to ".png.".
codecov[bot]
commented
3 years ago Merging #86 (1faccdc) into master (ba60608) will decrease coverage by
5.98%. The diff coverage is100.00%.
@@ Coverage Diff @@
## master #86 +/- ##
==========================================
- Coverage 88.03% 82.05% -5.99%
==========================================
Files 11 11
Lines 652 652
==========================================
- Hits 574 535 -39
- Misses 78 117 +39 | Impacted Files | Coverage Δ | |
|---|---|---|
| src/complex_matmul.jl | 100.00% <100.00%> (ø) |
|
| src/memory_buffer.jl | 36.84% <0.00%> (-31.58%) |
:arrow_down: |
| src/matmul.jl | 80.74% <0.00%> (-10.00%) |
:arrow_down: |
| src/global_constants.jl | 50.00% <0.00%> (-10.00%) |
:arrow_down: |
| src/block_sizes.jl | 94.91% <0.00%> (-1.70%) |
:arrow_down: |
Continue to review full report at Codecov.
Legend - Click here to learn more
Δ = absolute <relative> (impact),ø = not affected,? = missing dataPowered by Codecov. Last update ba60608...1faccdc. Read the comment docs.
Unfortunately, my M1 always segfaults whenever I interact with the REPL after running the benchmarks. Perhaps I should try making a 1-liner that runs the benchmarks and saves the benchmark results, so that I can load it from Julia+Rosetta and produce the plots. Maybe condensing it into 1 line would let me save before it crashes.
I made it so that BLASBenchmarksCPU on the M1 benchmarks AppleAccelerate, which of course dominates the others thanks to using the matrix coprocessor.
For now, I'm just adding the 10980XE in this PR. But more PRs are welcome.
Additional possibilities I can run it on: Xeon SIlver with a single AVX512 FMA unit, 1165G7 (laptop with single AVX512 FMA unit), 1950X (Zen1, single 256 bit fma).