Stream: git-cranelift

Topic: cranelift / Issue #1304 Encode stack load store

GitHub (Dec 20 2019 at 20:19):

bjorn3 commented on Issue #1304:

Could somebody help me with the encoding problem?

GitHub (Jan 25 2020 at 14:00):

bjorn3 commented on Issue #1304:

I fixed the encoding problem. (I forgot a .rex().w())

GitHub (Jan 25 2020 at 14:29):

bjorn3 commented on Issue #1304:

```
Benchmark #1: ./raytracer_cg_clif
Time (mean ± σ): 9.283 s ± 0.134 s [User: 9.224 s, System: 0.025 s]
Range (min … max): 9.124 s … 9.591 s 20 runs

Benchmark #2: ./raytracer_cg_clif_opt
Time (mean ± σ): 7.071 s ± 0.098 s [User: 7.023 s, System: 0.021 s]
Range (min … max): 6.997 s … 7.381 s 20 runs

Benchmark #3: ./raytracer_cg_clif_opt_stack_enc
Time (mean ± σ): 6.912 s ± 0.063 s [User: 6.872 s, System: 0.018 s]
Range (min … max): 6.861 s … 7.113 s 20 runs

Benchmark #4: ./raytracer_cg_clif_stack_enc
Time (mean ± σ): 8.966 s ± 0.109 s [User: 8.915 s, System: 0.023 s]
Range (min … max): 8.861 s … 9.343 s 20 runs

Benchmark #5: ./raytracer_cg_llvm
Time (mean ± σ): 7.252 s ± 0.201 s [User: 7.183 s, System: 0.029 s]
Range (min … max): 7.086 s … 7.663 s 20 runs

Warning: Statistical outliers were detected. Consider re-running this benchmark on a quiet PC without any interferences from other programs. It might help to use the '--warmup' or '--prepare' options.

Summary
'./raytracer_cg_clif_opt_stack_enc' ran
1.02 ± 0.02 times faster than './raytracer_cg_clif_opt'
1.05 ± 0.03 times faster than './raytracer_cg_llvm'
1.30 ± 0.02 times faster than './raytracer_cg_clif_stack_enc'
1.34 ± 0.02 times faster than './raytracer_cg_clif'

This is a performance win of a few percent.
~~~

GitHub (Feb 13 2020 at 10:58):

bnjbvr commented on Issue #1304:

(I tried it on our Spidermonkey wasm benchmark suite, and it's between a slowdown down to 1.8% and a speedup up to 2%, depending on the benchmark.)

GitHub (Feb 13 2020 at 11:52):

bjorn3 commented on Issue #1304:

I am curious about what would cause the slowdown. What kind of operations does that 1.8% slowdown benchmark perform. Does it use the same stack slot a lot? Or is is the generated code a bit bigger after this PR?

GitHub (Feb 28 2020 at 23:36):

alexcrichton commented on Issue #1304:

Thanks for the PR again, and as a procedural note the Cranelift repository has now merged into the wasmtime repository.

PRs are no longer landing in this repository, and unfortunately there's no "one button" solution to move a PR to the wasmtime repository. A script has been prepared, however, to assist you in transferring this PR to the wasmtime repo. Feel free to reach out on Zulip with any questions!

Last updated: Dec 23 2024 at 13:07 UTC