Stream: git-wasmtime

Topic: wasmtime / PR #1669 machinst: reduce the number of memory...

view this post on Zulip Wasmtime GitHub notifications bot (May 06 2020 at 17:43):

bnjbvr requested cfallin for a review on PR #1669.

view this post on Zulip Wasmtime GitHub notifications bot (May 06 2020 at 17:43):

bnjbvr opened PR #1669 from reduce-malloc to master:

This set of small changes in the code helps reducing the number of memory allocations, as reported by DHAT, during compilation of regex-rs.wasm with aarch64:

#bytes allocated #blocks allocated avg size of block
Before 1 086 592 728 2 734 386 397,38
After 1 078 766 780 2 516 317 428,71

So this represents an 8% reduction of the number of allocated memory blocks. This doesn't seem to show any compiler runtime performance improvement, but the whole programs compiles in less than 1.2 seconds on my machine, so it's a bit hard to distinguish what's noise and what's not.

Will need a regalloc.rs bump for https://github.com/bytecodealliance/regalloc.rs/pull/59, so CI will fail.

view this post on Zulip Wasmtime GitHub notifications bot (May 06 2020 at 17:51):

cfallin submitted PR Review.

view this post on Zulip Wasmtime GitHub notifications bot (May 06 2020 at 17:51):

cfallin submitted PR Review.

view this post on Zulip Wasmtime GitHub notifications bot (May 06 2020 at 17:51):

cfallin created PR Review Comment:

tiny nit: s/which/whose/ (and below too)

view this post on Zulip Wasmtime GitHub notifications bot (May 07 2020 at 09:47):

bnjbvr updated PR #1669 from reduce-malloc to master:

This set of small changes in the code helps reducing the number of memory allocations, as reported by DHAT, during compilation of regex-rs.wasm with aarch64:

#bytes allocated #blocks allocated avg size of block
Before 1 086 592 728 2 734 386 397,38
After 1 078 766 780 2 516 317 428,71

So this represents an 8% reduction of the number of allocated memory blocks. This doesn't seem to show any compiler runtime performance improvement, but the whole programs compiles in less than 1.2 seconds on my machine, so it's a bit hard to distinguish what's noise and what's not.

Will need a regalloc.rs bump for https://github.com/bytecodealliance/regalloc.rs/pull/59, so CI will fail.

view this post on Zulip Wasmtime GitHub notifications bot (May 07 2020 at 10:24):

bnjbvr merged PR #1669.

Last updated: Oct 23 2024 at 20:03 UTC