Wasm baseline: SIMD shuffle optimizations
Categories
(Core :: JavaScript: WebAssembly, enhancement, P5)
Tracking
()
People
(Reporter: lth, Unassigned)
References
(Blocks 1 open bug)
Details
Virtually all of the machinery we need to implement the shuffle optimizations in the baseline compiler is already available, because it's not Ion-specific. (The exception is some of the code generator bits but we can work on that, easily; and some of the code may need to be lifted out of the Jit and into, say, wasm/WasmSIMD.cpp and generalized slightly.) Thus baseline massively pessimizes shuffle. Since Liftoff is starting to optimize shuffle, and it's not a big deal for us to do the same, we should do so too.
Reporter | ||
Updated•5 years ago
|
Reporter | ||
Comment 1•5 years ago
|
||
Also on ARM64. This will entail generalizing the machinery that's now in x86-specific Ion code to a cross-architecture solution that also works for baseline. This still should be a modest amount of work, but the selection of optimizable patterns is somewhat different for ARM64: there are some new patterns to recognize, and the priority of the different patterns may be different due to different hardware support.
Reporter | ||
Updated•4 years ago
|
Reporter | ||
Updated•4 years ago
|
Description
•