Petr Penzin

Horizontal adds can be tricky. Currently the function does this: ``` float horizontal_add(float4 x) { auto partial_sum = x.xy + x.zw; return partial_sum.x + partial_sum.y; } ``` This moves low...

@nlewycky I'd take a look at the code. I am also curious about what happens when we go from wasm generated for this example to assembly (in Wasmer codegen or...

@nlewycky what you think has changed, if horizontal add stays the same across optimization then it is probably not useful. Same for LLVM IR vs native -- if IR is...

Ping @nlewycky Should the [results][m] be reproducible with Wasmer 0.6.0 from get.wasmer.io? For some reason I can't get the parity between Wasm and native on Ubuntu 18.04 (on Core-i7-7700K). I...

Thank you both for quick responses! @nlewycky thank you for the flags, that gives me something to work with. I was able to dump IR with this (built from master):...

Discard makes the address range "disappear" - while no indices need to change the underlying memory gets deallocated (this is realistically achieved by memory mapping in the runtime). > It's...

It changes the footprint - when you deallocate everything on a memory page, it then gets `discard`'ed (by `free()` for example) and memory consumption of the tab drops. Without that...