For the past few weeks I've been working on a Wasm to Go 
<https://github.com/ncruces/wasm2go> translator. It takes a Wasm module and 
converts it to a Go package. I plan to use it for my SQLite driver. I'm 
translating a Wasm build of SQLite into ~ 600 kLoC of Go 
<https://github.com/ncruces/go-sqlite3-wasm>. I've tested it across 20 
GOOS/GOARCH combinations.

I have found that GC produces suboptimal code in some situations.

Wasm is a little endian platform, so a memory load is something like 
(offset is a constant literal):

int32(binary.LittleEndian.Uint32(m.Memory[int64(ptr)+offset:]))

I've noticed that, on little endian platforms, this works much faster:

*(*int32)(unsafe.Pointer((*[4]byte)(m.Memory[int64(ptr)+offset:])))

I think because the first version has 2 bounds checks, and the later just 
one? This is unfortunate, as I have to generate 2 versions of the code (for 
little and big endian).

Am I missing any trick to get the compiler to generate better code?

Thanks! 

-- 
You received this message because you are subscribed to the Google Groups 
"golang-nuts" group.
To unsubscribe from this group and stop receiving emails from it, send an email 
to [email protected].
To view this discussion visit 
https://groups.google.com/d/msgid/golang-nuts/2bffffa0-e64d-4014-95e9-b15fe4961a63n%40googlegroups.com.

Reply via email to