bnjbvr opened PR #1915 from x64-wip
to master
:
This makes it possible to run a simple recursive fibonacci function in
wasmtime, with the new x64 backend. Plenty of todos implemented as assertions that may and will trigger on larger programs, plenty of paths are still untested, still a lot of missing opcodes lowering and code gen, plenty of optimizations, etc etc.
bnjbvr requested cfallin and julian-seward1 for a review on PR #1915.
bnjbvr requested cfallin and julian-seward1 for a review on PR #1915.
cfallin submitted PR Review.
bjorn3 submitted PR Review.
bjorn3 created PR Review Comment:
Does space mean size or location?
cfallin created PR Review Comment:
Size, in this case (the
aarch64
impl is the same (and as an aside, we hope to factor out the commonalities here soon)).
cfallin submitted PR Review.
bjorn3 submitted PR Review.
bjorn3 created PR Review Comment:
ty.bits()
bjorn3 submitted PR Review.
bjorn3 created PR Review Comment:
Why is it an
i64
then? A negative size doesn't make sense.
julian-seward1 submitted PR Review.
julian-seward1 created PR Review Comment:
pico-nit (smaller even than a nano-nit): These are now sign-extended loads and moves.
julian-seward1 submitted PR Review.
julian-seward1 created PR Review Comment:
atto-nit: (reg addr), but no imm
julian-seward1 submitted PR Review.
julian-seward1 created PR Review Comment:
Random comment: if at some point you add
tst
as an instruction, it has the same "shape" ascmp
and so would fit naturally here. (cmp = subtract operands, set flags, throw away result; tst = AND operands; set flags; throw away result).
bnjbvr updated PR #1915 from x64-wip
to master
:
This makes it possible to run a simple recursive fibonacci function in
wasmtime, with the new x64 backend. Plenty of todos implemented as assertions that may and will trigger on larger programs, plenty of paths are still untested, still a lot of missing opcodes lowering and code gen, plenty of optimizations, etc etc.
bnjbvr merged PR #1915.
bnjbvr submitted PR Review.
bnjbvr created PR Review Comment:
Thanks! Even updated the callers so they use ty.bits() instead.
bnjbvr submitted PR Review.
bnjbvr created PR Review Comment:
Good question; I'll take care of it during the Grand Refactoring.
bjorn3 created PR Review Comment:
*convention
bjorn3 submitted PR Review.
bjorn3 created PR Review Comment:
This should have used
next_vreg
instead ofnext_gpr
.
bjorn3 submitted PR Review.
Last updated: Dec 23 2024 at 12:05 UTC