Stream: git-wasmtime

Topic: wasmtime / issue #14694 Cranelift: missed optimization fo...

Wasmtime GitHub notifications bot (Oct 13 2026 at 02:27):

Takashiidobe opened issue #14694:

.clif Test Case

function %missing_division(i8) -> i8 {
block0(v0: i8):
    v1 = iconst.i8 7
    v10 = sdiv v0, v1
    v2 = iconst.i8 0
    return v2
}

Steps to Reproduce

• compile on any architecture

Expected Results

the sdiv cannot trap and is dead, so i would expect something like return 0

Actual Results

for x86_64 I see:

push rbp
mov rbp, rsp
mov rax, rdi
cbw
mov esi, 7
idiv sil    ; trap: int_ovf
xor eax, eax
mov rsp, rbp
pop rbp
ret

with the trap comment so I assume it's not properly deleted

Versions and Environment

Cranelift version or commit: 0.136.2

Operating system: linux

Architecture: x86_64, though this happens on all architectures since this isn't lowering specific

Extra Info

https://github.com/bytecodealliance/wasmtime/issues/5908 seems to indicate that work was done to handle this problem already.

i32 and i64 already properly handle this case, i8 and i16 do not.

I'm assuming that just extending i8 and i16 to i32 is a cheap fix for this particular problem, since they both just emit, which deletes the division entirely

push rbp
mov rbp, rsp
xor eax, eax
mov rsp, rbp
pop rbp
ret

Wasmtime GitHub notifications bot (Oct 13 2026 at 02:27):

Takashiidobe added the bug label to Issue #14694.

Wasmtime GitHub notifications bot (Oct 13 2026 at 02:27):

Takashiidobe added the cranelift label to Issue #14694.

Wasmtime GitHub notifications bot (Oct 13 2026 at 14:37):

alexcrichton added the cranelift:goal:optimize-speed label to Issue #14694.

Wasmtime GitHub notifications bot (Oct 19 2026 at 16:41):

alexcrichton closed issue #14694:

.clif Test Case

function %missing_division(i8) -> i8 {
block0(v0: i8):
    v1 = iconst.i8 7
    v10 = sdiv v0, v1
    v2 = iconst.i8 0
    return v2
}

Steps to Reproduce

• compile on any architecture

Expected Results

the sdiv cannot trap and is dead, so i would expect something like return 0

Actual Results

for x86_64 I see:

push rbp
mov rbp, rsp
mov rax, rdi
cbw
mov esi, 7
idiv sil    ; trap: int_ovf
xor eax, eax
mov rsp, rbp
pop rbp
ret

with the trap comment so I assume it's not properly deleted

Versions and Environment

Cranelift version or commit: 0.136.2

Operating system: linux

Architecture: x86_64, though this happens on all architectures since this isn't lowering specific

Extra Info

https://github.com/bytecodealliance/wasmtime/issues/5908 seems to indicate that work was done to handle this problem already.

i32 and i64 already properly handle this case, i8 and i16 do not.

I'm assuming that just extending i8 and i16 to i32 is a cheap fix for this particular problem, since they both just emit, which deletes the division entirely

push rbp
mov rbp, rsp
xor eax, eax
mov rsp, rbp
pop rbp
ret

Last updated: Oct 31 2026 at 23:03 UTC